Citation
Characterizations of low-cut filters in handheld digital recorders using long-term average sorted spectrum analysis

Material Information

Title:
Characterizations of low-cut filters in handheld digital recorders using long-term average sorted spectrum analysis
Creator:
Fazio, Dante Alexander
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Master's ( Master of science)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Music and Entertainment Industry Studies, CU Denver
Degree Disciplines:
Recording arts
Committee Chair:
Grigoras, Catalin
Committee Members:
Smith, Jeff M.
Begault, Durand

Notes

Abstract:
An audio signal's long-term average sorted spectrum can be used to reveal information about the source of a recording, which, in the case of this study, will focus on the low-cut filter option of small handheld digital recording devices. These devices are small, affordable, easy to operate and often come equipped from the manufacturer with a myriad of useful features, such as the low-cut filter. The shape of the low-cut filter slope could be a potential source of data in an authentication exam. It is the intention of this study to determine the forensic strength of applying long-term average sorted spectrum analysis to an audio signal to observe the slope of a low-cut filter in order to make comparisons and determine whether they may lead to an identification or an exclusion of a device. Three hypotheses will be examined:  It is possible to identify whether or not the LCF option of a recorder was on or off.  Low-cut filter signal processing differs significantly between recorders.  LTASS/LCF analysis can be used as a means of eliminating a device from a pool of suspected recorders.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
Copyright Dante Alexander Fazio. Permission granted to University of Colorado Denver to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item has the following downloads:


Full Text
CHARACTERIZATIONS OF LOW-CUT FILTERS
IN HANDHELD DIGITAL RECORDERS USING LONG-TERM AVERAGE SORTED SPECTRUM ANALYSIS
by
DANTE ALEXANDER FAZIO B.M., Berklee College of Music, 2006
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Master of Science Recording Arts Program
2019


©2019
DANTE ALEXANDER FAZIO ALL RIGHTS RESERVED
11


This thesis for the Master of Science degree by Dante Alexander Fazio has been approved for the Recording Arts Program by
Catalin Grigoras, Chair Jeff M. Smith Durand Begault
Date: December 14, 2019


Fazio, Dante Alexander (M.S., Recording Arts Program)
Characterizations of Low-Cut Filters in Handheld Digital Recorders Using Long-Term Average Sorted Spectrum Analysis
Thesis directed by Associate Professor Catalin Grigoras
ABSTRACT
An audio signal's long-term average sorted spectrum can be used to reveal information about the source of a recording, which, in the case of this study, will focus on the low-cut filter option of small handheld digital recording devices. These devices are small, affordable, easy to operate and often come equipped from the manufacturer with a myriad of useful features, such as the low-cut filter. The shape of the low-cut filter slope could be a potential source of data in an authentication exam. It is the intention of this study to determine the forensic strength of applying long-term average sorted spectrum analysis to an audio signal to observe the slope of a low-cut filter in order to make comparisons and determine whether they may lead to an identification or an exclusion of a device. Three hypotheses will be examined:
â–  It is possible to identify whether or not the LCF option of a recorder was on or off.
â–  Low-cut filter signal processing differs significantly between recorders.
â–  LTASS/LCF analysis can be used as a means of eliminating a device from a pool of suspected recorders.
The form and content of this abstract are approved. I recommend its publication.
Approved: Catalin Grigoras
IV


DEDICATION
To my family for their love, unyielding support, and genuine enthusiasm in whatever shiny thing
I may be interested in that week.
To my beautiful wife Monica, whose assistance in my completion of this degree program cannot be understated. I love you so much. Thank you for watching the farm while I was off becoming
a scientist.
To our pets, Kingston, Lady, Sherpa, Billie, Duke, Buster, Levi and Midnight. Were it not for your relentless appeals for affection, this thesis would have been completed in half the time.
v


ACKNOWLEDGMENTS
I would like to thank my teachers, Dr. Catalin Grigoras and Jeff Smith, for shepherding my education in this exciting field. Their knowledge and enthusiasm for media forensics is both impressive and contagious, and it has been a pleasure being their student.
I would also like to recognize Leah Haloin and Cole Whitecotton for their extraordinary contributions to the NCMF. Without them, the NCMF would be dark and full of terrors.
I must thank Dr. Durand Begault for being so generous with his time and resources and for introducing me to the NCMF, and it is an honor to have him on my thesis committee.
A huge thank you to Douglas Lacey, my classmate, unofficial instructor, good friend, and extremely generous contributor to the data set for my thesis, for all the knowledge he shared with us despite not receiving a stipend aside for an occasional dirty martini.
Thank you to my classmates for the camaraderie, the exchange of ideas, and for making the evening lectures something to regret in the morning.
vi


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION...........................................................1
Scope and Intention..................................................1
II. PERSPECTIVE............................................................2
Digital Recorders....................................................2
Audio Formats........................................................2
Low-Cut Filters......................................................2
Fast Fourier Transform...............................................3
Power Spectral Density...............................................3
Long-Term Average Spectrums..........................................4
Sorted Spectrums.....................................................5
III. MATERIALS & METHODS...................................................6
Digital Recorders....................................................6
FFmpeg...............................................................8
VIA U.AB.............................................................8
Testing Procedure....................................................8
IV. ANALYSIS..............................................................12
V. RESULTS...............................................................25
Hypothesis One......................................................25
vii


Hypothesis Two..................................................26
Hypothesis Three................................................27
VI. DISCUSSION & FUTURE RESEARCH......................................29
REFERENCES............................................................30
APPENDIX..............................................................31
viii


LIST OF TABLES
TABLE
1 - Technical Characteristics of the Digital Recorders Used.....................7
2 - CC Values of Olympus VP-10 vs Sample Database..............................13
3 - CC Values of Olympus WS-802 vs Sample Database.............................15
4 - CC Values of Sony ICD-PX370 vs Sample Database.............................17
5 - CC Values of Sony ICD-UX523 vs Sample Database.............................19
6 - CC Values of Zoom H4N 80Hz vs Sample Database..............................21
7 - CC Values of Olympus WS-700M vs Sample Database............................23
8 - CC Test Results............................................................28
IX


LIST OF FIGURES
FIGURE
1 - Example of PSD Analysis....................................................4
2 - Example of LTAS Analysis...................................................4
3 - Example of LTASS Analysis..................................................5
4 - Example of PSD, LTAS, and LTASS Plots Derived from MATLAB..................11
5 - ZOOM H4N Sorted Power Values of all LCF Options............................26
6 - Olympus DM-520 LCF On/Off Comparison.......................................31
7 - Olympus DM-620 LCF On/Off Comparison.......................................31
8 - Olympus DM-901 LCF On/Off Comparison.......................................32
9 - Olympus DS-40 LCF On/Off Comparison........................................32
10 - Olympus VN-8100PC LCF On/Off Comparison...................................33
11 - Olympus VP-10 LCF On/Off Comparison.......................................33
12 - Olympus WS-600S LCF On/Off Comparison.....................................34
13 - Olympus WS-700M LCF On/Off Comparison.....................................34
14 - Olympus WS-852 LCF On/Off Comparison......................................35
15 - Phillips DVT-5500 LCF On/Off Comparison...................................35
16 - Sony ICD-PX312 LCF On/Off Comparison......................................36
17 - Sony ICD-PX370 LCF On/Off Comparison......................................36
18 - Sony ICD-UX512 LCF On/Off Comparison......................................37
19 - Sony ICD-UX523 LCF On/Off Comparison......................................37
20 - Sony ICD-UX533 LCF On/Off Comparison.....................................38
21 - Sony ICD-UX560 LCF On/Off Comparison.....................................38
x


22 - Sony ICD-TX50 LCF On/Off Comparison....................................39
23 - Tascam DR-05 40Hz LCF On/Off Comparison................................39
24 - Tascam DR-05 80Hz LCF On/Off Comparison................................40
25 - Tascam DR-05 120Hz LCF On/Off Comparison...............................40
26 - Tascam DR-07 40Hz LCF On/Off Comparison................................41
27 - Tascam DR-07 80Hz LCF On/Off Comparison................................41
28 - Tascam DR-07 120Hz LCF On/Off Comparison...............................42
29 - Zoom HI LCF On/Off Comparison..........................................42
30 - Zoom H4N 80Hz LCF On/Off Comparison....................................43
31 - Zoom H4N 98Hz LCF On/Off Comparison....................................43
32 - Zoom H4N 115Hz LCF On/Off Comparison...................................44
33 - Zoom H4N 133Hz LCF On/Off Comparison...................................44
34 - Zoom H4N 150Hz LCF On/Off Comparison...................................45
35 - Zoom H4N 168Hz LCF On/Off Comparison.................................45
36 - Zoom H4N 185Hz LCF On/Off Comparison.................................46
37 - Zoom H4N 2031 Iz LCF On/Off Comparison.................................46
38 - Zoom H4N 2201 Iz LCF On/Off Comparison.................................47
39 - Zoom H4N 2371 Iz LCF On/Off Comparison.................................47
40 - Olympus DM-901 vs Olympus DS-40 - LCF On...............................48
41 - Olympus DM-520 vs Olympus DM-620 - LCF On..............................48
42 - Zoom HI vs Zoom H4N 237 - LCF On.......................................49
43 - Tascam DR-05 120Hz vs Tascam DR07 120Hz - LCF On.......................49
44 - Sony ICD-UX560 vs Tascam DR-05 1201 Iz - LCF On........................50
xi


45 - Sony ICD-UX523 vs Sony ICD-UX533 - LCF On.........................50
46 - Sony ICD-UX512 vs Sony ICD-UX523 - LCF On.........................51
47 - Sony ICD-PX370 vs Sony ICD-TX50 - LCF On..........................51
48 - Phillips DVT-5500 vs Sony ICD-PX312 - LCF On......................52
49 - Olympus WS-802 vs Olympus WS-852 - LCF On.........................52
50 - Olympus WS-600S vs Olympus WS-700M - LCF On.......................53
51 - Olympus VN-8100PC vs Olympus VP-10 - LCF On.......................53
52 - Olympus DM-520 vs Sony ICD-PX370 - LCF On.........................54
xii


GLOSSARY OF TERMS & ABBREVIATIONS
LCF - Low-Cut Filter
LTAS - Long-Term Average Spectrum
LTASS - Long-Term Average Spectrum
PSD - Power Spectral Density
CC - Correlation Coefficient
FFT - Fast Fourier Transform
DFFT - Discrete Fast Fourier Transform
SWGDE - Scientific Working Group on Digital Evidence
xm


I. INTRODUCTION
Handheld digital recorders are small, affordable, easy to operate and often come equipped from the manufacturer with a myriad of useful options and features. One of those features is the low-cut filter, an equalization tool with the primary function of reducing noise in a recording to help make the intended source more intelligible. The shape of the low-cut filter slope, or the pattern of its frequency attenuation, could be a potential source of data in an authentication exam. Manufacturers such as Sony, Olympus, Tascam, et al., choose a frequency and a slope, generally expressed in decibels per octave, to assign to the numerous models they manufacture each year. If the LCF pattern can be observed, or identified with LTASS, the analysis could be a powerful tool in an examiners toolbox to apply in conjunction with other analysis methods.
Scope and Intention
The scope of this examination was limited to digital handheld recorders that have a low-cut filter option. Recording samples were created in realistic “real-world” settings, quiet rooms with uncomplex ambient tone. It was not within the scope of this study to use pink noise or other similar calibration techniques. It is the intention of this study to determine the forensic strength of applying long-term average sorted spectrum analysis to an audio signal to observe the LCF slope in order to make comparisons and determine whether these comparisons may lead to an identification or exclusion of a device. Three hypotheses will be examined:
â–  It is possible to identify whether or not the LCF option of a recorder was on or off.
â–  Low-cut filter signal processing differs significantly between recorders.
â–  LTASS/LCF analysis can be used as a means of eliminating a device from a pool of suspected recorders.
1


II. PERSPECTIVE
Digital Recorders
The digital handheld recorders used in this study are generally consumer-level devices marketed for recording speeches, depositions, lectures and other types of close proximity conversations. Law enforcement officers may even carry these devices to help document their interactions with citizens, suspects and witnesses [1], Some models used have extra features for the audio production market, such as external XLR inputs, high sample rates or filtering options. These handheld recorders are particularly commonplace in the field of criminal and civil investigations. Most current models offer removeable/expandable storage and have built in USB plugs or ports for easy file transfers. They are relatively inexpensive, compact, easy to operate and can be relinquished as evidence far more easily than one’s personal mobile phone or tablet [2],
Audio Formats
The recording devices used in this study record digital audio in a variety of formats, both lossy and uncompressed. In the SWGDE document Best Practices for Forensic Audio, whenever performing a forensic examination it is always recommended to avoid degradation of the audio by limiting unnecessary conversions. When transcoding is necessary, one should convert to or between uncompressed formats, such as .WAV, and maintain the sampling rate and bit depth of the original recording [3], Transcoding is the process of de-coding and re-encoding to convert a file into another format [4],
Low-Cut Filters
A low-cut filter, also sometimes referred to as a high-pass filter, is a type of audio filter which removes low frequencies from a signal, usually with the intent of improving the
2


intelligibility of a source in a recording. Frequencies are removed below a determined frequency, and the shape of the frequency attenuation is referred as the slope. Handheld digital devices use these filters quite effectively, and generally the only option available is to turn the filter on or off. It is typical that the slope will begin at a frequency pre-determined by the manufacturer and may or may not be included in the device documentation. In some cases, manufacturers will give the user frequency options for the frequency where filtering should begin, as is the case with Tascam and Zoom brand models included in this examination.
Fast Fourier Transform
The FFT spectrum (Fast Fourier Transform) changes the time domain audio signal into the frequency domain. With the audio represented in this way, data can be plotted with frequency on the horizontal axis and amplitude along the vertical axis [5], The examiner has the option to display the signal as a spectrogram, a waterfall plot, or narrow band frequency display. The analysis can be useful to observe the frequency limits of a recording system, identify precise frequencies and pitches, along with many other characteristics of an audio signal [6],
Power Spectral Density
Power Spectral Density analysis (PSD) is one of the different ways to display a Fourier transform. Also often referred to as a signal’s frequency spectrum, it is a representation of that signal in the frequency domain, presented as magnitude (dB) versus frequency (Hz) [7], Spectral analysis can be helpful in identifying traces of file recompression or inconsistencies between the recorded evidence and the claimed recording device [8],
3


Low Cut Filter (LCF) analysis, PSD
2,
c
QJ
0 50 100 150 200 250 300 350 400 450
frequency [Hz]
Figure 1 - Example of PSD Analysis Long-Term Average Spectrums
The long-term average spectrum (LTAS) is an analysis method which can reveal additional information about a signal related to the environment and acoustics of the recording, the equipment used to create the recording and potential traces of re-compression or processing. It is a mathematical computation that can be used for purposes such as statistical comparisons of different signals. The LTAS is derived by dividing a signal into short time windows, typically overlapping by 50%, then frame functions are applied, FFT and PSD are computed for each frame and the results for each frame are averaged. The product of these processes is a two-column vector consisting of frequency and amplitude. From these values a histogram can be derived to show the number of appearances of each energy level. The global spectrum of the signal can be viewed to verify whether the curve is consistent with a typical recording obtained with the same methods, equipment and settings claimed by the recording operator [4] [6],
Figure 2 - Example of LTAS Analysis
4


Sorted Spectrums
Discrete Fast Fourier Transform (DFFT) computes the minimum, maximum and mean frequencies across all the frames in a recorded signal. These are known as M3 values. A perceptually coded signal will display a steep drop off in magnitude above a certain frequency, which is indicative of a lossy, or compressed, signal, whereas an uncompressed signal will be more consistent across the full range of available frequencies. In a sorted spectrum, the x-axis represents the signal’s frequency range and the y-axis represents their power values. Values can be sorted on a descending order to show signs of lossy compression amongst higher frequencies. When values are sorted on an ascending order, the characterizations of a low-cut filter’s slope can be observed [4] [7],
Figure 3 - Example of LTASS Analysis
5


III. MATERIALS & METHODS
Digital Recorders
Audio samples were collected from twenty-two different models of digital recorders from a variety of manufacturers. The devices were procured from the collections of three examiners and were chosen based on two primary requirements: the device must have an LCF option and access to transfer the digital file sample directly to a computer. The following table shows the record characteristics of each recorder as they were used in this study.
6


Table 1 - Technical Characteristics of the Digital Recorders Used
Recording Characteristics of Digital Recorders Used

Manufactur er Model Recording Format Bit Rate (kbps) Sample Rate (kHz) Bit Depth Chann els LCF Options (Hz)
Olympus DM- 520 .WAV PCM 44.1 16 2
Olympus DM- 620 .WAV PCM 48 16 2
Olympus DM- 901 .WAV PCM 44.1 16 2
Olympus DS-40 .WMA 128 44.1 16 2
Olympus VN- 8100PC .MP3 192 44.1 2
Olympus VP-10 .WAV PCM 22.05 16 2
Olympus WS- 600S .MP3 192 44.1 2
Olympus WS- 700M .WAV PCM 44.1 16 2
Olympus WS- 802 .WAV PCM 44.1 16 2
Olympus WS- 852 .MP3 128 44.1 2
Phillips DVT- 5500 .WAV PCM 44.1 16 2
Sony ICD- PX312 .MP3 192 44.1 2
Sony ICD- PX370 .MP3 192 44.1 2
Sony ICD- TX50 .WAV PCM 44.1 16 2
Sony ICD- UX512 .WAV PCM 44.1 16 2
Sony ICD- UX523 .WAV PCM 44.1 16 2
Sony ICD- UX533 .WAV PCM 44.1 16 2
Sony ICD- UX560 .WAV PCM 44.1 16 2
Tascam DR-05 .WAV PCM 96 24 1 40, 80, 120
Tascam DR-07 .WAV PCM 48 24 2 40, 80, 120
Zoom HI .WAV PCM 96 24 2
Zoom H4N .WAV PCM 44.1 16 2 80, 98, 115, 133, 150, 168, 185, 203, 220, 237
7


FFmpeg
FFmpeg is a command line tool useful for converting, creating, manipulating and streaming digital audio. In this case, it was used for splitting a stereo signal into multiple mono streams and for transcoding a lossy compressed format file into an uncompressed .WAV file. Two commands were primarily used during this examination.
Transcode audio into an uncompressed .WAV format: ffmpeg -i .\input.mp3 -acodec pcm_sl61e -ar 44100 -ac 1 OUT.wav
Splitting a stereo audio stream into two mono streams:
ffmpeg -i .\input.WAV -map_channel 0.0.0 OUTPUT_LEFT.wav -map_channel 0.0.1 OUTPUT_RIGHT.wav
MATLAB
MATLAB is programmable software that can be used to create custom scripts, algorithms and workflows that incorporate peer-reviewed authentication techniques. It served as the backbone of this examination and provided critical comparative data. In this examination, MATLAB was used to calculate a signal’s PSD, LTAS and LTASS, plot the spectrums to a visual figure and save the sorted power values to a document which was later used to create additional comparative periodogram figures as well as determine the correlation coefficient (CC) between two signals.
Testing Procedure
The test recordings were produced only on handheld digital recorders which had an LCF option. The recorders were set to the highest native resolution offered for each device, .WAV
8


PCM if available, but, if not, then the highest resolution of .MP3 or .WMA was selected. The recordings were made in a realistic quiet room environment consisting of uncomplex ambient room tone. Each recording was approximately five to ten minutes in length. It was decided that lengthier file durations would help to eliminate the effects of extremes when averaged. Recording samples were made with the LCF both on and off, and often multiple samples of the same device were taken. When more than one device of the same make and model were available, samples were recorded and labeled as a duplicate. In the cases where there were LCF options, samples were taken for each frequency.
Once recorded, the sample recordings were transferred directly to the examination computer either through an attached USB plug or via a mini USB cable connection. The files were renamed to follow a standardized naming convention:
MAKE MODEL DUPLICATE (WHEN APPLICABLE) LCF ON/OFF LCF FREQ SAMPLE
This convention allowed for easy file sorting to find all instances of samples taken on a specific device.
Once named, the recording samples were evaluated to see whether the files were recorded in a compressed format, and whether they were mono or stereo. If the file was in a compressed format, it was necessary to transcode the file in an uncompressed .WAV format. This was accomplished using FFmpeg in Windows PowerShell. If the ensuing .WAV file was stereo, it was then split into two mono files, with the Left channel (channel 0) being renamed with the addition of MONO at the end and retained for testing, while the Right channel (channel 1) was deleted.
9


The remaining mono .WAV file for each recording sample was then imported into the initial MATLAB script created for this examination. In brief, the script calculation steps were as follows:
â–  Input a mono .WAV audio signal.
â–  Calculate a one-sided spectrum to display the PSD.
â–  Divide the spectrum into short time windows, one second in length and overlapping each other by 50%.
â–  Compute the FFT and PSD for each resulting frame. The FFT order was determined by the sampling frequency of the file being analyzed.
â–  After splitting a signal on N short time windows and computing FFT and PSD, compute the min, max, mean and std values for the N vector.
â–  Sort the power values of the LTAS signal on an ascending scale and save to a text document for separate analysis purposes.
â–  Separately plot the three spectrums, PSD, LTAS and LTASS, to a figure with a limited frequency range up to 500Hz.
10


OLYMPUS-DM-520-LCFoNmONO Low Cut Filter (LCF) analysis, PSD
i o-
S'-20 -§ -40 -
cr
0 50 100 150 200 250 300 350 400 450
frequency [Hz]
Figure 4 - Example of PSD, LTAS, and LTASS Plots Derived from MATLAB
A second MATLAB script was utilized to determine the correlation coefficient between two signals using the sorted power values exported by the initial MATLAB process to measure the statistical relationship between the two variables. This is expressed by a number between 0 and 1 with 1 meaning the variables are identical and 0 representing no correlation whatsoever. These values were then placed in a table to show the CC values of the entire testing database against a chosen exemplar sample.
Additionally, the sorted power values were utilized to plot two spectrums to a figure to create a visual comparison of two audio signals representing a single recorder with the LCF both on and off and of two audio signals from different recorders comparing the LCF slope for each.
11


IV. ANALYSIS
Figures 6 - 39, see Appendix, show the intra-variability of a recorders output signal by comparing the LTASS values of samples taken with the LCF function both on and off for each recording device used in this examination. The intention is to show how LTASS analysis can identify whether or not the LCF function was active while recording. The frequency range is limited to 500Hz, and power values vary for each plot to reflect the minimum value represented in each computation. The solid line represents the spectrum of the signal that had the LCF feature turned off, and the dotted line represents when the LCF feature was instantiated.
Figures 40 - 52, see Appendix, compare the inter-variability of the LTASS values of samples taken from two different recording devices, both with the LCF function on, with the intention of showing how the slope of the LCF differs significantly between recorders. Again, the frequency range is limited to 500Hz, and power values vary for each plot to reflect the minimum value represented in each computation. A legend in the figure identifies which recorder is represented by the solid or dotted line.
Tables 2-7 list the CC value of an exemplar, named at the top of column B, compared with each of the LCF On audio samples created for this examination, all of which are listed in column A. The CC value is a means to show the level of inter-variability within the database. CC values were limited to the fourth digit. The results are sorted in descending order with the cell containing the highest CC value shaded gray, identifying the device whose sorted spectrum has with the highest mathematical relationship to the exemplar.
12


Table 2 - CC Values of Olympus VP-10 vs Sample Database
LCF-On Data Set Correlation to OLYMPUS VP10-
LCF ON A MONO.wav

Standard Deviation: .0387
SONY-ICD-UX560-LCF ON B MONO.wav 0.9969
ZOOM-H4N-LCF ON 133Hz MONO.wav 0.9953
SONY-ICD-UX560-LCF ON A MONO.wav 0.9944
ZOOM-H4N-LCF ON 150Hz MONO.wav 0.9918
TASCAM-DR-07-LCF ON 120Hz A MONO.wav 0.9859
ZOOM-H4N-LCF ON 168Hz MONO.wav 0.9856
OLYMPUS-VP-10-LCF ON C MONO.wav 0.9849
SONY-ICD-PX370-LCF ON B MONO.wav 0.9809
SONY-ICD-UX560-LCF ON C MONO.wav 0.9806
ZOOM-H4N-LCF ON 185Hz MONO.wav 0.9767
OLYMPUS-WS-852-LCF ON B MONO.wav 0.9738
OLYMPUS-WS-802-DUP-LCF ON MONO.wav 0.9736
ZOOM-H1-LCF ON MONO.wav 0.9722
ZOOM-H4N-LCF ON 115Hz MONO.wav 0.9718
TASCAM-DR-05-LCF ON 120Hz MONO 0.9661
OLYMPUS-DM-520-LCF ON MONO.wav 0.9631
OLYMPUS-VP 10-LCF ON B MONO.wav 0.9607
SONY-ICD-PX312-LCF ON MONO.wav 0.9562
SONY-ICD-UX523-LCF ON A MONO.wav 0.9559
OLYMPUS-VN-8100PC-LCF ON MONO.wav 0.9552
OLYMPUS-DM-620-LCF ON MONO.wav 0.9549
ZOOM-H4N-LCF ON 203Hz MONO.wav 0.9547
OLYMPUS-WS-700M-DUP-LCF ON MONO.wav 0.9545
SONY-ICD-UX512-DUP-LCF ON MONO.wav 0.9543
SONY-ICD-UX523-LCF ON B MONO.wav 0.9531
OLYMPUS-WS-600S-LCF ON MONO.wav 0.9528
SONY-ICD-UX523-LCF ON C MONO.wav 0.9524
OLYMPUS-DS-40-LCF ON MONO.wav 0.9523
SONY-ICD-PX370-LCF ON A MONO.wav 0.9499
SONY-ICD-UX5 3 3 -LCF ON MONO.wav 0.9492
TASCAM-DR-07-LCF ON 80Hz B MONO.wav 0.9478
OLYMPUS-DM-901 -LCF ON MONO.wav 0.9460
OLYMPUS-WS-700M-LCF ON MONO.wav 0.9356
ZOOM-H4N-LCF ON 220Hz MONO.wav 0.9352
SONY-ICD-UX512-LCF ON C MONO.wav 0.9339
SONY-ICD-TX5 0-LCF ON MONO.wav 0.9282
OLYMPUS-WS-802-LCF ON C MONO.wav 0.9279
TASCAM-DR-05-LCF ON 80Hz MONO 0.9275
13


Table 2 Continued
LCF-On Data Set Correlation to OLYMPUS VP10-
LCF ON A MONO.wav
ZOOM-H4N-LCF ON 237Hz MONO.wav 0.9255
TASCAM-DR-07-LCF ON 80Hz A MONO.wav 0.9243
OLYMPUS-WS-802-LCF ON A MONO.wav 0.9237
OLYMPUS-WS-802-LCF ON B MONO.wav 0.9233
SONY-ICD-UX512-LCF ON B MONO.wav 0.9227
SONY-ICD-UX512-LCF ON A MONO.wav 0.9199
ZOOM-H4N-LCF ON 80Hz B MONO.wav 0.9122
ZOOM-H4N-LCF ON 80Hz A MONO.wav 0.9002
TASCAM-DR-05-LCF ON 40Hz MONO.wav 0.8920
ZOOM-H4N-LCF ON 98Hz MONO.wav 0.8910
OLYMPUS-WS-852-LCF ON A MONO.wav 0.8855
TASCAM-DR-07-LCF ON 40Hz A MONO.wav 0.8578
PHILLIPS-DVT-5500-LCF ON MONO.wav 0.7797
14


Table 3 - CC Values of Olympus WS-802 vs Sample Database
LCF-On Data Set Correlation to OLYMPUS-WS-802-LCF ON_A MONO.wav

Standard Deviation: .0716
OLYMPUS-WS-802-LCF ON B MONO.wav 0.9997
OLYMPUS-WS-802-LCF ON C MONO.wav 0.9993
OLYMPUS-WS-700M-LCF ON MONO.wav 0.9965
SONY-ICD-TX50-LCF ON MONO.wav 0.9951
SONY-ICD-UX512-LCF ON C MONO.wav 0.9888
SONY-ICD-UX512-LCF ON B MONO.wav 0.9881
OLYMPUS-WS-700M-DUP-LCF ON MONO.wav 0.9872
SONY-ICD-UX512-LCF ON A MONO.wav 0.9869
OLYMPUS-WS-600S-LCF ON MONO.wav 0.9862
SONY-ICD-UX5 3 3 -LCF ON MONO.wav 0.9839
OLYMPUS-DM-520-LCF ON MONO.wav 0.9816
SONY-ICD-PX370-LCF ON B MONO.wav 0.9772
SONY-ICD-PX312-LCF ON MONO.wav 0.9771
SONY-ICD-UX523-LCF ON C MONO.wav 0.9769
SONY-ICD-UX512-DUP-LCF ON MONO.wav 0.9769
SONY-ICD-UX523-LCF ON B MONO.wav 0.9756
OLYMPUS-WS-802-DUP-LCF ON MONO.wav 0.9743
SONY-ICD-UX523-LCF ON A MONO.wav 0.9723
OLYMPUS-DM-620-LCF ON MONO.wav 0.9632
OLYMPUS-VP10-LCF ON B MONO.wav 0.9533
OLYMPUS-DM-901 -LCF ON MONO.wav 0.9531
OLYMPUS-DS-40-LCF ON MONO.wav 0.9519
TASCAM-DR-07-LCF ON 120Hz A MONO.wav 0.9482
SONY-ICD-UX560-LCF ON C MONO.wav 0.9347
OLYMPUS-VP-10-LCF ON C MONO.wav 0.9336
OLYMPUS-VP 10-LCF ON A MONO.wav 0.9237
SONY-ICD-UX560-LCF ON B MONO.wav 0.9173
ZOOM-H4N-LCF ON 133Hz MONO.wav 0.9140
ZOOM-H4N-LCF ON 168Hz MONO.wav 0.9127
ZOOM-H4N-LCF ON 150Hz MONO.wav 0.9097
OLYMPUS-WS-852-LCF ON B MONO.wav 0.9092
ZOOM-H4N-LCF ON 185Hz MONO.wav 0.9073
ZOOM-H1-LCF ON MONO.wav 0.9045
ZOOM-H4N-LCF ON 203Hz MONO.wav 0.8963
SONY-ICD-UX560-LCF ON A MONO.wav 0.8943
ZOOM-H4N-LCF ON 220Hz MONO.wav 0.8916
OLYMPUS-VN-8100PC-LCF ON MONO.wav 0.8859
ZOOM-H4N-LCF ON 237Hz MONO.wav 0.8858
15


Table 3 Continued
LCF-On Data Set Correlation to OLYMPUS-WS-
802-LCF ON_A MONO.wav
TASCAM-DR-05-LCF ON 80Hz MONO 0.8816
ZOOM-H4N-LCF ON 115Hz MONO.wav 0.8787
SONY-ICD-PX3 70-LCF ON A MONO.wav 0.8726
TASCAM-DR-05-LCF ON 120Hz MONO 0.8644
OLYMPUS-WS-852-LCF ON A MONO.wav 0.8574
TASCAM-DR-05-LCF ON 40Hz MONO.wav 0.8490
TASCAM-DR-07-LCF ON 80Hz B MONO.wav 0.8460
ZOOM-H4N-LCF ON 80Hz B MONO.wav 0.8238
TASCAM-DR-07-LCF ON 80Hz A MONO.wav 0.8136
ZOOM-H4N-LCF ON 80Hz A MONO.wav 0.7861
ZOOM-H4N-LCF ON 98Hz MONO.wav 0.7739
TASCAM-DR-07-LCF ON 40Hz A MONO.wav 0.7308
PHILLIPS-DVT-5500-LCF ON MONO.wav 0.6955
16


Table 4 - CC Values ofSonyICD-PX370 vs Sample Database
LCF-On Data Set Correlation to SONY-ICD-PX370-LCF ON_A MONO.wav

Standard Deviation: .0724
ZOOM-H4N-LCF ON 203Hz MONO.wav 0.9908
ZOOM-H4N-LCF ON 220Hz MONO.wav 0.9901
ZOOM-H4N-LCF ON 237Hz MONO.wav 0.9892
ZOOM-H4N-LCF ON 185Hz MONO.wav 0.9864
ZOOM-H4N-LCF ON 168Hz MONO.wav 0.9775
OLYMPUS-WS-852-LCF ON A MONO.wav 0.9773
OLYMPUS-WS-852-LCF ON B MONO.wav 0.9754
OLYMPUS-DS-40-LCF ON MONO.wav 0.9684
TASCAM-DR-07-LCF ON 120Hz A MONO.wav 0.9649
SONY-ICD-UX560-LCF ON A MONO.wav 0.9617
OLYMPUS-DM-901 -LCF ON MONO.wav 0.9601
OLYMPUS-DM-620-LCF ON MONO.wav 0.9588
SONY-ICD-UX560-LCF ON B MONO.wav 0.9545
ZOOM-H4N-LCF ON 133Hz MONO.wav 0.9542
OLYMPUS-VP10-LCF ON A MONO.wav 0.9499
OLYMPUS-WS-802-DUP-LCF ON MONO.wav 0.9453
SONY-ICD-UX512-DUP-LCF ON MONO.wav 0.9427
SONY-ICD-PX312-LCF ON MONO.wav 0.9364
SONY-ICD-PX370-LCF ON B MONO.wav 0.9322
SONY-ICD-UX5 3 3 -LCF ON MONO.wav 0.9317
ZOOM-H4N-LCF ON 150Hz MONO.wav 0.9232
OLYMPUS-VN-8100PC-LCF ON MONO.wav 0.9221
SONY-ICD-UX523-LCF ON C MONO.wav 0.9215
SONY-ICD-UX523-LCF ON B MONO.wav 0.9177
OLYMPUS-DM-520-LCF ON MONO.wav 0.9149
SONY-ICD-UX523-LCF ON A MONO.wav 0.9135
TASCAM-DR-05-LCF ON 120Hz MONO 0.9100
SONY-ICD-UX512-LCF ON C MONO.wav 0.9096
OLYMPUS-VP-10-LCF ON C MONO.wav 0.9012
OLYMPUS-WS-700M-DUP-LCF ON MONO.wav 0.9007
SONY-ICD-UX512-LCF ON B MONO.wav 0.9001
ZOOM-H1-LCF ON MONO.wav 0.8995
SONY-ICD-TX5 0-LCF ON MONO.wav 0.8974
SONY-ICD-UX512-LCF ON A MONO.wav 0.8974
OLYMPUS-WS-700M-LCF ON MONO.wav 0.8948
OLYMPUS-WS-600S-LCF ON MONO.wav 0.8948
SONY-ICD-UX560-LCF ON C MONO.wav 0.8828
OLYMPUS-WS-802-LCF ON C MONO.wav 0.8773
17


Table 4 Continued
LCF-On Data Set Correlation to SONY-ICD-
PX370-LCF ON_A MONO.wav
ZOOM-H4N-LCF ON 115Hz MONO.wav 0.8727
OLYMPUS-WS-802-LCF ON A MONO.wav 0.8726
OLYMPUS-WS-802-LCF ON B MONO.wav 0.8709
OLYMPUS-VP10-LCF ON B MONO.wav 0.8587
TASCAM-DR-07-LCF ON 80Hz B MONO.wav 0.8301
TASCAM-DR-05-LCF ON 80Hz MONO 0.8255
TASCAM-DR-07-LCF ON 80Hz A MONO.wav 0.7851
TASCAM-DR-05-LCF ON 40Hz MONO.wav 0.7806
ZOOM-H4N-LCF ON 80Hz B MONO.wav 0.7757
ZOOM-H4N-LCF ON 80Hz A MONO.wav 0.7718
ZOOM-H4N-LCF ON 98Hz MONO.wav 0.7425
TASCAM-DR-07-LCF ON 40Hz A MONO.wav 0.7335
PHILLIPS-DVT-5500-LCF ON MONO.wav 0.6817
18


Table 5 - CC Values of Sony ICD-UX523 vs Sample Database
LCF-On Data Set Correlation to SONY-ICD-UX523-LCF ON_A MONO.wav

Standard Deviation: .0483
SONY-ICD-UX523-LCF ON B MONO.wav 0.9987
SONY-ICD-UX523-LCF ON C MONO.wav 0.9971
OLYMPUS-DM-520-LCF ON MONO.wav 0.9907
SONY-ICD-UX512-LCF ON C MONO.wav 0.9878
SONY-ICD-UX512-LCF ON A MONO.wav 0.9850
SONY-ICD-UX512-LCF ON B MONO.wav 0.9835
SONY-ICD-PX370-LCF ON B MONO.wav 0.9790
OLYMPUS-WS-802-DUP-LCF ON MONO.wav 0.9788
SONY-ICD-UX512-DUP-LCF ON MONO.wav 0.9786
SONY-ICD-UX5 3 3 -LCF ON MONO.wav 0.9783
OLYMPUS-DM-901 -LCF ON MONO.wav 0.9768
OLYMPUS-WS-802-LCF ON C MONO.wav 0.9747
OLYMPUS-WS-802-LCF ON B MONO.wav 0.9737
OLYMPUS-WS-600S-LCF ON MONO.wav 0.9728
OLYMPUS-WS-700M-DUP-LCF ON MONO.wav 0.9728
TASCAM-DR-07-LCF ON 120Hz A MONO.wav 0.9728
OLYMPUS-WS-700M-LCF ON MONO.wav 0.9724
OLYMPUS-WS-802-LCF ON A MONO.wav 0.9723
SONY-ICD-TX50-LCF ON MONO.wav 0.9723
SONY-ICD-PX312-LCF ON MONO.wav 0.9703
OLYMPUS-DM-620-LCF ON MONO.wav 0.9651
OLYMPUS-DS-40-LCF ON MONO.wav 0.9593
OLYMPUS-VP10-LCF ON B MONO.wav 0.9569
OLYMPUS-VP10-LCF ON A MONO.wav 0.9559
OLYMPUS-VP-10-LCF ON C MONO.wav 0.9545
OLYMPUS-WS-852-LCF ON B MONO.wav 0.9515
SONY-ICD-UX560-LCF ON C MONO.wav 0.9508
ZOOM-H4N-LCF ON 133Hz MONO.wav 0.9473
ZOOM-H4N-LCF ON 150Hz MONO.wav 0.9473
OLYMPUS-VN-8100PC-LCF ON MONO.wav 0.9468
SONY-ICD-UX560-LCF ON B MONO.wav 0.9467
ZOOM-H1-LCF ON MONO.wav 0.9447
ZOOM-H4N-LCF ON 168Hz MONO.wav 0.9427
SONY-ICD-UX560-LCF ON A MONO.wav 0.9369
ZOOM-H4N-LCF ON 185Hz MONO.wav 0.9359
TASCAM-DR-05-LCF ON 80Hz MONO 0.9285
ZOOM-H4N-LCF ON 115Hz MONO.wav 0.9259
TASCAM-DR-05-LCF ON 120Hz MONO 0.9248
19


Table 5 Continued
LCF-On Data Set Correlation to SONY-ICD-
UX523-LCF ON_A MONO.wav
ZOOM-H4N-LCF ON 203Hz MONO.wav 0.9190
SONY-ICD-PX3 70-LCF ON A MONO.wav 0.9135
ZOOM-H4N-LCF ON 220Hz MONO.wav 0.9110
ZOOM-H4N-LCF ON 237Hz MONO.wav 0.9065
TASCAM-DR-07-LCF ON 80Hz B MONO.wav 0.9062
TASCAM-DR-05-LCF ON 40Hz MONO.wav 0.8831
OLYMPUS-WS-852-LCF ON A MONO.wav 0.8788
ZOOM-H4N-LCF ON 80Hz B MONO.wav 0.8786
TASCAM-DR-07-LCF ON 80Hz A MONO.wav 0.8681
ZOOM-H4N-LCF ON 80Hz A MONO.wav 0.8650
ZOOM-H4N-LCF ON 98Hz MONO.wav 0.8424
TASCAM-DR-07-LCF ON 40Hz A MONO.wav 0.8179
PHILLIPS-DVT-5500-LCF ON MONO.wav 0.7630
20


Table 6 - CC Values of Zoom H4N 80Hz vs Sample Database
LCF-On Data Set Correlation to ZOOM-H4N-LCF_ONJOHz_A_MONO.wav

Standard Deviation: .0818
ZOOM-H4N-LCF ON 98Hz MONO.wav 0.9914
ZOOM-H4N-LCF ON 80Hz B MONO.wav 0.9895
TASCAM-DR-07-LCF ON 80Hz B MONO.wav 0.9828
TASCAM-DR-07-LCF ON 80Hz A MONO.wav 0.9812
TASCAM-DR-07-LCF ON 40Hz A MONO.wav 0.9811
TASCAM-DR-05-LCF ON 80Hz MONO 0.9692
ZOOM-H4N-LCF ON 115Hz MONO.wav 0.9690
TASCAM-DR-05-LCF ON 120Hz MONO 0.9496
TASCAM-DR-05-LCF ON 40Hz MONO.wav 0.9482
ZOOM-H1-LCF ON MONO.wav 0.9450
ZOOM-H4N-LCF ON 150Hz MONO.wav 0.9322
SONY-ICD-UX560-LCF ON C MONO.wav 0.9317
OLYMPUS-VP-10-LCF ON C MONO.wav 0.9278
OLYMPUS-VN-8100PC-LCF ON MONO.wav 0.9211
OLYMPUS-VP 10-LCF ON B MONO.wav 0.9135
PHILLIPS-DVT-5500-LCF ON MONO.wav 0.9035
OLYMPUS-VP 10-LCF ON A MONO.wav 0.9002
SONY-ICD-UX560-LCF ON B MONO.wav 0.8889
ZOOM-H4N-LCF ON 133Hz MONO.wav 0.8880
SONY-ICD-UX560-LCF ON A MONO.wav 0.8829
SONY-ICD-UX523-LCF ON A MONO.wav 0.8650
TASCAM-DR-07-LCF ON 120Hz A MONO.wav 0.8649
OLYMPUS-DM-520-LCF ON MONO.wav 0.8641
SONY-ICD-PX370-LCF ON B MONO.wav 0.8595
OLYMPUS-WS-852-LCF ON B MONO.wav 0.8585
SONY-ICD-UX523-LCF ON B MONO.wav 0.8487
SONY-ICD-UX523-LCF ON C MONO.wav 0.8469
ZOOM-H4N-LCF ON 168Hz MONO.wav 0.8360
OLYMPUS-WS-802-DUP-LCF ON MONO.wav 0.8339
OLYMPUS-WS-600S-LCF ON MONO.wav 0.8328
OLYMPUS-WS-700M-DUP-LCF ON MONO.wav 0.8297
ZOOM-H4N-LCF ON 185Hz MONO.wav 0.8112
SONY-ICD-PX312-LCF ON MONO.wav 0.8047
SONY-ICD-UX512-DUP-LCF ON MONO.wav 0.8023
SONY-ICD-TX5 0-LCF ON MONO.wav 0.7971
SONY-ICD-UX512-LCF ON C MONO.wav 0.7936
SONY-ICD-UX5 3 3 -LCF ON MONO.wav 0.7915
OLYMPUS-WS-802-LCF ON C MONO.wav 0.7910
21


Table 6 Continued
LCF-On Data Set Correlation to ZOOM-H4N-
LCF ON_80Hz A MONO.wav
OLYMPUS-WS-700M-LCF ON MONO.wav 0.7890
OLYMPUS-WS-802-LCF ON B MONO.wav 0.7882
OLYMPUS-DM-901 -LCF ON MONO.wav 0.7880
OLYMPUS-DM-620-LCF ON MONO.wav 0.7880
OLYMPUS-WS-802-LCF ON A MONO.wav 0.7861
SONY-ICD-UX512-LCF ON B MONO.wav 0.7777
OLYMPUS-DS-40-LCF ON MONO.wav 0.7761
SONY-ICD-UX512-LCF ON A MONO.wav 0.7759
SONY-ICD-PX3 70-LCF ON A MONO.wav 0.7718
ZOOM-H4N-LCF ON 203Hz MONO.wav 0.7560
ZOOM-H4N-LCF ON 220Hz MONO.wav 0.7193
ZOOM-H4N-LCF ON 237Hz MONO.wav 0.7043
OLYMPUS-WS-852-LCF ON A MONO.wav 0.6500
22


Table 7- CC Values of Olympus WS-700M vs Sample Database
LCF-On Data Set Correlation to OLYMPUS-WS-700M-LCF_ON_MONO.wav

Standard Deviation: .0710
SONY-ICD-TX50-LCF ON MONO.wav 0.9976
OLYMPUS-WS-802-LCF ON C MONO.wav 0.9970
OLYMPUS-WS-802-LCF ON A MONO.wav 0.9965
OLYMPUS-WS-802-LCF ON B MONO.wav 0.9960
OLYMPUS-WS-700M-DUP-LCF ON MONO.wav 0.9927
OLYMPUS-WS-600S-LCF ON MONO.wav 0.9910
SONY-ICD-UX5 3 3 -LCF ON MONO.wav 0.9905
SONY-ICD-UX512-LCF ON C MONO.wav 0.9883
SONY-ICD-UX512-LCF ON B MONO.wav 0.9865
SONY-ICD-PX312-LCF ON MONO.wav 0.9860
SONY-ICD-UX512-DUP-LCF ON MONO.wav 0.9849
SONY-ICD-UX512-LCF ON A MONO.wav 0.9845
OLYMPUS-DM-520-LCF ON MONO.wav 0.9844
OLYMPUS-WS-802-DUP-LCF ON MONO.wav 0.9843
SONY-ICD-PX370-LCF ON B MONO.wav 0.9835
SONY-ICD-UX523-LCF ON C MONO.wav 0.9778
SONY-ICD-UX523-LCF ON B MONO.wav 0.9759
OLYMPUS-DM-620-LCF ON MONO.wav 0.9752
SONY-ICD-UX523-LCF ON A MONO.wav 0.9724
OLYMPUS-DS-40-LCF ON MONO.wav 0.9644
OLYMPUS-DM-901 -LCF ON MONO.wav 0.9614
TASCAM-DR-07-LCF ON 120Hz A MONO.wav 0.9593
OLYMPUS-VP-10-LCF ON C MONO.wav 0.9403
SONY-ICD-UX560-LCF ON C MONO.wav 0.9397
OLYMPUS-VP 10-LCF ON A MONO.wav 0.9356
OLYMPUS-VP 10-LCF ON B MONO.wav 0.9356
SONY-ICD-UX560-LCF ON B MONO.wav 0.9315
ZOOM-H4N-LCF ON 168Hz MONO.wav 0.9291
ZOOM-H4N-LCF ON 133Hz MONO.wav 0.9273
ZOOM-H4N-LCF ON 185Hz MONO.wav 0.9254
OLYMPUS-WS-852-LCF ON B MONO.wav 0.9239
ZOOM-H4N-LCF ON 150Hz MONO.wav 0.9205
ZOOM-H4N-LCF ON 203Hz MONO.wav 0.9162
ZOOM-H1-LCF ON MONO.wav 0.9127
ZOOM-H4N-LCF ON 220Hz MONO.wav 0.9120
SONY-ICD-UX560-LCF ON A MONO.wav 0.9106
ZOOM-H4N-LCF ON 237Hz MONO.wav 0.9065
OLYMPUS-VN-8100PC-LCF ON MONO.wav 0.8961
23


Table 7 Continued
LCF-On Data Set Correlation to OLYMPUS-WS-
700M-LCF_ON_MONO.wav
SONY-ICD-PX3 70-LCF ON A MONO.wav 0.8948
ZOOM-H4N-LCF ON 115Hz MONO.wav 0.8865
TASCAM-DR-05-LCF ON 80Hz MONO 0.8836
OLYMPUS-WS-852-LCF ON A MONO.wav 0.8796
TASCAM-DR-05-LCF ON 120Hz MONO 0.8769
TASCAM-DR-07-LCF ON 80Hz B MONO.wav 0.8500
TASCAM-DR-05-LCF ON 40Hz MONO.wav 0.8481
ZOOM-H4N-LCF ON 80Hz B MONO.wav 0.8244
TASCAM-DR-07-LCF ON 80Hz A MONO.wav 0.8185
ZOOM-H4N-LCF ON 80Hz A MONO.wav 0.7890
ZOOM-H4N-LCF ON 98Hz MONO.wav 0.7771
TASCAM-DR-07-LCF ON 40Hz A MONO.wav 0.7335
PHILLIPS-DVT-5500-LCF ON MONO.wav 0.6934
24


V. RESULTS
The aim of this study was to determine whether sufficient salient data could be derived from the LCF effects in audio recordings made with handheld digital recorders using LTASS analysis to reliably make comparisons, exclusions or possible identifications. Three hypotheses were examined using the data created for this examination.
Hypothesis One
It is possible to identify whether or not the LCF option of a recorder was on or off.
This hypothesis has been confirmed to be true. Samples were made with the LCF both on and off for each recorder used in the study. When processed with the initial analysis script in MATLAB, the power values of the LTASS were exported and saved to a text document. The on and off power values were then plotted to a periodogram. These figures show the inter-variability of a recorder’s output, LCF on vs off.
As can be visually observed in figures 6 - 39, the signal with the LCF turned on differs greatly from the signal with the LCF turned off, with varying degrees of significant differentiation. Figure 21, showing the spectrums of a Sony ICD-TX50 recorder, is an excellent example to show a clear distinction of LCF on vs. off. Other examples are more subtle, such as figure 10, showing the spectrums of an Olympus VP-10. All devices tested from manufacturers Olympus, Sony, Phillips and Zoom exhibited expected results. An interesting characteristic of the Zoom H4N model, the device with the highest number of frequency roll-off options, did not seem to exhibit any substantial difference in frequency among the samples, only changes in the slope severity (figures 30 - 39). Below is a figure showing the overlaid sorted power values of all the LCF options; 80Hz, 98Hz, 115Hz, 133Hz, 150Hz, 168Hz, 185Hz, 203Hz, 220Hz, and 237Hz.
25


Figure 5 - ZOOM H4N Sorted Power Values of all LCF Options
Additionally, the Tascam DR-05 model showed less low frequency activity in the LCF off sample than in the samples taken with the LCF on and set to the three frequency options of 40Hz, 80Hz, and 120Hz, seen in figures 23 - 25. Additional samples or even additional DR-05 devices in a future study could help understand this anomaly. However, visually, there was a clear frequency slope drop seen in figure 25 which shows the sorted spectrum with the LCF set to 120Hz.
Hypothesis Two
Low-cut filter signal processing differs significantly between recorders.
This hypothesis has been confirmed to be true. To compare the inter-variability of the LTASS values of samples taken from two different recording devices, sorted power values from two different recorders, both with the LCF on, were overlaid in a single periodogram. As shown in figures 40 - 52, in most cases, there is a clear distinction between the LCF processing of one
26


recorder to the other. It is particularly evident in figure 47, which shows the spectrums for the Sony ICD-PX370 and the Sony ICD-TX50. A more subtle example of LCF slope distinction can be seen in figure 41 where the Olympus DM-520 and Olympus DM-620 are compared.
Some inter-variability comparisons displayed interesting results. In figure 46, the Sony ICD-UX512 spectrum is very similar to that of the Sony ICD-UX523. A similar event is seen in figure 50 between the spectrums of the Olympus WS-600S and the Olympus WS-700M. A possible explanation is that the manufacturers copied much of the processing architecture to a subsequent model, giving them similar LCF functionality.
Hypothesis Three
LTASS/LCF analysis can be used as a means of eliminating a device from a pool of
recorders.
This hypothesis has been confirmed to be true. To test whether LTASS analysis can reliably identify the LCF signature of a purported recording device, correlation coefficients were calculated for a single device sample against all of the sample library used in this study. CC values were charted for six comparative studies shown in tables 2-7. The highest CC value was highlighted to show which recorder sample had the highest mathematical relationship to the exemplar. The results are as follows:
27


Table 8 - CC Test Results
Exemplar Recorder Recorder with Highest CC Sample
Test 1 OLYMPUS VP 10-LCF ON A MONO.wav SONY-ICD-UX560-LCF ON B MONO.wav
Test 2 OLYMPUS-WS-802-LCF ON A MONO.wav OLYMPUS-WS-802-LCF ON B MONO.wav
Test 3 SONY-ICD-PX370-LCF ON A MONO.wav OLYMPUS-WS-852-LCF ON A MONO.wav
Test 4 SONY-ICD-UX523-LCF ON A MONO.wav SONY-ICD-UX523-LCF ON B MONO.wav
Test 5 ZOOM-H4N- LCF ON 80Hz A MONO.wav ZOOM-H4N- LCF ON 98Hz MONO.wav
Test 6 OLYMPUS-WS-700M-LCF ON MONO.wav SONY-ICD-TX50-LCF ON MONO.wav
Tests 2, 4 and 5 were accurately able to determine which recording device was used to create the exemplar recording because another audio sample made with the same recorder measured the highest CC values. However, tests 1, 3 and 6 resulted in a make and model of recorder other than the exemplar as having the highest CC value.
While LTASS analysis cannot reliably identify the LCF signature of specific handheld digital recorders, it may still be used to make an exclusion through spectrum comparisons on a periodogram. The scope of this study only included recorders with the LCF option, but there are numerous recorders on the market that do not have this option. Should a signal after analysis show evidence of an LCF, but the purported recorder does not have this option, this information could be useful to an examiner.
28


VI. DISCUSSION & FUTURE RESEARCH
This study was able to show that LTASS analysis can be a very useful technique when analyzing the effects of an LCF on an audio signal from a handheld digital recorder. Its use does have limitations, but when used in conjunction with other recommended techniques can provide valuable data for an investigation.
Possible areas of interest for future research could be:
â–  Samples may be created in a calibrated environment involving pink noise or frequency sweeps.
■ Samples may be created in more varied “real-world” situations to test how background noise effects the reliability of LTASS analysis.
â–  Samples may be created at the extremes of available resolution in a recorder to further test intra-variability.
29


REFERENCES
1. Maher, R.C. and S.R. Shaw, Gunshot Recordings from Digital Voice Recorders, in 54th International Conference: Audio Forensics. 2014: London, England.
2. Pruett, S., An Exploratory Study of Manual Pause-Record Events in Digital Audio Recordings. 2019, University of Colorado Denver.
3. SWGDE, Best Practices for Forensic Audio, Version 2.2. October, 2016.
4. Rappaport, D.L., Establishing a standardfor digital audio authenticity : a critical analysis of tools, methodologies, and challenges. 2012, University of Colorado Denver.
5. Koenig, B.E., D.S. Lacey, and S.A. Killion, Forensic Enhancement of Digital Audio Recordings. Journal of the Audio Engineering Society, 2007. 55(5): p. 352-370.
6. Grigoras, C., D.L. Rappaport, and J.M. Smith, Analytical Framework for Digital Audio Authentication, in AES 46th International Conference: Audio Forensics. 2012: Denver, CO, USA.
7. Grigoras, C., Statistical Tools for Multimedia Forensics, in 39 th International Conference: Audio Forensics: Practices and Challenges. 2010: Hillerod, Denmark.
8. Grigoras, C. and J.M. Smith, Audio Enhancement and Authentication, in Encyclopedia of Forensic Sciences. 2013. p. 315-326.
30


APPENDIX
Figure 6 - Olympus DM-520 LCF On/Off Comparison
Figure 7 - Olympus DM-620 LCF On/Off Comparison
31


Figure 8 - Olympus DM-901 LCF On/Off Comparison
Figure 9 - Olympus DS-40 LCF On/Off Comparison
32


Figure 10 - Olympus VN-8100PC LCF On/Off Comparison
Figure 11 - Olympus VP-10 LCF On/Off Comparison
33


-40
OLYMPUS WS-600S On/Off Comparison of Sorted Power Values
0 100 200 300 400 500
Frequency(Hz)
Figure 12 - Olympus WS-600S LCF On/Off Comparison
Figure 13 - Olympus WS-700M LCF On/Off Comparison
34


Figure 14 - Olympus WS-852 LCF On/Off Comparison
Figure 15 - Phillips DVT-5500 LCF On/Off Comparison
35


-40
SONY ICD-PX312 On/Off Comparison of Sorted Power Values
-45 -
0 100 200 300 400 500
Frequency(Hz)
Figure 16 - Sony 1CD-PX312 LCF On/Off Comparison
Figure 17 - Sony ICD-PX3 70 LCF On/O ff Comparison
36


Figure 18 - Sony ICD-UX512 LCF On/Off Comparison
Figure 19 - Sony 1CD-UX523 LCF On/Off Comparison
37


Figure 20 - Sony ICD-UX533 LCF On/Off Comparison
Figure 21 - Sony ICD-UX560 LCF On/Off Comparison
38


Figure 22 - Sony ICD-TX50 LCF On/Off Comparison
Figure 23 - Tascam DR-05 40Hz LCF On/Off Comparison
39


Figure 24 - Tascam DR-05 80HzLCF On/Off Comparison
Figure 25 - Tascam DR-05 120Hz LCF On/Off Comparison
40


Figure 26 - Tascam DR-07 40HzLCF On/Off Comparison
Figure 27 - Tascam DR-07 80HzLCF On/Off Comparison
41


Figure 28 - Tascam DR-07 120Hz LCF On/Off Comparison
Figure 29 - Zoom HI LCF On/Off Comparison
42


Figure 30 - Zoom H4N 80Hz LCF On/Off Comparison
Figure 31 - Zoom H4N 98HzLCF On/Off Comparison
43


Figure 32 - Zoom H4N 115Hz LCF On/Off Comparison
Figure 33 - Zoom H4N 133Hz LCF On/Off Comparison
44


Figure 34- Zoom H4N 150Hz LCF On/Off Comparison
Figure 35 - Zoom H4N168Hz LCF On/Off Comparison
45


Figure 36- Zoom H4N 185Hz LCF On/Off Comparison
Figure 37 - Zoom H4N 203Hz LCF On/Off Comparison
46


Figure 38 - Zoom H4N 220Hz LCF On/Off Comparison
Figure 39 - Zoom H4N 23 7Hz LCF On/Off Comparison
47


Figure 40 - Olympus DM-901 vs Olympus DS-40 - LCF On
Figure 41 - Olympus DM-520 vs Olympus DM-620 - LCF On
48


Figure 42 - Zoom HI vs Zoom H4N237 - LCF On
Figure 43 - Tascam DR-05 120Hz vs Tascam DR07120Hz - LCF On
49


Figure 44 - Sony ICD-UX560 vs Tascam DR-05 120Hz - LCF On
Figure 45 - Sony ICD-UX523 vs Sony ICD-UX533 - LCF On
50


Figure 46 - Sony ICD-UX512 vs Sony ICD-UX523 - LCF On
Figure 47- Sony ICD-PX370 vs Sony ICD-TX50 -LCF On
51


Figure 48 - Phillips DVT-5500 vs Sony ICD-PX312 - LCF On
Figure 49 - Olympus WS-802 vs Olympus WS-852 - LCF On
52


Figure 50 - Olympus WS-600S vs Olympus WS-700M-LCF On
Figure 51 - Olympus VN-8100PC vs Olympus VP-10 - LCF On
53


Figure 52 - Olympus DM-520 vs Sony ICD-PX3 70 - LCF On
54


Full Text

PAGE 1

CHARACTERIZATIONS OF LOW CUT FILTERS IN HANDHELD DIGITAL RECORDERS USING LONGTERM AVERAGE SORTED SPECTRUM ANALYSIS by DANTE ALEXANDER FAZIO B.M., Berklee College of Music, 2006 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirement s for the degree of Master of Science Recording Arts Program 2019

PAGE 2

ii 2019 DANTE ALEXANDER FAZIO ALL RIGHTS RESERVED

PAGE 3

iii This thesis for the Master of Science degree by Dante Alexander Fazio has been approved for the Recording Arts Program by Catalin Grigoras , Chair Jeff M. Smith Durand Begault Date : December 14, 2019

PAGE 4

iv Fazio, Dante Alexander (M.S ., Recording Arts Program ) Characterizations of Low Cut Filters in Handheld Digital Recorders Using LongTerm Average Sorted Spectrum Analysis Thesis directed by Associate Professor Catalin Grigoras ABSTRACT A n audio signal's longterm average sorted spectrum can be used to reveal information about the source of a recording, which, in the case of this study , will focus on the low cut filter option of small handheld digital recording devices. These devices are small, affordable, easy to operate and often come equipped from the manufacturer with a myriad of useful features , such as the low cut filter. The shape of the low cut filter slope could be a potential source of data in an authentication exam. It is the intention of this study to determine the forensic strength of appl ying longterm average sorted spectrum analysis to an audio signal to observe the slope of a low cut filter in order to make comparisons and determine whether they may lead to an identification or an exclusion of a device. Three hypotheses will be examine d: It is possible to identify whether or not the LCF option of a recorder was on or off. Low cut filter signal processing differs significantly between recorders. LTASS/LCF analysis can be used as a means of eliminating a device from a pool of suspected recorders. Th e form and content of this abstract are approved. I recommend its publication. Approved: Catalin Grigora s

PAGE 5

v DEDICATION To my family for their love, unyielding support , and genuine enthusiasm in whatever shiny thing I may be interested in that week. To my beautiful wife Monica , whose assistance in my completion of this degree program cannot be understated. I love you so much. T hank you for watching the farm while I was off becoming a scientist. To our pets, Kingston, Lady, Sherpa, Billie, Duke, Buster, Levi and Midnight. Were it not for your relentless appeals for affection, this thesis would have been completed in half the time.

PAGE 6

vi ACKNOWLEDGMENTS I would like to thank my teachers, Dr. Catalin Grigoras and Jeff Smith , for shepherding my education in this exciting field. Their knowledge and enthusiasm for media forensics is both impressive and contagious, and it has been a pleasure being their student. I would also like to recognize Leah Haloin a nd Cole Whitecotton for their extraordinary contributions to the NCMF. Without them, the NCMF would be dark and full of terrors. I must thank Dr. Durand Begault for being so generous with his time and resources and for introduc ing me to the NCMF, and it is an honor to have him on my thesis committee. A huge thank you to Doug las Lacey , my classmate, unofficial instructor, good friend, and extremely generous contributor to the data set for my thesis , for all the knowledge he share d with us despite not receiving a stipend aside for an occasional dirty martini. Thank you to my classmates for the camaraderie, the exchange of ideas , and for making the evening lectures something to regret in the morning .

PAGE 7

vii TABLE OF CONTENTS C HAP TER I. INTRODUCTION ................................................................................................................... 1 Scope and Intention ............................................................................................................. 1 II. PERSPECTIVE ....................................................................................................................... 2 Digital Recorders ................................................................................................................ 2 Audio Formats .................................................................................................................... 2 Low Cut Filters ................................................................................................................... 2 Fast Fourier Transform ....................................................................................................... 3 Power Spectral Density ....................................................................................................... 3 LongTerm Average Spectrums .......................................................................................... 4 Sorted Spectrums ................................................................................................................ 5 III. MATERIALS & METHODS .................................................................................................. 6 Digital Recorders ................................................................................................................ 6 FFmpeg ............................................................................................................................... 8 MATLAB ............................................................................................................................ 8 Testing Procedure ............................................................................................................... 8 IV. ANALYSIS ........................................................................................................................... 12 V. RESULTS .............................................................................................................................. 25 Hypothesis One ................................................................................................................. 25

PAGE 8

viii Hypothesis Two ................................................................................................................ 26 Hypothesis Three .............................................................................................................. 27 VI. DISCUSSION & FUTURE RESEARCH ............................................................................. 29 REFERENCES ............................................................................................................................. 30 APPENDIX ................................................................................................................................... 31

PAGE 9

ix LIST OF TABLES TABLE 1 – Technical C haracteristics of the D igital R ecorders U sed ......................................................... 7 2 – CC V alues of Olympus VP –10 vs Sample Database .............................................................. 13 3 – CC V alues of Olympus WS –802 vs Sample Database ........................................................... 15 4 – CC V alues of Sony ICD –PX370 vs Sample Database ........................................................... 17 5 – CC V alues of Sony ICD –UX523 vs Sample Database ........................................................... 19 6 – CC V alues of Zoom H4N 80Hz vs Sample Database ............................................................. 21 7 – CC V alues of Olympus WS –700M vs Sample Database ....................................................... 23 8 – CC Test Results ....................................................................................................................... 28

PAGE 10

x LIST OF FIGURES FIGURE 1 – Example of PSD Analysis ......................................................................................................... 4 2 – Example of LTAS Analysis ...................................................................................................... 4 3 – Example of LTASS Analysis .................................................................................................... 5 4 – Example of PSD, LTAS, and LTASS P lots D erived from MATLAB ................................... 11 5 – ZOOM H4N Sorted Power Values of all LCF Options .......................................................... 26 6 – Olympus DM 520 LCF On/Off Comparison .......................................................................... 31 7 – Olympus DM 620 LCF On/Off Comparison .......................................................................... 31 8 – Olympus DM 901 LCF On/Off Comparison .......................................................................... 32 9 – Olympus DS 40 LCF On/Off Comparison ............................................................................. 32 10 – Olympus VN 8100PC LCF On/Off Comparison ................................................................. 33 11 – Olympus VP 10 LCF On/Off Comparison ........................................................................... 33 12 – Olympus WS 600S LCF On/Off Comparison ...................................................................... 34 13 – Olympus WS 700M LCF On/Off Comparison ..................................................................... 34 14 – Olympus WS 852 LCF On/Off Comparison ........................................................................ 35 15 – Phillips DVT 5500 LCF On/Off Comparison ...................................................................... 35 16 – Sony ICD PX312 LCF On/Off Comparison......................................................................... 36 17 – Sony ICD PX370 LCF On/Off Comparison......................................................................... 36 18 – Sony ICD UX512 LCF On/Off Comparison ........................................................................ 37 19 – Sony ICD UX523 LCF On/Off Comparison ........................................................................ 37 20 – Sony ICD UX533 LCF On/Off Comparison ........................................................................ 38 21 – Sony ICD UX560 LCF On/Off Comparison ........................................................................ 38

PAGE 11

xi 22 – Sony ICD TX50 LCF On/Off Comparison .......................................................................... 39 23 – Tascam DR 05 40Hz LCF On/Off Comparison ................................................................... 39 24 – Tascam DR 05 80Hz LCF On/Off Comparison ................................................................... 40 25 – Tascam DR 05 120Hz LCF On/Off Comparison ................................................................. 40 26 – Tascam DR 07 40Hz LCF On/Off Comparison ................................................................... 41 27 – Tascam DR 07 80Hz LCF On/Off Comparison ................................................................... 41 28 – Tascam DR 07 120Hz LCF On/Off Comparison ................................................................. 42 29 – Zoom H1 LCF On/Off Comparison ...................................................................................... 42 30 Zoom H4N 80Hz LCF On/Off Comparison .......................................................................... 43 31 – Zoom H4N 98Hz LCF On/Off Comparison ......................................................................... 43 32 – Zoom H4N 115Hz LCF On/Off Comparison ....................................................................... 44 33 – Zoom H4N 133Hz LCF On/Off Comparison ....................................................................... 44 34 – Zoom H4N 150Hz LCF On/Off Comparison ....................................................................... 45 35 – Zoom H4N 168Hz LCF On/Off Comparison ....................................................................... 45 36 – Zoom H4N 185Hz LCF On/Off Comparison ....................................................................... 46 37 – Zoom H4N 203Hz LCF On/Off Comparison ....................................................................... 46 38 – Zoom H4N 220Hz LCF On/Off Comparison ....................................................................... 47 39 – Zoom H4N 237Hz LCF On/Off Comparison ....................................................................... 47 40 – Olympus DM 901 vs Olympus DS 40 LCF On ................................................................. 48 41 – Olympus DM 520 vs Olympus DM 620 LCF On ............................................................. 48 42 – Zoom H1 vs Zoom H4N 237 LCF On ................................................................................ 49 43 – Tascam DR 05 120Hz vs Tascam DR07 120Hz LCF On .................................................. 49 44 – Sony ICD UX560 vs Tascam DR 05 120Hz LCF On ....................................................... 50

PAGE 12

xii 45 – Sony ICD UX523 vs Sony ICD UX533 LCF On .............................................................. 50 46 – Sony ICD UX512 vs Sony ICD UX523 LCF On .............................................................. 51 47 – Sony ICD PX370 vs Sony ICD TX50 LCF On ................................................................. 51 48 – Phillips DVT 5500 vs Sony ICD PX312 LCF On ............................................................. 52 49 – Olympus WS 802 vs Olympus WS 852 LCF On .............................................................. 52 50 – Olympus WS 600S vs Olympus WS 700M LCF On ......................................................... 53 51 – Olympus VN 8100PC vs Olympus VP 10 LCF On ........................................................... 53 52 – Olympus DM 520 vs Sony ICD PX370 LCF On .............................................................. 54

PAGE 13

xiii GLOSSARY OF TERMS & ABBREVIATIONS LCF – Low –Cut Filter LTAS – Long –Term Average Spectrum LTASS – Long–Term Average Spectrum PSD – Power Spectral Density CC – Correlation Coefficient FFT – Fast Fourier Transform DFFT – Discrete Fast Fourier Transform SWGDE – Scientific Working Group on Digital Evidence

PAGE 14

1 I. INTRODUCTION Handheld digital recorders are small, afforda ble, easy to operate and often come equipped from the manufacturer with a myriad of useful options and features. One of those features is the low cut filter, an equalization tool with the primary function of reducing noise in a recording to help make the intended source more intelligible. The shape of the low cut filter slope, or the pattern of its frequency attenuation, could be a potential source of data in an authentication exam. Manufacturers such as Sony, Olympus, Tascam, et al., choose a frequency and a slope, generally expressed in decibels per octave, to assign to the numerous models they manufacture each year. If the LCF pattern can be observed, or identified with LTASS, the analysis could be a powerful tool in an examiners toolbox to apply in c onjunction with other analysis methods. Scope and Intention The scope of this examination was limited to digital handheld recorders that have a low cut filter option. Recording samples were created in realistic “real world” settings, quiet rooms with uncomplex ambient tone. It was not within the scope of this study to use pink noise or other similar calibration techniques. It is the intention of this study to determine the forensic strength of applying longterm average sorted spectrum analysis to an audio signal to observe the LCF slope in order to make comparisons and determine whether these comparisons may lead to an identification or exclusion of a device. Three hypotheses will be examined: It is possible to identify whether or not the LCF option of a recorder was on or off. Low cut filter signal processing differs significantly between recorders. LTASS/LCF analysis can be used as a means of eliminating a device from a pool of suspected recorders.

PAGE 15

2 II. PERSPECTIVE Digital Recorders The digital handheld recorders used in this study are generally consumer level devices marketed for recording speeches, depositions, lectures and other types of close proximity conversations. Law enforcement officers may even carry these devices to help document their intera ctions with citizens, suspects and witnesses [1] . Some models used have extra features for the audio production market, such as external XLR inputs, high sample rates or filtering options . These handheld recorders are particularly commonplace in the field of criminal and civil investigations. Most current models offer removeable/expandable storage and have built in USB plugs or ports for easy file transfers. They are relatively inexpensive, compact , easy to operate and can be relinquished as evidence far more easily than one’s personal mobile phone or tablet [2] . Audio Formats The recording devices used in this study record digital audio in a variety of formats, both lossy and uncompressed . In the SWGDE document Best Practices for Forensic Audio , whenever performing a forensic examination it is always recommended to avoid degra dation of the audio by limiting unnecessary conversions. When transcoding is necessary, one should convert to or between uncompressed formats, such as .WAV, and maintain the sampling rate and bit depth of the original recording [3] . Transcoding is the process of de coding and re encoding to convert a file into another format [4] . LowCut Filters A lowcut filter, also sometimes referred to as a high pass filter , is a type of audio filter which remove s low frequencies from a signal, usually with the intent of improving the

PAGE 16

3 intelligibility of a source in a recording. Frequencies are removed below a determined frequency, and the shape of the frequency attenuation is referred as the slope. Handheld digital devices use these filters quite effectively, and generally the only option available is to turn the filter on or off. It is typical that the slope will begin at a frequency pre determined by the manufacturer and may or may not be included in the device documentation. In some cases , manufacturers will give the user frequency options for the frequency where filtering should begin, as is the case with Tascam and Zoom brand models included in this examination. Fast Fourier Transform The FFT spectrum ( Fast Fourier Transform ) changes the time domain audio signal into the frequency domain. With the audio represented in this way, data can be plotted with frequency on the horizontal axis and amplitude along the vertical axis [5] . The examiner has the option to display the signal as a spectrogram, a waterfall plot, or narrow band frequency display. The analysis can be useful to observe the frequency limits of a recording system, identify precise frequencies and pitches, along with many other character istics of an audio signal [6] . Power Spectral Density Power Spectral Density analysis (PSD) is one of the different w ays to display a Fourier transform. Also often referred to as a signal’s frequency spectrum, it is a representation of that signal in the frequency domain, presented as magnitude (dB) versus frequency (Hz) [7] . Spectral analysis can be helpful in identifying traces of file recompression or inconsistencies between the recorded evidence and the claimed recording device [8] .

PAGE 17

4 Figure 1 – Example of PSD Analysis Long Term Average Spectrums The longterm average spectrum (LTAS) is an analysis method which can reveal additional info rmation about a signal related to the environment and acoustics of the recording, the equipment used to create the recording and potential traces of re compression or processing. It is a mathematical computation that can be used for purposes such as stati stical comparisons of different signals. The LTAS is derived by dividing a signal into short time windows, typically overlapping by 50%, then frame functions are applied, FFT and PSD are computed for each frame and the results for each frame are averaged. The product of these processes is a two column vector consisting of frequency and amplitude. From these values a histogram can be derived to show the number of appearances of each energy level. The global spectrum of the signal can be viewed to verify whether the curve is consistent with a typical recording obtained with the same methods, equipment and settings claimed by the recording operator [4] [6] . Figure 2 – Example of LTAS Analysis

PAGE 18

5 Sorted Spectrums Discrete Fast Fourier Transform (DFFT) computes the minimum, maximum and mean frequencies across all the frames in a recorded signal. These are known as M3 values. A perceptually coded signal will display a steep drop off in magnitude above a certain frequency, which is indicative of a lossy, or compressed, signal, whereas an uncompressed signal w ill be more consistent across the full range of available frequencies. In a sorted spectrum, the x axis represents the signal’s frequency range and the yaxis represent s their power values. Values can be sorted on a descending order to show signs of lossy com pression amongst higher frequencies . W hen values are sorted on an ascending order, the characterizations of a low cut filter’s slope can be observed [4] [7] . Figure 3 – Example of LTASS Analysis

PAGE 19

6 III. MATERIALS & METHODS Digital Recorders A udio samples were collected from twenty two different models of digital recorders from a variety of manufacturers. The devices were procured from the collections of three examiners and were chosen based on two primary requirements : the device must have an LCF option and access to transfer the digital file sample directly to a computer. The following table shows the record characteristics of each recorder as they were used in this study.

PAGE 20

7 Table 1 – Technical Characteristics of the D igital R ecorders U sed Recording Characteristics of Digital Recorders Used Manufactur er Model Recording Format Bit Rate (kbps) Sample Rate (kHz) Bit Depth Chann els LCF Options (Hz) Olympus DM 520 .WAV PCM 44.1 16 2 Olympus DM 620 .WAV PCM 48 16 2 Olympus DM 901 .WAV PCM 44.1 16 2 Olympus DS 40 .WMA 128 44.1 16 2 Olympus VN 8100PC .MP3 192 44.1 2 Olympus VP 10 .WAV PCM 22.05 16 2 Olympus WS 600S .MP3 192 44.1 2 Olympus WS 700M .WAV PCM 44.1 16 2 Olympus WS 802 .WAV PCM 44.1 16 2 Olympus WS 852 .MP3 128 44.1 2 Phillips DVT 5500 .WAV PCM 44.1 16 2 Sony ICD PX312 .MP3 192 44.1 2 Sony ICD PX370 .MP3 192 44.1 2 Sony ICD TX50 .WAV PCM 44.1 16 2 Sony ICD UX512 .WAV PCM 44.1 16 2 Sony ICD UX523 .WAV PCM 44.1 16 2 Sony ICD UX533 .WAV PCM 44.1 16 2 Sony ICD UX560 .WAV PCM 44.1 16 2 Tascam DR 05 .WAV PCM 96 24 1 40, 80, 120 Tascam DR 07 .WAV PCM 48 24 2 40, 80, 120 Zoom H1 .WAV PCM 96 24 2 Zoom H4N .WAV PCM 44.1 16 2 80, 98, 115, 133, 150, 168, 185, 203, 220, 237

PAGE 21

8 FFmpeg FFmpeg is a command line tool useful for converting, creating, manipulating and streaming digital audio. In this case, it was used for splitting a stereo signal into multiple mono streams and f or transcoding a lossy compressed format file into a n uncompressed .WAV file . Two commands were pr imarily used during this examination . Transcode audio into an uncompressed .WAV format: ffmpeg -i .\input.mp3 -acodec pcm_s16le -ar 44100 -ac 1 OUT.wav Splitting a stereo audio stream into two mono streams: ffmpeg -i .\input.WAV -map_channel 0.0.0 OUTPUT_LEFT.wav map_channel 0.0.1 OUTPUT_RIGHT.wav MATLAB MATLAB is programmable software that can be used to create custom scripts, algorithms and workflows that incorporate peerreviewed authentication techniques. It served as the backbone of this examination and provided critical comparative data. In this examination, MATLAB was used to calculate a signal’s PSD, LTAS and LTASS, plot the spectrums to a visual figure and save the sorted power values to a document which was later used to create additional comparative periodogram figures as well as determine the correlation coefficient (CC) between two signals. Testing Procedure The t est recordings were produc ed only on handheld digital recorders which had a n LCF option. The recorders were set to the highest native resolution offered for each device , .WAV

PAGE 22

9 PCM if available , but , if not , then the highest resolution of .MP3 or .WMA was selected . The recordings w ere made in a realistic quiet room environment consisting of uncomplex ambient room tone. Each recording was approximately five to ten minute s in length. It was decided that lengthier file durations would help to eliminate the effects of extreme s when averaged. Recording samples were made with the LCF both on and off, and often multiple samples of the same device were taken. When more than one device of the same make and model were available, samples were recorded and labeled as a duplicate. In the cases where there were LCF options, samples were taken for each frequency. Once recorded, t he sample recordings were transferred directly to the examination computer either through an attached USB plug or via a mini USB cable connection. The files were renam ed to follow a standardized naming convention: MAKE_MODEL_DUPLICATE (WHEN APPLICABLE)_LCF ON/OFF_LCF FREQ_SAMPLE This convention allowed for easy file sorting to find all instances of samples taken on a specific device. Once named, the recording samples were evaluated to see whether the files were recorded in a compressed format, and whether they were mono or stereo. If the file was in a compressed format, it was necessary to transcode the file in an uncompressed .WAV format. This was accomplished using F Fmpeg in Windows PowerShell . If the ensuing .WAV file was stereo, it w as then split into two mono files, with the Left channel (channel 0) being renamed with the addition of _MONO at the end and retained for testing, while the Right channel (channel 1) was deleted.

PAGE 23

10 The remaining mono .WAV file for each recording sample was then imported into the initial MATLAB script created for this examination. In brief, the script calculation steps were as follows: Input a mono .WAV audio signal. Calculate a onesided spectrum to display the PSD . Divide the spectrum into short time windows , one second in length and overlapping each other by 50%. Compute t he FFT and PSD for each resulting frame. The FFT order was determined by the sampling frequency of the file being analyzed. After splitting a signal on N short time windows and computing FFT and PSD, compute the min, max, mean and std values for the N vector. Sort the power values of the LTAS signal on an ascending scale and save to a text document for separate analysis purposes. Separately plot t he three spectrums, PSD, LTAS and LTASS, to a figure with a limited frequency range up to 500Hz.

PAGE 24

11 Figure 4 – Example of PSD, LTAS, and LTASS P lots D erived from MATLAB A second MATLAB script was utilized to determine the correlation coefficient between two signals using the sorted power values exported by the initial MATLAB process to measure the statistical relationship between the two variables. This is expressed by a number between 0 and 1 with 1 meaning the variables are identical and 0 representing no correlation whatsoever. These values were then placed in a table to show the CC values of the entire testing database against a chosen exemplar sample. Additionally , the sorted power values were utilized to plot two spectrums to a figure to create a visual comparison of two audio signals representing a single recorder with the LCF both on and off and of two audio signals from different recorders comparing the LCF slope for each.

PAGE 25

12 IV. ANALYSIS Figures 6 – 39, see A ppendix, show the intra variability of a rec orders output signal by compar ing the LTASS values of samples taken with the LCF function both on and off for each recording device used in this examination. The intention is to show how LTASS analysis can identify whether or not the LCF function was acti ve while recording. The frequency range is limited to 500Hz, and power values vary for each plot to reflect the minimum value represented in each computation. The solid line represents the spectrum of the signal that had the LCF feature turned off, and t he dotted line represents when the LCF feature was instantiated. Figures 40 – 52, see A ppendix, compare the inter variability of the LTASS values of samples taken from two different recording devices, both with the LCF function on, with the intention of showing how the slope of the LCF differs significantly between recorders. Again, t he frequency range is limited to 500Hz, and power values vary for each plot to reflect the minimum value represented in each computation. A legend in the figure identifies which recorder is represented by the solid or dotted line. Tables 2 – 7 list the CC value of an exemplar , named at the top of column B, compared with each of the LCF On audio samples created for this examination , all of which are listed in col umn A . The CC value is a means to show the level of inter variability within the database. CC values were limited to the fourth digit. The results are sorted in descending order with t he cell containing the highest CC value shaded gray, identifying the device whose sorted spectrum has with the highest mathematical relationship to the exemplar.

PAGE 26

13 Table 2 – CC V alues of Olympus VP -10 vs Sample Database LCF -On Data Set Correlation to OLYMPUS_VP10 LCF_ON_A_MONO.wav Standard Deviation: .0387 SONY ICD UX560 LCF_ON_B_MONO.wav 0.9969 ZOOM H4N LCF_ON_133Hz_MONO.wav 0.9953 SONY ICD UX560 LCF_ON_A_MONO.wav 0.9944 ZOOM H4N LCF_ON_150Hz_MONO.wav 0.9918 TASCAM DR 07 LCF_ON_120Hz_A_MONO.wav 0.9859 ZOOM H4N LCF_ON_168Hz_MONO.wav 0.9856 OLYMPUS VP 10 LCF_ON_C_MONO.wav 0.9849 SONY ICD PX370 LCF_ON_B_MONO.wav 0.9809 SONY ICD UX560 LCF_ON_C_MONO.wav 0.9806 ZOOM H4N LCF_ON_185Hz_MONO.wav 0.9767 OLYMPUS WS 852 LCF_ON_B_MONO.wav 0.9738 OLYMPUS WS 802 DUP LCF_ON_MONO.wav 0.9736 ZOOM H1 LCF_ON_MONO.wav 0.9722 ZOOM H4N LCF_ON_115Hz_MONO.wav 0.9718 TASCAM DR 05 LCF_ON_120Hz_MONO 0.9661 OLYMPUS DM 520 LCF_ON_MONO.wav 0.9631 OLYMPUS VP10 LCF_ON_B_MONO.wav 0.9607 SONY ICD PX312 LCF_ON_MONO.wav 0.9562 SONY ICD UX523 LCF_ON_A_MONO.wav 0.9559 OLYMPUS VN 8100PC LCF_ON_MONO.wav 0.9552 OLYMPUS DM 620 LCF_ON_MONO.wav 0.9549 ZOOM H4N LCF_ON_203Hz_MONO.wav 0.9547 OLYMPUS WS 700M DUP LCF_ON_MONO.wav 0.9545 SONY ICD UX512 DUP LCF_ON_MONO.wav 0.9543 SONY ICD UX523 LCF_ON_B_MONO.wav 0.9531 OLYMPUS WS 600S LCF_ON_MONO.wav 0.9528 SONY ICD UX523 LCF_ON_C_MONO.wav 0.9524 OLYMPUS DS 40 LCF_ON_MONO.wav 0.9523 SONY ICD PX370 LCF_ON_A_MONO.wav 0.9499 SONY ICD UX533 LCF_ON_MONO.wav 0.9492 TASCAM DR 07 LCF_ON_80Hz_B_MONO.wav 0.9478 OLYMPUS DM 901 LCF_ON_MONO.wav 0.9460 OLYMPUS WS 700M LCF_ON_MONO.wav 0.9356 ZOOM H4N LCF_ON_220Hz_MONO.wav 0.9352 SONY ICD UX512 LCF_ON_C_MONO.wav 0.9339 SONY ICD TX50 LCF_ON_MONO.wav 0.9282 OLYMPUS WS 802 LCF_ON_C_MONO.wav 0.9279 TASCAM DR 05 LCF_ON_80Hz_MONO 0.9275

PAGE 27

14 Table 2 Continued LCF -On Data Set Correlation to OLYMPUS_VP10 LCF_ON_A_MONO.wav ZOOM H4N LCF_ON_237Hz_MONO.wav 0.9255 TASCAM DR 07 LCF_ON_80Hz_A_MONO.wav 0.9243 OLYMPUS WS 802 LCF_ON_A_MONO.wav 0.9237 OLYMPUS WS 802 LCF_ON_B_MONO.wav 0.9233 SONY ICD UX512 LCF_ON_B_MONO.wav 0.9227 SONY ICD UX512 LCF_ON_A_MONO.wav 0.9199 ZOOM H4N LCF_ON_80Hz_B_MONO.wav 0.9122 ZOOM H4N LCF_ON_80Hz_A_MONO.wav 0.9002 TASCAM DR 05 LCF_ON_40Hz_MONO.wav 0.8920 ZOOM H4N LCF_ON_98Hz_MONO.wav 0.8910 OLYMPUS WS 852 LCF_ON_A_MONO.wav 0.8855 TASCAM DR 07 LCF_ON_40Hz_A_MONO.wav 0.8578 PHILLIPS DVT 5500 LCF_ON_MONO.wav 0.7797

PAGE 28

15 Table 3 – CC V alues of Olympus WS -802 vs Sample Database LCF -On Data Set Correlation to OLYMPUS -WS 802-LCF_ON_A_MONO.wav Standard Deviation: .0716 OLYMPUS WS 802 LCF_ON_B_MONO.wav 0.9997 OLYMPUS WS 802 LCF_ON_C_MONO.wav 0.9993 OLYMPUS WS 700M LCF_ON_MONO.wav 0.9965 SONY ICD TX50 LCF_ON_MONO.wav 0.9951 SONY ICD UX512 LCF_ON_C_MONO.wav 0.9888 SONY ICD UX512 LCF_ON_B_MONO.wav 0.9881 OLYMPUS WS 700M DUP LCF_ON_MONO.wav 0.9872 SONY ICD UX512 LCF_ON_A_MONO.wav 0.9869 OLYMPUS WS 600S LCF_ON_MONO.wav 0.9862 SONY ICD UX533 LCF_ON_MONO.wav 0.9839 OLYMPUS DM 520 LCF_ON_MONO.wav 0.9816 SONY ICD PX370 LCF_ON_B_MONO.wav 0.9772 SONY ICD PX312 LCF_ON_MONO.wav 0.9771 SONY ICD UX523 LCF_ON_C_MONO.wav 0.9769 SONY ICD UX512 DUP LCF_ON_MONO.wav 0.9769 SONY ICD UX523 LCF_ON_B_MONO.wav 0.9756 OLYMPUS WS 802 DUP LCF_ON_MONO.wav 0.9743 SONY ICD UX523 LCF_ON_A_MONO.wav 0.9723 OLYMPUS DM 620 LCF_ON_MONO.wav 0.9632 OLYMPUS VP10 LCF_ON_B_MONO.wav 0.9533 OLYMPUS DM 901 LCF_ON_MONO.wav 0.9531 OLYMPUS DS 40 LCF_ON_MONO.wav 0.9519 TASCAM DR 07 LCF_ON_120Hz_A_MONO.wav 0.9482 SONY ICD UX560 LCF_ON_C_MONO.wav 0.9347 OLYMPUS VP 10 LCF_ON_C_MONO.wav 0.9336 OLYMPUS VP10 LCF_ON_A_MONO.wav 0.9237 SONY ICD UX560 LCF_ON_B_MONO.wav 0.9173 ZOOM H4N LCF_ON_133Hz_MONO.wav 0.9140 ZOOM H4N LCF_ON_168Hz_MONO.wav 0.9127 ZOOM H4N LCF_ON_150Hz_MONO.wav 0.9097 OLYMPUS WS 852 LCF_ON_B_MONO.wav 0.9092 ZOOM H4N LCF_ON_185Hz_MONO.wav 0.9073 ZOOM H1 LCF_ON_MONO.wav 0.9045 ZOOM H4N LCF_ON_203Hz_MONO.wav 0.8963 SONY ICD UX560 LCF_ON_A_MONO.wav 0.8943 ZOOM H4N LCF_ON_220Hz_MONO.wav 0.8916 OLYMPUS VN 8100PC LCF_ON_MONO.wav 0.8859 ZOOM H4N LCF_ON_237Hz_MONO.wav 0.8858

PAGE 29

16 Table 3 Continued LCF -On Data Set Correlation to OLYMPUS -WS 802-LCF_ON_A_MONO.wav TASCAM DR 05 LCF_ON_80Hz_MONO 0.8816 ZOOM H4N LCF_ON_115Hz_MONO.wav 0.8787 SONY ICD PX370 LCF_ON_A_MONO.wav 0.8726 TASCAM DR 05 LCF_ON_120Hz_MONO 0.8644 OLYMPUS WS 852 LCF_ON_A_MONO.wav 0.8574 TASCAM DR 05 LCF_ON_40Hz_MONO.wav 0.8490 TASCAM DR 07 LCF_ON_80Hz_B_MONO.wav 0.8460 ZOOM H4N LCF_ON_80Hz_B_MONO.wav 0.8238 TASCAM DR 07 LCF_ON_80Hz_A_MONO.wav 0.8136 ZOOM H4N LCF_ON_80Hz_A_MONO.wav 0.7861 ZOOM H4N LCF_ON_98Hz_MONO.wav 0.7739 TASCAM DR 07 LCF_ON_40Hz_A_MONO.wav 0.7308 PHILLIPS DVT 5500 LCF_ON_MONO.wav 0.6955

PAGE 30

17 Table 4 – CC V alues of Sony ICD -PX370 vs Sample Database LCF -On Data Set Correlation to SONY -ICD PX370-LCF_ON_A_MONO.wav Standard Deviation: .0724 ZOOM H4N LCF_ON_203Hz_MONO.wav 0.9908 ZOOM H4N LCF_ON_220Hz_MONO.wav 0.9901 ZOOM H4N LCF_ON_237Hz_MONO.wav 0.9892 ZOOM H4N LCF_ON_185Hz_MONO.wav 0.9864 ZOOM H4N LCF_ON_168Hz_MONO.wav 0.9775 OLYMPUS WS 852 LCF_ON_A_MONO.wav 0.9773 OLYMPUS WS 852 LCF_ON_B_MONO.wav 0.9754 OLYMPUS DS 40 LCF_ON_MONO.wav 0.9684 TASCAM DR 07 LCF_ON_120Hz_A_MONO.wav 0.9649 SONY ICD UX560 LCF_ON_A_MONO.wav 0.9617 OLYMPUS DM 901 LCF_ON_MONO.wav 0.9601 OLYMPUS DM 620 LCF_ON_MONO.wav 0.9588 SONY ICD UX560 LCF_ON_B_MONO.wav 0.9545 ZOOM H4N LCF_ON_133Hz_MONO.wav 0.9542 OLYMPUS VP10 LCF_ON_A_MONO.wav 0.9499 OLYMPUS WS 802 DUP LCF_ON_MONO.wav 0.9453 SONY ICD UX512 DUP LCF_ON_MONO.wav 0.9427 SONY ICD PX312 LCF_ON_MONO.wav 0.9364 SONY ICD PX370 LCF_ON_B_MONO.wav 0.9322 SONY ICD UX533 LCF_ON_MONO.wav 0.9317 ZOOM H4N LCF_ON_150Hz_MONO.wav 0.9232 OLYMPUS VN 8100PC LCF_ON_MONO.wav 0.9221 SONY ICD UX523 LCF_ON_C_MONO.wav 0.9215 SONY ICD UX523 LCF_ON_B_MONO.wav 0.9177 OLYMPUS DM 520 LCF_ON_MONO.wav 0.9149 SONY ICD UX523 LCF_ON_A_MONO.wav 0.9135 TASCAM DR 05 LCF_ON_120Hz_MONO 0.9100 SONY ICD UX512 LCF_ON_C_MONO.wav 0.9096 OLYMPUS VP 10 LCF_ON_C_MONO.wav 0.9012 OLYMPUS WS 700M DUP LCF_ON_MONO.wav 0.9007 SONY ICD UX512 LCF_ON_B_MONO.wav 0.9001 ZOOM H1 LCF_ON_MONO.wav 0.8995 SONY ICD TX50 LCF_ON_MONO.wav 0.8974 SONY ICD UX512 LCF_ON_A_MONO.wav 0.8974 OLYMPUS WS 700M LCF_ON_MONO.wav 0.8948 OLYMPUS WS 600S LCF_ON_MONO.wav 0.8948 SONY ICD UX560 LCF_ON_C_MONO.wav 0.8828 OLYMPUS WS 802 LCF_ON_C_MONO.wav 0.8773

PAGE 31

18 Table 4 Continued LCF -On Data Set Correlation to SONY -ICD PX370-LCF_ON_A_MONO.wav ZOOM H4N LCF_ON_115Hz_MONO.wav 0.8727 OLYMPUS WS 802 LCF_ON_A_MONO.wav 0.8726 OLYMPUS WS 802 LCF_ON_B_MONO.wav 0.8709 OLYMPUS VP10 LCF_ON_B_MONO.wav 0.8587 TASCAM DR 07 LCF_ON_80Hz_B_MONO.wav 0.8301 TASCAM DR 05 LCF_ON_80Hz_MONO 0.8255 TASCAM DR 07 LCF_ON_80Hz_A_MONO.wav 0.7851 TASCAM DR 05 LCF_ON_40Hz_MONO.wav 0.7806 ZOOM H4N LCF_ON_80Hz_B_MONO.wav 0.7757 ZOOM H4N LCF_ON_80Hz_A_MONO.wav 0.7718 ZOOM H4N LCF_ON_98Hz_MONO.wav 0.7425 TASCAM DR 07 LCF_ON_40Hz_A_MONO.wav 0.7335 PHILLIPS DVT 5500 LCF_ON_MONO.wav 0.6817

PAGE 32

19 Table 5 – CC V alues of Sony ICD -UX523 vs Sample Database LCF -On Data Set Correlation to SONY-ICD UX523 -LCF_ON_A_MONO.wav Standard Deviation: .0483 SONY ICD UX523 LCF_ON_B_MONO.wav 0.9987 SONY ICD UX523 LCF_ON_C_MONO.wav 0.9971 OLYMPUS DM 520 LCF_ON_MONO.wav 0.9907 SONY ICD UX512 LCF_ON_C_MONO.wav 0.9878 SONY ICD UX512 LCF_ON_A_MONO.wav 0.9850 SONY ICD UX512 LCF_ON_B_MONO.wav 0.9835 SONY ICD PX370 LCF_ON_B_MONO.wav 0.9790 OLYMPUS WS 802 DUP LCF_ON_MONO.wav 0.9788 SONY ICD UX512 DUP LCF_ON_MONO.wav 0.9786 SONY ICD UX533 LCF_ON_MONO.wav 0.9783 OLYMPUS DM 901 LCF_ON_MONO.wav 0.9768 OLYMPUS WS 802 LCF_ON_C_MONO.wav 0.9747 OLYMPUS WS 802 LCF_ON_B_MONO.wav 0.9737 OLYMPUS WS 600S LCF_ON_MONO.wav 0.9728 OLYMPUS WS 700M DUP LCF_ON_MONO.wav 0.9728 TASCAM DR 07 LCF_ON_120Hz_A_MONO.wav 0.9728 OLYMPUS WS 700M LCF_ON_MONO.wav 0.9724 OLYMPUS WS 802 LCF_ON_A_MONO.wav 0.9723 SONY ICD TX50 LCF_ON_MONO.wav 0.9723 SONY ICD PX312 LCF_ON_MONO.wav 0.9703 OLYMPUS DM 620 LCF_ON_MONO.wav 0.9651 OLYMPUS DS 40 LCF_ON_MONO.wav 0.9593 OLYMPUS VP10 LCF_ON_B_MONO.wav 0.9569 OLYMPUS VP10 LCF_ON_A_MONO.wav 0.9559 OLYMPUS VP 10 LCF_ON_C_MONO.wav 0.9545 OLYMPUS WS 852 LCF_ON_B_MONO.wav 0.9515 SONY ICD UX560 LCF_ON_C_MONO.wav 0.9508 ZOOM H4N LCF_ON_133Hz_MONO.wav 0.9473 ZOOM H4N LCF_ON_150Hz_MONO.wav 0.9473 OLYMPUS VN 8100PC LCF_ON_MONO.wav 0.9468 SONY ICD UX560 LCF_ON_B_MONO.wav 0.9467 ZOOM H1 LCF_ON_MONO.wav 0.9447 ZOOM H4N LCF_ON_168Hz_MONO.wav 0.9427 SONY ICD UX560 LCF_ON_A_MONO.wav 0.9369 ZOOM H4N LCF_ON_185Hz_MONO.wav 0.9359 TASCAM DR 05 LCF_ON_80Hz_MONO 0.9285 ZOOM H4N LCF_ON_115Hz_MONO.wav 0.9259 TASCAM DR 05 LCF_ON_120Hz_MONO 0.9248

PAGE 33

20 Table 5 Continued LCF -On Data Set Correlation to SONY -ICD UX523 -LCF_ON_A_MONO.wav ZOOM H4N LCF_ON_203Hz_MONO.wav 0.9190 SONY ICD PX370 LCF_ON_A_MONO.wav 0.9135 ZOOM H4N LCF_ON_220Hz_MONO.wav 0.9110 ZOOM H4N LCF_ON_237Hz_MONO.wav 0.9065 TASCAM DR 07 LCF_ON_80Hz_B_MONO.wav 0.9062 TASCAM DR 05 LCF_ON_40Hz_MONO.wav 0.8831 OLYMPUS WS 852 LCF_ON_A_MONO.wav 0.8788 ZOOM H4N LCF_ON_80Hz_B_MONO.wav 0.8786 TASCAM DR 07 LCF_ON_80Hz_A_MONO.wav 0.8681 ZOOM H4N LCF_ON_80Hz_A_MONO.wav 0.8650 ZOOM H4N LCF_ON_98Hz_MONO.wav 0.8424 TASCAM DR 07 LCF_ON_40Hz_A_MONO.wav 0.8179 PHILLIPS DVT 5500 LCF_ON_MONO.wav 0.7630

PAGE 34

21 Table 6 – CC V alues of Zoom H4N 80Hz vs Sample Database LCF -On Data Set Correlation to ZOOM H4N LCF_ON_80Hz_A_MONO.wav Standard Deviation: .0818 ZOOM H4N LCF_ON_98Hz_MONO.wav 0.9914 ZOOM H4N LCF_ON_80Hz_B_MONO.wav 0.9895 TASCAM DR 07 LCF_ON_80Hz_B_MONO.wav 0.9828 TASCAM DR 07 LCF_ON_80Hz_A_MONO.wav 0.9812 TASCAM DR 07 LCF_ON_40Hz_A_MONO.wav 0.9811 TASCAM DR 05 LCF_ON_80Hz_MONO 0.9692 ZOOM H4N LCF_ON_115Hz_MONO.wav 0.9690 TASCAM DR 05 LCF_ON_120Hz_MONO 0.9496 TASCAM DR 05 LCF_ON_40Hz_MONO.wav 0.9482 ZOOM H1 LCF_ON_MONO.wav 0.9450 ZOOM H4N LCF_ON_150Hz_MONO.wav 0.9322 SONY ICD UX560 LCF_ON_C_MONO.wav 0.9317 OLYMPUS VP 10 LCF_ON_C_MONO.wav 0.9278 OLYMPUS VN 8100PC LCF_ON_MONO.wav 0.9211 OLYMPUS VP10 LCF_ON_B_MONO.wav 0.9135 PHILLIPS DVT 5500 LCF_ON_MONO.wav 0.9035 OLYMPUS VP10 LCF_ON_A_MONO.wav 0.9002 SONY ICD UX560 LCF_ON_B_MONO.wav 0.8889 ZOOM H4N LCF_ON_133Hz_MONO.wav 0.8880 SONY ICD UX560 LCF_ON_A_MONO.wav 0.8829 SONY ICD UX523 LCF_ON_A_MONO.wav 0.8650 TASCAM DR 07 LCF_ON_120Hz_A_MONO.wav 0.8649 OLYMPUS DM 520 LCF_ON_MONO.wav 0.8641 SONY ICD PX370 LCF_ON_B_MONO.wav 0.8595 OLYMPUS WS 852 LCF_ON_B_MONO.wav 0.8585 SONY ICD UX523 LCF_ON_B_MONO.wav 0.8487 SONY ICD UX523 LCF_ON_C_MONO.wav 0.8469 ZOOM H4N LCF_ON_168Hz_MONO.wav 0.8360 OLYMPUS WS 802 DUP LCF_ON_MONO.wav 0.8339 OLYMPUS WS 600S LCF_ON_MONO.wav 0.8328 OLYMPUS WS 700M DUP LCF_ON_MONO.wav 0.8297 ZOOM H4N LCF_ON_185Hz_MONO.wav 0.8112 SONY ICD PX312 LCF_ON_MONO.wav 0.8047 SONY ICD UX512 DUP LCF_ON_MONO.wav 0.8023 SONY ICD TX50 LCF_ON_MONO.wav 0.7971 SONY ICD UX512 LCF_ON_C_MONO.wav 0.7936 SONY ICD UX533 LCF_ON_MONO.wav 0.7915 OLYMPUS WS 802 LCF_ON_C_MONO.wav 0.7910

PAGE 35

22 Table 6 Continued LCF -On Data Set Correlation to ZOOM H4N LCF_ON_80Hz_A_MONO.wav OLYMPUS WS 700M LCF_ON_MONO.wav 0.7890 OLYMPUS WS 802 LCF_ON_B_MONO.wav 0.7882 OLYMPUS DM 901 LCF_ON_MONO.wav 0.7880 OLYMPUS DM 620 LCF_ON_MONO.wav 0.7880 OLYMPUS WS 802 LCF_ON_A_MONO.wav 0.7861 SONY ICD UX512 LCF_ON_B_MONO.wav 0.7777 OLYMPUS DS 40 LCF_ON_MONO.wav 0.7761 SONY ICD UX512 LCF_ON_A_MONO.wav 0.7759 SONY ICD PX370 LCF_ON_A_MONO.wav 0.7718 ZOOM H4N LCF_ON_203Hz_MONO.wav 0.7560 ZOOM H4N LCF_ON_220Hz_MONO.wav 0.7193 ZOOM H4N LCF_ON_237Hz_MONO.wav 0.7043 OLYMPUS WS 852 LCF_ON_A_MONO.wav 0.6500

PAGE 36

23 Table 7 – CC V alues of Olympus WS -700M vs Sample Database LCF -On Data Set Correlation to OLYMPUS -WS 700M -LCF_ON_MONO.wav Standard Deviation: .0710 SONY ICD TX50 LCF_ON_MONO.wav 0.9976 OLYMPUS WS 802 LCF_ON_C_MONO.wav 0.9970 OLYMPUS WS 802 LCF_ON_A_MONO.wav 0.9965 OLYMPUS WS 802 LCF_ON_B_MONO.wav 0.9960 OLYMPUS WS 700M DUP LCF_ON_MONO.wav 0.9927 OLYMPUS WS 600S LCF_ON_MONO.wav 0.9910 SONY ICD UX533 LCF_ON_MONO.wav 0.9905 SONY ICD UX512 LCF_ON_C_MONO.wav 0.9883 SONY ICD UX512 LCF_ON_B_MONO.wav 0.9865 SONY ICD PX312 LCF_ON_MONO.wav 0.9860 SONY ICD UX512 DUP LCF_ON_MONO.wav 0.9849 SONY ICD UX512 LCF_ON_A_MONO.wav 0.9845 OLYMPUS DM 520 LCF_ON_MONO.wav 0.9844 OLYMPUS WS 802 DUP LCF_ON_MONO.wav 0.9843 SONY ICD PX370 LCF_ON_B_MONO.wav 0.9835 SONY ICD UX523 LCF_ON_C_MONO.wav 0.9778 SONY ICD UX523 LCF_ON_B_MONO.wav 0.9759 OLYMPUS DM 620 LCF_ON_MONO.wav 0.9752 SONY ICD UX523 LCF_ON_A_MONO.wav 0.9724 OLYMPUS DS 40 LCF_ON_MONO.wav 0.9644 OLYMPUS DM 901 LCF_ON_MONO.wav 0.9614 TASCAM DR 07 LCF_ON_120Hz_A_MONO.wav 0.9593 OLYMPUS VP 10 LCF_ON_C_MONO.wav 0.9403 SONY ICD UX560 LCF_ON_C_MONO.wav 0.9397 OLYMPUS VP10 LCF_ON_A_MONO.wav 0.9356 OLYMPUS VP10 LCF_ON_B_MONO.wav 0.9356 SONY ICD UX560 LCF_ON_B_MONO.wav 0.9315 ZOOM H4N LCF_ON_168Hz_MONO.wav 0.9291 ZOOM H4N LCF_ON_133Hz_MONO.wav 0.9273 ZOOM H4N LCF_ON_185Hz_MONO.wav 0.9254 OLYMPUS WS 852 LCF_ON_B_MONO.wav 0.9239 ZOOM H4N LCF_ON_150Hz_MONO.wav 0.9205 ZOOM H4N LCF_ON_203Hz_MONO.wav 0.9162 ZOOM H1 LCF_ON_MONO.wav 0.9127 ZOOM H4N LCF_ON_220Hz_MONO.wav 0.9120 SONY ICD UX560 LCF_ON_A_MONO.wav 0.9106 ZOOM H4N LCF_ON_237Hz_MONO.wav 0.9065 OLYMPUS VN 8100PC LCF_ON_MONO.wav 0.8961

PAGE 37

24 Table 7 Continued LCF -On Data Set Correlation to OLYMPUS -WS 700M -LCF_ON_MONO.wav SONY ICD PX370 LCF_ON_A_MONO.wav 0.8948 ZOOM H4N LCF_ON_115Hz_MONO.wav 0.8865 TASCAM DR 05 LCF_ON_80Hz_MONO 0.8836 OLYMPUS WS 852 LCF_ON_A_MONO.wav 0.8796 TASCAM DR 05 LCF_ON_120Hz_MONO 0.8769 TASCAM DR 07 LCF_ON_80Hz_B_MONO.wav 0.8500 TASCAM DR 05 LCF_ON_40Hz_MONO.wav 0.8481 ZOOM H4N LCF_ON_80Hz_B_MONO.wav 0.8244 TASCAM DR 07 LCF_ON_80Hz_A_MONO.wav 0.8185 ZOOM H4N LCF_ON_80Hz_A_MONO.wav 0.7890 ZOOM H4N LCF_ON_98Hz_MONO.wav 0.7771 TASCAM DR 07 LCF_ON_40Hz_A_MONO.wav 0.7335 PHILLIPS DVT 5500 LCF_ON_MONO.wav 0.6934

PAGE 38

25 V. RESULTS The aim of this study was to determine whether sufficient salient data could be derived from the LCF effects in audio recordings made with handheld digital recorders using LTASS analysis to reliably make comparisons, exclusions or possible identifications. Three hypotheses were examined using the data created for this examination. Hypothesis One It is possible to identify whether or not the LCF option of a recorder was on or off. This hypothesis has been confirmed to be t rue. Samples were made with the LCF both on a nd off for each recorder used in the study. When processed with the initial analysis script in MATLAB, the power values of the LTASS were exported and saved to a text document. The on and off power values were then plotted to a periodogram. These figure s show the inter variability of a recorder ’ s output, LCF on vs off. As can be visually observed in figure s 6 – 39, the signal with the LCF turned on differs greatly from the signal with the LCF turned off, with varying degrees of significant differentiatio n. Figure 21, showing the spectrums of a Sony ICD TX50 recorder, is an excellent example to show a clear distinction of LCF on vs . off. Other examples are more subtle, such as figure 10 , showing the spectrums of an Olympus VP 10. All devices te s ted from manufacturers Olympus, Sony, Phillips and Zoom exhibited expected results. An interesting characteristic of the Zoom H4N model , the device with the highest number of frequency roll off options, did not seem to exhibit any substantial difference in freque ncy among the samples, only changes in the slope severity (figure s 30 39) . Below is a figure showing the overlaid sorted power values of all the LCF options; 80Hz, 98Hz, 115Hz, 133Hz, 150Hz, 168Hz, 185Hz, 203Hz, 220Hz, and 237Hz.

PAGE 39

26 Figure 5 – ZOOM H4N Sorted Power Values of all LCF Options Additionally, the Tascam DR 05 model showed less low frequency activity in the LCF off sample than in the samples taken with the LCF on and set to the three frequency options of 40Hz, 80Hz, and 120Hz, seen in figures 23 25. Additional samples or even additional DR 05 devices in a future study could help understand this anomaly. However, visually, there was a clear frequency slope drop seen in figure 2 5 which shows the sorted spectrum with the LCF set to 120Hz. Hypothesis Two Lowcut filter signal processing differs significantly between recorders. This hypothesis has been confirmed to be true. To compare the inter variability of the LTASS values of samples taken from two different recording devices, sorted power values from two different recorders, both with the LCF on, were overlaid in a single periodogram. As shown in figures 40 – 52, in most cases, there is a clear distinction between the LCF processing of one

PAGE 40

27 r ecorder to the other. It is particularly evident in figure 4 7, which shows the spectrums for the Sony ICD PX370 and the Sony ICD TX50. A more subtle example of LCF slope distinction can be seen in figure 41 where the Olympus DM 520 and Olympus DM 620 are compared. Some inter variability comparisons displayed interesting results. In figure 4 6, the Sony ICDUX512 spectrum is very similar to that of the Sony ICD UX523. A similar event is seen in figure 50 between the spectrums of the Olympus WS 600S and the Olympus WS 700M. A possible explanation is that the manufacturers copied much of the processing architecture to a subsequent model, giving them similar LCF functionality. Hypothesis Three LTASS/LCF analysis can be used as a means of eliminating a device from a pool of recorders. This hypothesis has been confirmed to be true. To test whether LTASS analysis can reliably identify the LCF signature of a purported recording device, correlation coefficients were calculated for a single device sample against all of the sample library used in this study. CC values were charted for six comparative studies shown in tables 2 – 7. The highest CC value was highlighted to show which recorder sample had the highest mathematical relationship to the exemplar. The results are as follows:

PAGE 41

28 Table 8 – CC Test Results Exemplar Recorder Recorder with Highest CC Sample Test 1 OLYMPUS_VP10 LCF_ON_A_MONO.wav SONY ICD UX560 LCF_ON_B_MONO.wav Test 2 OLYMPUS WS 802 LCF_ON_A_MONO.wav OLYMPUS WS 802 LCF_ON_B_MONO.wav Test 3 SONY ICD PX370 LCF_ON_A_MONO.wav OLYMPUS WS 852 LCF_ON_A_MONO.wav Test 4 SONY ICD UX523 LCF_ON_A_MONO.wav SONY ICD UX523 LCF_ON_B_MONO.wav Test 5 ZOOM H4N LCF_ON_80Hz_A_MONO.wav ZOOM H4N LCF_ON_98Hz_MONO.wav Test 6 OLYMPUS WS 700M LCF_ON_MONO.wav SONY-ICD TX50 -LCF_ON_MONO.wav Tests 2, 4 and 5 were accurately able to determine which recording device was used to create the exemplar recording because another audio sample made with the same recorder measured the highest CC values. However, tests 1, 3 and 6 resulted in a make and model of recorder other than the exemplar as having the highest CC value. While LTASS analysis cannot reli ably identify the LCF signature of specific handheld digital recorders, it may still be used to make an exclusion through spectrum comparisons on a periodogram. The scope of this study only included recorders with the LCF option, but there are numerous re corders on the market that do not have this option. Should a signal after analysis show evidence of an LCF , but the purported recorder does not have this option, this information could be useful to an examiner.

PAGE 42

29 VI. DISCUSSION & FUTURE RESEARCH This study was able to show that LTASS analysis can be a very useful technique when analyzing the effects of an LCF on an audio signal from a handheld digital recorder. Its use does have limitations , but when used in conjunction with other recommended techniqu es can provide valuable data for an investigation. Possible areas of interest for future research could be: Samples may be created in a calibrated environment involving pink noise or frequency sweeps. Samples may be created in m ore varied “real world” situ ations to test how background noise effects the reliability of LTASS analysis. Samples may be created at the extremes of available resolution in a recorder to further test intra variability.

PAGE 43

30 R EFERENCES 1. Maher, R.C. and S.R. Shaw, Gunshot Recordings from Digital Voice Recorders, in 54th International Conference: Audio Forensics . 2014: London, England. 2. Pruett, S., An Exploratory Study of Manual Pause Record Events in Digital Audio Recordings . 2019, Unive rsity of Colorado Denver. 3. SWGDE, Best Practices for Forensic Audio, Version 2.2. October, 2016. 4. Rappaport, D.L., Establishing a standard for digital audio authenticity : a critical analysis of tools, methodologies, and challenges . 2012, University of Colorado Denver. 5. Koenig, B.E., D.S. Lacey, and S.A. Killion, Forensic Enhancement of Digital Audio Recordings. Journal of the Audio Engineering Society, 2007. 55 (5): p. 352 370. 6. Grigoras, C., D.L. Rappaport, and J.M. Smith, Analytical F ramework for Digital Audio Authentication, in AES 46th International Conference: Audio Forensics . 2012: Denver, CO, USA. 7. Grigoras, C., Statistical Tools for Multimedia Forensics , in 39th International Conference: Audio Forensics: Practices and Challenge s . 2010: Hillerd, Denmark. 8. Grigoras, C. and J.M. Smith, Audio Enhancement and Authentication, in Encyclopedia of Forensic Sciences. 2013. p. 315326.

PAGE 44

31 APPENDIX Figure 6 – Olympus DM -520 LCF On/Off Comparison Figure 7 – Olympus DM -6 20 LCF On/Off Comparison

PAGE 45

32 Figure 8 – Olympus DM -901 LCF On/Off Comparison Figure 9 – Olympus DS -40 LCF On/Off Comparison

PAGE 46

33 Figure 10 – Olympus VN 8100PC LCF On/Off Comparison Figure 11 – Olympus VP10 LCF On/Off Comparison

PAGE 47

34 Figure 12 – Olympus WS 600S LCF On/Off Comparison Figure 13 – Olympus WS 700M LCF On/Off Comparison

PAGE 48

35 Figure 14 – Olympus WS 852 LCF On/Off Comparison Figure 15 – Phillips DVT-5500 LCF On/Off Comparison

PAGE 49

36 Figure 16 – Sony ICD -PX312 LCF On/Off Comparison Figure 17 – Sony ICD -PX370 LCF On/Off Comparison

PAGE 50

37 Figure 18 – Sony ICD -UX512 LCF On/Off Comparison Figure 19 – Sony ICD -UX523 LCF On/Off Comparison

PAGE 51

38 Figure 20 – Sony ICD -UX533 LCF On/Off Comparison Figure 21 – Sony ICD -UX560 LCF On/Off Comparison

PAGE 52

39 Figure 22 – Sony ICD -TX50 LCF On/Off Comparison Figure 23 – Tascam DR -05 40Hz LCF On/Off Comparison

PAGE 53

40 Figure 24 – Tascam DR -05 8 0Hz LCF On/Off Comparison Figure 25 – Tascam DR -05 120Hz LCF On/Off Comparison

PAGE 54

41 Figure 26 – Tascam DR -0 7 40Hz LCF On/Off Comparison Figure 27 – Tascam DR -07 8 0Hz LCF On/Off Comparison

PAGE 55

42 Figure 28 – Tascam DR -07 120Hz LCF On/Off Comparison Figure 29 – Zoom H1 LCF On/Off Comparison

PAGE 56

43 Figure 30 – Zoom H4N 8 0Hz LCF On/Off Comparison Figure 31 – Zoom H4N 98 Hz LCF On/Off Comparison

PAGE 57

44 Figure 32 – Zoom H4N 115 Hz LCF On/Off Comparison Figure 33 – Zoom H4N 133 Hz LCF On/Off Comparison

PAGE 58

45 Figure 34 – Zoom H4N 150 Hz LCF On/Off Comparison Figure 35 – Zoom H4N 168 Hz LCF On/Off Comparison

PAGE 59

46 Figure 36 – Zoom H4N 185 Hz LCF On/Off Comparison Figure 37 – Zoom H4N 203 Hz LCF On/Off Comparison

PAGE 60

47 Figure 38 – Zoom H4N 220 Hz LCF On/Off Comparison Figure 39 – Zoom H4N 237 Hz LCF On/Off Comparison

PAGE 61

48 Figure 40 – Olympus DM -901 vs Olympus DS -40 LCF On Figure 41 – Olympus DM -520 vs Olympus DM -620 LCF On

PAGE 62

49 Figure 42 – Zoom H1 vs Zoom H4N 237 LCF On Figure 43 – Tascam DR -05 120Hz vs Tascam DR07 120Hz LCF On

PAGE 63

50 Figure 44 – Sony ICD -UX560 vs Tascam DR -05 120Hz LCF On Figure 45 – Sony ICD -UX523 vs Sony ICD -UX533 LCF On

PAGE 64

51 Figure 46 – Sony ICD -UX512 vs Sony ICD -UX523 LCF On Figure 47 – Sony ICD -PX370 vs Sony ICD -TX50 LCF On

PAGE 65

52 Figure 48 – Phillips DVT-5500 vs Sony ICD -PX312 LCF On Figure 49 – Olympus WS 802 vs Olympus WS -852 LCF On

PAGE 66

53 Figure 50 – Olympus WS 600S vs Olympus WS -700M LCF On Figure 51 – Olympus VN 8100PC v s Olympus VP-10 LCF On

PAGE 67

54 Figure 52 – Olympus DM -520 vs Sony ICD -PX370 LCF On