Citation
Analysis of voice activated artifacts

Material Information

Title:
Analysis of voice activated artifacts
Creator:
Black, Kelsey ( author )
Place of Publication:
Denver, Colo.
Publisher:
University of Colorado Denver
Publication Date:
Language:
English
Physical Description:
1 electronic file (74 pages) : ;

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

Subjects

Subjects / Keywords:
Sound -- Recording and reproducing ( lcsh )
Voice ( lcsh )
Automatic speech recognition ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
This purpose of this thesis is to analyze voice-activated recording artifacts, using a playback audio created in Adobe Audition. To show how an automated voice recorder with standby mode treats the silence of a recording. This thesis focuses on the WAV PCM format. The WS-550M, WS-560M, and the DM-520 recorders did not have the option to create a WAV PCM file, therefore the WS-550M and the 560M created MP3 files and the DM-520 created a WMA file. Each of the recorders have automated standby mode. The recorders were set to create a WAV PCM that was a 16-bit stereo file at 44kHz. Below is a list of the devices that will be used in this study. Olympus DM-520 ; Olympus DM-620 ; Olympus WS-550M ; Olympus WS-560M ; Olympus WS-700M ; Olympus WS-700M ; Olympus WS-750M ; Olympus WS-760M ; Olympus WS-802 ; Olympus WS-822 ; Olympus WS-823 Philips Voice Tracer.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: Adobe Reader.
Statement of Responsibility:
by Kelsey Black.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
on10208 ( NOTIS )
1020853560 ( OCLC )
on1020853560
Classification:
LD1193.A70 2017m B53 ( lcc )

Downloads

This item has the following downloads:


Full Text
ANALYSIS OF VOICE ACTIVATED ARTIFACTS
by
KELSEY BLACK B.S., Marshall University, 2015 B.S., Marshall University, 2015
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
2017


This thesis for the Master of Science degree by
Kelsey Black
has been approved for the Recording Arts Program by
Catalin Grigoras, Chair Jeff M. Smith Scott Burgess
Date: December 16, 2017


Black, Kelsey (M.S., Recording Arts Program)
Analysis of Voice Activated Artifacts
Thesis directed by Assistant Professor Catalin Grigoras
ABSTRACT
This purpose of this thesis is to analyze voice-activated recording artifacts, using a playback audio created in Adobe Audition. To show how an automated voice recorder with standby mode treats the silence of a recording. This thesis focuses on the WAV PCM format. The WS-550M, WS-560M, and the DM-520 recorders did not have the option to create a WAV PCM file, therefore the WS-550M and the 560M created MP3 files and the DM-520 created a WMA file. Each of the recorders have automated standby mode. The recorders were set to create a WAV PCM that was a 16-bit stereo file at 44kHz. Below is a list of the devices that will be used in this study.
Olympus DM-520 Olympus DM-620 Olympus WS-550M Olympus WS-560M Olympus WS-700M Olympus WS-700M Olympus WS-750M Olympus WS-760M Olympus WS-802 Olympus WS-822 Olympus WS-823 Philips Voice Tracer
The form and content of this abstract are approved. I recommend its publication.
Approved: Catalin Grigoras


DEDICATION
For my sweet daughter, Jaslynn, without you I wouldnt be where or who I am today. You make me want to become a better person, and strive to be the best that I can be for you. Becoming a parent in graduate school, I gave up more time with you than I would have liked to have given up. Everything I do, I do for you, to ensure that you can have the best of everything you could ever want. Im extremely proud to be your mommy, I love you baby girl.
I would like to thank my parents and the rest of my family for always being supportive and understanding. Thank you for helping me get through the last seven years of college and graduate school. Thank you for standing by me during those times of extreme stress, swift attitude changes, and everything in between. Most importantly thank you for helping with Jaslynn while I continue my education, I love you.
I would like to thank Wanda Dyke, Brian Morgan, and Josh Brunty for all your help not only throughout my time at Marshall University but for all the help you have given me now. Each of you have had a huge role in helping me get to where I am today, whether it is writing recommendation letters to begin graduate school and my internship, checking up on me, or simply supporting me. The guidance you gave me showed me just who I want to be, and what I am capable of accomplishing. I am truly grateful for each of you, and I honestly cant thank you enough for your support and help.
IV


ACKNOWLEDGEMENTS
I would like to thank my professors Jeff Smith and Catalin Grigoras. The last two years have been challenging, eventful, but most importantly extremely educational. When I first started in the Recording Arts emphasis Media Forensics I was beyond certain that I was way out of my league. That first year you were supportive as I became a parent, you were understanding when I reached out to you. I have learned so much in the last two years, so much of that is already being utilized. I took an incredible opportunity for an honors internship during a very crucial class. During that time, you were supportive, understanding, and helped me to find a way to participate in the class.
I would like to thank Leah and Emma for answering my random questions, helping me get into the classroom, and the helpful reminders. Without the help I received from you I would be lost and behind in the program.
Thank you, Scott Burgess, even though we have not met you agreed to be on my committee. I greatly appreciate you taking the time to help me complete my masters degree education.
v


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION........................................ 1
Previous research..............................2
II. PREPARATIONS.........................................7
Methods........................................7
Materials......................................7
Testing.......................................11
III. ANALYSIS...............................................13
IV. RESULTS................................................20
V. FUTURE RESEARCH........................................21
BIBLIOGRAPHY................................................23
APPENDIX A-OLYMPUS DM-520...................................24
APPENDIX B OLYMPUS DM-620.................................27
APPENDIX C OLYMPUS WS-550M................................30
APPENDIX D OLYMPUS WS-560M................................33
VI


APPENDIX E OLYMPUS WS-700M
36
APPENDIX F OLYMPUS WS-700M.............................39
APPENDIX G OLYMPUS WS-750M.............................42
APPENDIX H OLYMPUS WS-760M.............................45
APPENDIX I OLYMPUS WS-802..............................48
APPENDIX J OLYMPUS WS-822..............................51
APPENDIX K OLYMPUS WS-823..............................54
APPENDIX L PHILIPS VOICE TRACER........................57
APPENDIX M EXCEL SPREADSHEETS..........................60
vii


LIST OF FIGURES
FIGURE
1. Diagram................................................................8
2. WAV PCM file...........................................................9
3. Generate 1khz tones settings..........................................10
4. Generate silence settings.............................................11
5. Olympus DM-520 recording..............................................13
6. Olympus DM-520 recording break 1....................................14
7. Olympus DM-520 recording samples....................................14
8. Olympus DM-520 recording amplitude statistics.......................15
9. Olympus DM-520 recording frequency analysis.........................15
10. Excel spreadsheet samples all breaks..............................17
11. Excel spreadsheet samples mean, standard deviation................18
12. Olympus DM-520 MATLAB waveform and energy plot....................19
20
viii
13. Excel spreadsheet samples time duration


LIST OF ABBREVIATIONS
1. WMA Windows Media Audio
2. WAV PCM Waveform Audio File Format, Pulse-Code Modulation
3. MP3 Motion Picture Experts Group Layer-3


CHAPTER I
INTRODUCTION
This purpose of this thesis is to analyze voice-activated recording artifacts, using a playback audio created in Adobe Audition. To show how an automated voice recorder with standby mode treats the silence of a recording. This thesis focuses on the WAV PCM format. The goal is to show that each recorder with the automatic standby mode records the silence different, whether it be different from recorder to recorder or being different in the length of time being captured from the playback audio. Looking at the different sets of data that was captured from each recorder will show how the WAV PCM format captures the specific playback audio from model to model on the recorders.
As technology is advancing, the need for better understanding on how the automatic standby mode affects a recording is crucial for the Digital Forensics field. Looking at the WAV PCM format is just one area that needs better understanding. This thesis will look at one WMA file, nine WAV PCM files, and two MP3 files, however the focus will be on the WAV PCM files. Right now, what is known is if you have a ten-second generated silence, the automated standby mode of the recorder is entering standby mode as the silence begins. This research will show what happens when the recorders enter standby mode and how it affects the captured playback audio. In the Digital Forensics field recorders are becoming more commonly used to capture audio, so this thesis will help to understand what the recorder does when a silence occurs, and how that affects the outcome of the recording.
1


Previous Research
The topic of automated standby mode, which is also considered to be voice-activated, is a topic that does not have extensive prior research. There are instances in previous works that a study on it has been suggested. Instances of pause in records were mentioned in an article titled Tape Analysis and Authentication Using Multi-Track Recorders (Begault, Brustad, & Stanley, 2005). In their article they said
In many cases, an audio forensic expert is called upon to examine taped evidence to provide an opinion on whether or not a tape has been edited or doctored in any way. Specifically, this translates into an analysis of the temporal sequence of events found on the tape that correspond to record start, pause, and stop operations of one or more tape recording devices.
After reading that, it was evident that a study about the voice-activated artifacts needed to be performed. With the way the technology is changing, and the criminals are evolving with the technology, a study regarding the automated standby mode is becoming increasingly more required. The forensic field needs that understanding of how a recorder handles the automated standby mode. This study goes on to discuss the waveform analysis, which this study shows the waveform of each of the recordings that were taken. However, this study goes on to say that the type of analysis needed for the pause signatures of a recording are not within the scope of that article.
2


The next article that was reviewed was the Test Audio Recordings and Their Use in Authenticity Examinations. Database of Properties of Digital Audio Recorders and Recordings. (Michalek, 2016). This article discusses the digital audio recorders that are becoming available, and how they have so many capabilities among which comes a voice-activated recording. It explains that the evidentiary value of being able to verify the model of a recorder and the parameters associated with that recorder. The author does not go into detail regarding the pause or voice-activated functions, as this article is geared more towards the authenticity of a recording more than it is to provide an understanding of the pause or voice-activated functions. The evidentiary value of knowing how a recorder reacts to an automated standby mode could also be very helpful to the forensics field, especially when it comes to recorded conversations.
Next is an article called Overview of Audio Forensics (Maher, 2010), in this article the voice-activated functionality is referred to as a gate. Maher (2010) stated:
The noise gate compares the short-time level of its input signal with a pre-determined level threshold. If the signal level is below the threshold, the gate closes and no signal is let through, while if the signal level is above the threshold, the gate opens and allows the signal to pass.
When thinking about what the voice-activation functionality is doing, it could be considered as a gate. In this instance the gate as Maher called it, would be the standby mode once a period of silence has been detected. On a digital recorder like one of the ones used in this thesis study the background noise of the recording area
3


was recognized as silence and the only noise to be recorded was the tones that were supposed to be recorded. So essentially when the recorder recognized the sounds of the air conditioning unit as a silence it had become a gate and was not allowing those sounds to activate the recorder to enter back into recording mode. But once again it does not go into further details about what the voice-activated does, after the gate comparison.
Next is Audio Forensic Examination (Maher, 2009), this article is discussing that during an examination the examiner is looking for consistency marks. They are looking for those identifying areas like the record, erase and the pause as well. Mainly the transitions from each of these modes can be seen on the magnetic development of the recording. However, this article does explain that these markers can be helpful during the examination process to identify the recorders. It does not continue or explain more about the functionality of the voice-activated or pause areas to support their evidentiary value. They explain that the examiner is using these markers to determine authenticity which having a better understanding of the voice-activated function could provide an easier identifier for the authenticity of a recording.
Next is the SWGDE Best Practices for Digital Audio Authentication (Scientific Working Group on Digital Evidence, 2017), in this article they are laying out the best methods for which an examiner should go about their authentication process. They list questions that each examiner should ask prior to their examination. They state:
4


Additionally, discontinuities and alterations within the questioned recording should be included in the report with an explanation of the cause. Examples would include recording pauses, voice activation discontinuities, etc.
They are saying that should the examiner find an instance of a pause or other discontinuities they should put it in their report, that they discovered it. Then after they have completed their examination they can say with a reasonable degree of consistency whether the device was paused like the client stated it was or was not.
In Appendix A: Sample Case (Scientific Working Group on Digital Evidence, 2017) they show a case in which they have two versions of what happened to a recording. One instance it was continuous and the other instance it was paused and restarted, so they use the waveform, and spectrogram analysis to determine if there were any instances of pause/restart in the recording. After they finished testing with the waveform and the spectrogram they used the energy analysis to confirm the results. This article does show that you can determine if there are any instances of pause/restart in each recording. Even though this article does talk about pause/restart further research is needed to gain a better understanding.
Next is the Forensic Enhancement of Digital Audio Recordings (Koenig, Lacey, & Killion, 2007). In this article is discussing some of the techniques that can be used to enhance an audio recording. It discusses the process of examination from the equipment and software to the expert testimony. They can discuss how they are able to improve the intelligibility of a voice from a recording. They explain that when they receive an audio file that is not playable they have to include that in
5


their report along with the possible reasons as to why the audio file is not playable. But much like the other articles it does not going into detail about the voice-activated/automated standby. This article is more geared toward enhancing what they are receiving in evidence versus understanding what the voice-activated/automated standby is doing to the recordings.
The last article that was reviewed was Forensic Authentication of Digital Audio Recordings (Koenig & Lacey, 2009). This article starts by discussing the materials that are at hand for these types of examinations to take place. They go on to discuss the authenticity examination protocol in which they are determining one of four options for the recording. One of the options is:
Has discontinuities in the recording process, including record stops,
starts, stop/starts, and amplitude-activated pauses.
So, for this article the pauses or voice-activated areas are discontinuities. They continue to discuss how they create their test recordings. They go through a case in which an authentication of an audio file was conducted. However, there is a mention of the voice-activated or automated standby but its just that a mention. They do not go further into the subject. Which is why the research on this topic should be done, to show what it looks like from a recorder using a specific set of settings.
6


CHAPTER II
PREPARATIONS
The test that has been designed for this study has a set time limit. In that time limit the noise has its own time frame followed by a specific time frame of generated silence. The playback audio that has been created will be recorded on each of the chosen recorders that have the automated standby mode. The testing will use twelve recorders. Each recorder will record the playback audio that was created for this study. After each recorder has recorded the playback audio recording, the file will then be transferred from the recorder to be analyzed after the testing has completed.
Materials
The software materials that were used in this study were: Microsoft Excel, Mathworks MATLAB, and Adobe Audition. There were two different brands of recorders being used in this study, Olympus and Phillips were the only two types of recorders with automated standby modes being used. There were eleven different Olympus recorders and only one Phillips recorder. Nearly all the recorders are Olympus, this gives the opportunity to see how different models of the same brand react when using the standby mode.
Methods
Using Adobe Audition the playback audio recording was made. The playback audio had 1kHz tones, 2kHz tones, and silence areas. Each area lasted for ten-seconds, starting with the 1kHz tone, then going to a ten-second silence, then going
7


to the 2kHz tone. The playback audio was setup so that it would have two rounds of the 1kHz tones, and the 2kHz tones, and then it would contain three areas of silence. The reason this study used the 1 kHz tones and the 2kHz tones instead of a broadband noise goes back to the information that was found while reviewing the previous research. Previous research shows that some recorders will introduce a gap when using the pause-record mode. It was unknown as to whether the recorders being used in this test would introduce a gap, therefore it was needed to use the 1 kHz tones and the 2kHz tones to differentiate between them on the spectrogram.
The following figures provides a better understanding of the layout of the playback audio recording that was used. The first figure shows the diagram of how the playback audio should be generated to get the best results. The second figure shows what the playback audio file looks like after being generated.
Figure 1.: Diagram showing the layout of the playback audio recording.
8


This setup for the playback audio recording allows for each recorder to have more than one sample area of silence. It starts with a 1 kHz tone lasting for ten-seconds, then it will go to a silence for ten-seconds, then to the 2kHz tone for ten-seconds, then to another silence for ten-seconds, then back to the 1 kHz tone for another ten-seconds, then to another silence for ten-seconds, and then it will end with a 2kHz tone for ten-seconds. This test is looking at the areas of silence to see how each recorder captured that ten-seconds of silence.
When generating the 1 kHz tone, the base frequency was set to 1000Hz. The modulate and modulation frequency were set to zero. For the frequency components, the first component was set to 100, and components two through five were set to zero. In the phasing area, all settings were set to zero. In the dB Volume area, both the Left and the Right were set to -6. The DC offset was set to zero. The Flavor was set to sine. The duration was set for ten-seconds. Figure 3. shows the settings for the 1 kHz tone.
9


Figure 3.: Adobe Audition 3.0, Generating 1kHz tones.
When generating the 2kHz tone, it needed to have most of the same settings as the 1 kHz tone. The base frequency was set to 2000Hz. The flavor characteristic was set to one. The modulate and modulation frequency were set to zero. The frequency components were the same with the first component being set to 100, and components two through five were set to zero. In the phasing area, all settings were set to zero. In the dB Volume area, both the Left and the Right were set to -6. The DC offset was set to zero. The Flavor was set to sine. The duration was set for ten-seconds.
When generating the silence, it was much simpler as the settings had one option that could be changed and that was the duration for which the generated
10


silence was supposed to last. The setting was changed to show ten-seconds. Figure 4. shows the settings for generating silence.
Figure 4.: Adobe Audition 3.0, generating a ten-second silence.
Testing
After the audio file has been created, the testing phase can begin. In the testing phase, the area for testing has background noise to give the test a more realistic setting. The test was setup using two desktop computer speakers. The speakers were placed approximately three inches apart. The recorder was laying down on a desk during the recording time, this allowed for the recorder to be placed approximately one inch centered from the speakers. For the Olympus brand recorders, the microphone is located at the top of the recorders. For the Philips recorder the microphone was in the top of the recorder also. The recorders were placed between the speakers to ensure that both the left and right channel received the tones at the same time at the same volume. To begin the recording process the record button on the recorder was pushed to begin recording, allowing to enter standby mode while pushing the start button to begin playing the recording button. The playback audio file lasts for approximately one minute and twenty seconds. The computer volume is set to 100%. Once the recorder has finished recording the playback audio the record button is pushed once more to complete the recording phase for that device. After the recording has been completed, the recorded file is
11


then transferred to the computer for the analysis phase. Only one recorder is running at a time, this was done to ensure that the recorder could capture the playback audio file as it was being played through both computer speakers, without have the extra sounds of each recorder as they are being started and stopped.
During the testing time, if a noise other than the background noise that the recorder noticed during the beginning of the recording is picked up the recording was started over. The main area of concern were the areas of silence in between the 1 kHz and 2kHz tones. The areas of silence before the first 1 kHz tone was not a concern as the recorder would have picked up the sound of the recorder being started and then went into standby mode until the playback audio is started. The same goes with the ending of the recording, after the end of the 2kHz tone the recorder would pick up the sound of the recording being stopped. So, to conduct a proper test of the areas of silence, the beginning and end were not taken into consideration for this study.
12


CHAPTER III
ANALYSIS
The analysis phase was done using Adobe Audition and MATLAB. Using Audition, each recording was loaded into the software to get a better view of what was recorded. Figure 5. will show what the recording looks like after it has been loaded into Adobe Audition.
Figure 5.: Olympus DM-520 recording of the playback audio file.
Looking at Figure 5. the information that is being analyzed is the silences, those silences can easily be seen in this recording. Each of those spaces were set for a ten-second time duration which is why the entire playback audio file was one minute and twenty seconds. When looking at the silences we are calculating the number of samples per area of silence, however, the silence at the beginning of the recording will not be calculated and nor will the silence at the end of the recording. This is to ensure that the silence area that is being calculated has not been affect by the sounds of the recorder being started and stopped. When analyzing each of the
13


recordings, the number of samples, the amplitude statistics, and the frequency analysis are taken from each of the middle three silence areas.
Figure 6.: Olympus DM-520, zoomed in to an area of silence to be analyzed.
Looking at Figure 6. it shows a zoomed in version of the audio recording. It is showing a 1 kHz tone followed by a silence and then to the 2kHz tone. That area of silence is what is being analyzed, the region of that silence that will be selected for analysis will be directly following the end of the 1kHz tone and stopping immediately prior to the start of the 2kHz tone.
Figure 7.: Olympus DM-520, Number of samples.
14


Figure 8.: Olympus DM-520, Amplitude Statistics.
Figure 7. shows the number of samples per the selected area of silence. In this case looking at the Olympus DM-520 recorder, the number of samples contained in the first selected silence was 26942. The samples started at 247140 and then ended at 274082. Now that the number of samples for that silence has been determined, its time to go on to the amplitude statistics. The amplitude statistics shows the Minimum Sample Value, Maximum Sample Value, Peak Amplitude, Possibly Clipped Samples, DC Offset, Minimum RMS Power, Maximum RMS Power, Average RMS Power, and Total RMS Power, and Actual Bit Depth.
15


The main information that is being analyzed from this is the Peak Amplitude, the Minimum RMS Power, the Maximum RMS Power, the Average RMS Power, and the Total RMS Power. Figure 8. shows all the information that was obtained looking at the amplitude statistics. Figure 9. shows the frequency of the sampled area, which shows that the area that was sampled does not contain the sound of the tones.
After each of the recordings were completed, they were transferred to the computer one at a time. After transferring, the brand, the model, and the serial number of each recorder was recorded for the results phase. The format for each recording was recorded for the results, to ensure the results of the WAV PCM files were not mistaken with another format. Looking at the number of samples contained within the first silence among all twelve recorders you can see a variation.
Looking at the information in the samples area of Figure 10. the variation can be seen. A similarity can be seen between the model numbers that are close, but thats just for the first silence in each of these recordings. Using the same recording from the Olympus DM-520, the samples for the second are totaled at 25953, with the beginning starting at 494580 and ending at 520533. Comparing the second silence sampling area to the first sampling area there is less than one thousand samples difference. At this point we can see that this Olympus DM-520 recorder has a similar pattern for their number of samples per the first and second area of silence for this recording.
Now to look at the third silence for this recording. Using the same Olympus DM-520 recording, the third silence number of samples is very close to the number of samples contained in the first silence. The number of samples between the first
16


and the third are only five samples difference. For the third silence the total number of samples was 26947, the number of samples for this area began at 741000 and then ended at 767947. While the beginning and end numbers for the samples are not very close however the total number of samples among all three silences is similar being that are less than one thousand samples in difference.
Looking at the samples for each break, the total number of samples per break was used to calculate the mean and the standard deviation of each recording. The mean and the standard deviation can be seen in the excel spreadsheet shown in Figure 11.
Samples
Recorder Format Break #1 Break #2 Break #3
Brand Model WAV PCM/ MP3/WMA Begin End Length Begin End Length Begin End Length
Olympus DM-520 WMA 247140 274082 26942 494580 520533 25953 741000 767947 26947
Olympus DM-620 WAV PCM 460274 495650 35376 936652 974200 37548 1415200 1450579 35379
Olympus WS-550M MP3 460000 470000 10000 914080 929416 15336 1373000 1383000 10000
Olympus WS-560M MP3 460000 470000 10000 914061 929354 15293 1373000 1383000 10000
Olympus WS-700M WAV PCM 442200 493955 51755 935072 987130 52058 1427881 1481726 53845
Olympus WS-700M WAV PCM 442579 496432 53853 937467 989361 51894 1430300 1484200 53900
Olympus WS-750M WAV PCM 450000 485000 35000 935113 988918 53805 1440000 1475000 35000
Olympus WS-760M WAV PCM 450000 485000 35000 934812 988650 53838 1440000 1475000 35000
Olympus WS-802 WAV PCM 470000 510000 40000 958100 1011825 53725 1460000 1500000 40000
Olympus WS-822 WAV PCM 470000 515000 45000 961200 1017061 55861 1465000 1505000 40000
Olympus WS-823 WAV PCM 475000 515000 40000 965000 1020886 55886 1466000 1512000 46000
Philips Voice Tracer WAV PCM 465549 597150 131601 1039143 1170716 131573 1612650 1744191 131541
FigurelO.: Excel Spreadsheet showing the samples for each break.
17


Number of Samples per Break
Format Recorder Make/Model Serial Number Break 1 Break 2 Break 3 Mean Standard Deviation
Olympus DM-620 100115567 35376 37548 35379 36101 1253.14
Olympus WS-700M 100124078 51755 52058 53845 52552.67 1129.40
Olympus WS-700M 100126397 53853 51894 53900 53215.67 1144.84
Olympus WS-750M 200104369 35000 53805 35000 41268.33 10857.07
WAV PCM Olympus WS-760M 200110592 35000 53838 35000 41279.33 10876.12
Olympus WS-802 100137893 40000 53725 40000 44575 7924.13
Olympus WS-822 100169832 45000 55861 40000 46953.67 8108.97
Olympus WS-823 100258938 40000 55886 46000 47295.33 8021.82
Philips Voice Tracer LFH0882 131601 131573 131541 131571.67 30.02

MP3 Olympus WS-550M 200137081 10000 15336 10000 11778.67 3080.74
Olympus WS-560M 200126197 10000 15293 10000 11764.33 3055.91

WMA Olympus DM-520 100104915 26942 25953 26947 26614 572.45
Figure 11:. Excel Spreadsheet showing the mean and standard deviation from total samples of each break.
After noticing the similarities amongst the number of samples just from the Olympus DM-520 a more in depth look at each of the recordings was needed. Not all the recorders had the same closeness in the number of samples for each silence as the Olympus DM-520 does. Looking at the frequency analysis for each of the recordings, it was clear that during the areas of silence there was just the background noise from the room in which the testing was performed. The amplitude statistics of each recording gave us the RMS power information for the left and the right of each recording.
After the information had been put into an excel spreadsheet, MATLAB was then used to run a script to plot the waveform and the energy plots for each of the twelve recordings. Looking at the plots, plot 1 and plot 3 show the same output signal, the amplitude looks alike, however the energy in the right channel looks different.
18


DM520997.WMA, Left Channel
5 6
samples
DM520997.WMA, Left Channel
xicr
O) 0.8 dj 0.6
c 0.4 LLI Q 2
4 5 6 7
samples
DM520997.WMA, Right Channel
samples
DM520997.WMA, Right Channel
10 x 105
x 10
a! 0 .3 CD 0.6 c 0.4 LU 0 2
5 6
samples
10 x 105
Figure 12.: Olympus DM-520 MATLAB waveform and energy plot for both the left and the right.
During the testing of these recorders, the number of samples per area of silence varied. That is due to a few of the recordings having a zero value at the end of the tones. So, to not have a sample number of zero, the sample size for those files were selected as a smaller area than the rest of the audio files had. This allowed for the number of samples per that area of silence to be something other than zero.
19


RESULTS
This thesis started with a playback audio file that was generated using Adobe Audition. That audio file was set to have a start of ten-seconds of 1 kHz tone, followed by ten-seconds of silence, then ten-seconds of 2kHz tones, then ten-seconds of silence again, then back to ten-seconds of 1 kHz tones, then another ten-seconds of silence, and ending with ten-seconds of 2kHz tones. The entire audio file was one minute and twenty seconds long.
Now lets look at the duration of the twelve recorded audio files, the playback audio file was one minute and twenty seconds long. Figure 13. shows the duration of the recorded file from each recorder.
Recorded File Duration
Brand Model Serial Number Recording Length
Olympus DM-520 100104915 46 seconds
Olympus DM-620 100115567 43 seconds
Olympus WS-550M 200137081 41 seconds
Olympus WS-560M 200126197 41 seconds
Olympus WS-700M 100124078 44 seconds
Olympus WS-700M 100126397 46 seconds
Olympus WS-750M 200104369 44 seconds
Olympus WS-760M 200110592 44 seconds
Olympus WS-802 100137893 45 seconds
Olympus WS-822 100169832 45 seconds
Olympus WS-823 100258938 45 seconds
Phillips Voice Tracer LFH0882 52 seconds
Figure 13.: Excel Spreadsheet showing the time duration for the recorded file on each recorder.
Just by looking at the duration of each of the recordings you can see that
even though the playback audio file had a set time frame of ten-seconds per break of silence, the recorder did not capture that ten-seconds once the recorder entered standby mode.
20


CHAPTER IV
FUTURE RESEARCH
In this study, the areas of silence to show that the recorder had gone into standby mode showed that the even though the playback audio file had a set ten-seconds of silence the recorder did not record that. The playback audio file was one minute and twenty seconds, after the testing and analysis was completed none of the recorders had a one minute and twenty second recordings. The recordings were less than one minute in duration. Once the recorder has gone into standby mode and essentially paused the recording process the playback audio file was still going which is how the recordings were shorter than the playback audio file. Therefore, the results could show that after the recorder has adapted the background noise of a recording the recorder will remain in standby mode until a different noise level is present before the recording will continue and the recording time itself will remain at the time in which the recorder entered standby mode.
This study was just the beginning of the information that is needed to be explored regarding the voice-activated or automated standby modes. The focus of this study was on the automated standby mode recorders, and WAV PCM files, there was one WMA and two MP3 files that were tested. However, when it comes to the WMA and MP3 files further research is requested, along with the manual standby mode recorders. Testing to determine if there is a possibility that the recorder can show a difference of amplitude among the channels is needed. In this study the 1 kHz tones do not have a spike but there is a spike present on the 2kHz tones further research is needed to gain a better understanding of this issue. Testing
21


using broadband noise, real life noises, as well as speech and other noises is requested. There is a need for further research using a different set of computer speakers to determine if the spikes are coming from the speakers or if they are coming from the recorders. This study is just the beginning of the information that needed to understand the topic of voice-activated artifacts.


BIBLIOGRAPHY
Begault, D. R., Brustad, B. M., & Stanley, A. M. (2005). Tape Analysis and Authentication Using Multi-Track Recorders. 1-7.
Koenig, B. E., & Lacey, D. S. (2009). Forensic Authentication of Digital Audio Recordings. 662-695.
Koenig, B. E., Lacey, D. S., & Killion, S. A. (2007). Forensic Enhancement of Digital Audio Recordings. 352-371.
Maher, R. C. (2009). Audio Forensic Examination. 84-94.
Maher, R. C. (2010). Overview of Audio Forensics. 127-144.
Michalek, M. (2016). Test Audio Recordings and Their Use In Authenticity Examinations. Database of Properties of Digital Audio Recorders and Recordings. Problems of Forensic Sciences, 355-369.
Scientific Working Group on Digital Evidence. (2017). SWGDE Best Practices for Digital Audio Authentication. 1-27.
23


APPENDIX A
OLYMPUS DM-520
Images of the Olympus DM-520 recording, settings, and MATLAB energy
This shows the frequency statistics of the first break.
24


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
0.5 0 -0.5 -1
0 1 23 4 56709 10
DM520997.WMA, Left Channel
J________________!_______________J________________I_______________!_______________!_______________I_______________1________________I_______________L
samples
DM520997.WMA, Left Channel
xIO"3
1 -0.3 -0.6 -0.4 -0.2 -
4 5 6 7
samples
DM520997.WMA, Right Channel
samples
DM520997.WMA, Right Channel
10 x 105
x 10
1
0.3
0.6
0.4
0.2
5 6
samples
10 x 105
This shows the MATLAB Amplitude and Energy plots.
26


APPENDIX B
OLYMPUS DM-620
Images of the Olympus DM-620 recording, settings, and MATLAB energy
plots.
This shows the file that was created when recording the playback audio file.
This shows the frequency statistics of the first break.
27


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
DM620108.WAV, Left Channel
8 10 samples
DM620108.WAV, Left Channel
samples
DM620108.WAV, Right Channel
8 10 12 samples
DM620108.WAV, Right Channel
14 16

x 10
x10
x 10
samples
This shows the MATLAB Amplitude and Energy plots.
x 10
29


APPENDIX C
OLYMPUS WS-550M
Images of the Olympus WS-550M recording, settings, and MATLAB energy
30


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.
31


This shows the amplitude statistics of the third break.
< -1
WS550121 .MP3, Left Channel
J____________________I___________________I___________________I____________________I_________________U_____________________I___________________L
0 2 4 6 a 10 12 14 16 13
samples xIO
WS550121.MP3, Left Channel
I T
j
ID
*o
Q.
E
<
0 2 4 6 3 10 12 14 16 18
1
0 -1
samples * io5
WS550121.MP3, Right Channel
0 2 4 6 8 10 12 14 16 18
samples
x 10
g>
cG
c
LU
WS550121.MP3, Right Channel
I i 1
1
0 2 4 6 8 10 12 14 16 18
samples xio5
This shows the MATLAB Amplitude and Energy plots.
32


APPENDIX D
OLYMPUS WS-560M
Images of the Olympus WS-560M recording, settings, and MATLAB energy
plots.
33


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
34


This shows the amplitude statistics of the third break.
WS560074.MP3, Left Channel
8 10 samples
WS560074.MP3, Left Channel
x 10
>.
oi
i 8 10 12
samples
WS560074.MP3, Right Channel
samples
WS560074.MP3, Right Channel
14
16
18
x105
X105
tu 0.5 c LU
8 10 samples
12
14
16
18
x 10s
This shows the MATLAB Amplitude and Energy plots.
35


APPENDIX E
OLYMPUS WS-700M
Images of the Olympus WS-700M recording, settings, and MATLAB energy
36


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.


WS700190.WAV, Left Channel
8 10 12 samples
WS700190.WAV, Left Channel
x10J
8 10 12 samples
WS700190.WAV, Right Channel
14
8 10 12 samples
WS 7 00190.WAV, R i g ht C h a n nel
16
18
x10J
x 10
>
E? 0.1 -to
0.05
10
samples
12
14
16
18
x 10s
This shows the MATLAB Amplitude and Energy plots.
38


APPENDIX F
OLYMPUS WS-700M
Images of the second Olympus WS-700M recording, settings, and MATLAB energy plots.
This shows the file that was created when recording the playback audio file.
39


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
4*
This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.
40


Energy Amplitude Energy Amplitude
o 666 oo 6 o
This shows the amplitude statistics of the third break.
WS700084.WAV, Left Channel
0 02 0.4 0.6 0.0 1 12 1.4 1.6 10 2
samples x 10
WS700084.WAV, Left Channel
0 0.2 0.4 0.6 00 1 1.2 1.4 1.6 10 2
samples x 106
WS700084.WAV, Right Channel
0 0.2 0.4 0.6 00 1 12 1.4 1.6 1.8 2
samples x 106
WS700084.WAV, Right Channel
0 02 0.4 0.6 00 1 12 1.4 1.6 10 2
samples x io
This shows the MATLAB Amplitude and Energy plots.
41


APPENDIX G
OLYMPUS WS-750M
Images of the Olympus WS-750M recording, settings, and MATLAB energy
This shows the frequency statistics of the first break.
42


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
plitude Statistics 1.
1 General j Histogram |
1 Minimum Sample Value: 16638 B ?^76 T3
1 Maximum Sample VaMe: 7220 0 16867 10
1 Peak AmpHrxJe. 5.58 dB 0 565 cS is
1 Posstblr Clipped Samples lj 0
DC Olliet 01* .002*
l Maximum RMS Power: ** 0 mi8dB B 1048
1 Average RMS Power: 2$.17d8 -3011 dB
Total RMS Power: 21.42dB -21.78 d8
Actual Bit Depth 6
Copy Data to Clpboatd
WindowWidth: 50 ms
Recalculate RMS 1 w* .1

This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.


This shows the amplitude statistics of the third break.
WS750096.WAV, Left Channel
8 10 12 samples
WS750096.WAV, Left Channel
xIO3
UJ
8 10 12 samples
WS750096.WAV, Right Channel
14
18
samples
WS750096.WAV, Right Channel
samples
18
x 10s
x 10b
x 10
This shows the MATLAB Amplitude and Energy plots.
44


APPENDIX H
OLYMPUS WS-760M
Images of the Olympus WS-760M recording, settings, and MATLAB energy
This shows the frequency statistics of the first break.
45


This shows the amplitude statistics of the first break.
46


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
0.5 0 -0.5 -1
0 2 4 8 8 10 12 14 16 13
samples xios
WS760196.WAV, Left Channel
1 --------1--------1--------1--------1--------1--------1--------1--------1--------1-------
0.5 -
0 2 4 6 8 10 12 14 16 18
WS760196.WAV, Left Channel
samples X10S
WS76Q196.WAV, Right Channel
0.5 0 0.5
8 10 samples
12
14
16
18
x 103
WS76D196.WAV, Right Channel
i-------1-------1--------1-------1-------1-------1-------r
0.5 -
samples
x10S
This shows the MATLAB Amplitude and Energy plots.
47


APPENDIX I
OLYMPUS WS-802
Images of the Olympus WS-802 recording, settings, and MATLAB energy
This shows the frequency statistics of the first break.
48


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.
49


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
802g061.WAV, Left Channel
0 0.2 0:4 0.6 O.S 1 12 1.4 1.6 1.3 2
samples xIO
802g061.WAV, Left Channel
samples x1o6
8020061.WAV, Right Channel
This shows the MATLAB Amplitude and Energy plots.
50


APPENDIX J
OLYMPUS WS-822
Images of the Olympus WS-822 recording, settings, and MATLAB energy
plots.
This shows the file that was created when recording the playback audio file.
This shows the frequency statistics of the first break.
51


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.

This shows the frequency statistics of the third break.
52


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
16121 7q058.WAV, Left Channel
0.5
C-.2 0.4 0.5
0,8 1 1.2 samples
16121 7jj058.WAV, Left Channel
1.4 1.6 1.8
samples
161217g058.WAV, Right Channel
samples
1612170058.WAV, Right Channel
samples
x 10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
x 10
0 0:2 0.4 0.6, 0.8 1 12 1.4 1.6. 1.8 2
x 10
x 10
This shows the MATLAB Amplitude and Energy plots.
53


APPENDIX K
OLYMPUS WS-823
Images of the Olympus WS-823 recording, settings, and MATLAB energy
This shows the frequency statistics of the first break.
54


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
This shows the frequency statistics of the third break.
55


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
samples X106
170825fl067.WAV, Right Channel
1708250067.WAV, Right Channel
This shows the MATLAB Amplitude and Energy plots.
56


APPENDIX L
PHILIPS VOICE TRACER
Images of the Philips Voice Tracer recording, settings, and MATLAB energy
This shows the frequency statistics of the first break.
57


This shows the amplitude statistics of the first break.
This shows the frequency statistics of the second break.
This shows the amplitude statistics of the second break.
58


Energy Amplitude Energy Amplitude
This shows the amplitude statistics of the third break.
0 0.5 1 1,5 2
samples xtO
DVT00590401000.WAV, Left Channel
A 0 0
0 05 1 1.5 2
samples x 106
DVTaOO509O4O10OOO.WAV, Right Channel
0 05 1 1.5 2
samples x106
DVTaOO509O4O10OOO.WAV, Right Channel
0 0.5 1 1.5 2
samples xio6
This shows the MATLAB Amplitude and Energy plots.
59


APPENDIX M
EXCEL SPREADSHEETS
BREAK#1
Images of the Excel spreadsheets that contain all the information that was analyzed from the recordings.
Break #1
Recorder Format Samples Peak Amplitude (dB) Minimum RMS Power (dB) Maximum RMS Power (dB) Average RMS Power (dB) Total RMS Power (dB)
Brand Model WAV PCM/ MP3/WMA Begin End Length Left Right Left Right Left Right Left Right Left Right
Olympus DM-520 WMA 247140 274082 26942 -9.19 -6.65 -44.29 -42.24 -14.03 -12.09 -28.77 -26.81 -22.27 -20.36
Olympus DM-620 WAV PCM 460274 495650 35376 -4.5 -5.04 -39.47 -40.86 -10.54 -11.01 -27.84 -28.56 -19.88 -20.36
Olympus WS-550M MP3 460000 470000 10000 -28.58 -28.91 -40.5 -40.91 -38.82 -38.96 -39.31 -39.81 -39.29 -39.85
Olympus WS-560M MP3 460000 470000 10000 -28.97 -27.8 -41.24 -40.61 -40.27 -39.32 -40.75 -39.89 -40.66 -39.78
Olympus WS-700M WAV PCM 442200 493955 51755 -15.86 -14.5 -40.88 -41.35 -22.12 -20.07 -36.04 -35.59 -32.51 -30.72
Olympus WS-700M WAV PCM 442579 496432 53853 -4.08 -9.73 -43.31 -46.28 -10.26 -15.84 -30.82 -35.52 -21.53 -27.08
Olympus WS-750M WAV PCM 450000 485000 35000 -22.91 -24.85 -37.13 -38.8 -33.45 -35.4 -35.43 -37.31 -35.38 -37.26
Olympus WS-760M WAV PCM 450000 485000 35000 -26.07 -23.81 -40.12 -38.11 -36.11 -33.08 -37.78 -34.77 -37.72 -34.71
Olympus WS-802 WAV PCM 470000 510000 40000 -22.48 -21.74 -36.62 -36.35 -33.09 -32.47 -34.54 -34.09 -34.47 -34.03
Olympus WS-822 WAV PCM 470000 515000 45000 -21.55 -21.14 -36.82 -36.07 -31.23 -30.21 -33.6 -32.8 -33.54 -32.73
Olympus WS-823 WAV PCM 475000 515000 40000 -23.97 -21.43 -37.01 -34.65 -33.77 -31.63 -35.36 -33.34 -35.35 -33.32
Philips Voice Tracer WAV PCM 465549 597150 131601 -15.12 -15.3 -50.59 -51.31 -18.74 -18.78 -42.82 -43.22 -33.76 -33.85
60


BREAK #2
Break #2
Recorder Format Samples Peak Amplitude (dB) Minimum RMS Power (dB) Maximum RMS Power (dB) Average RMS Power (dB) Total RMS Power (dB)
Brand Model WAV PCM/ MP3/WMA Begin End Length Left Right Left Right Left Right Left Right Left Right
Olympus DM-520 WMA 494580 520533 25953 -7.28 -6.74 -44.1 -42.35 -11 45 -10.47 -26.62 -25.61 -19 59 -18.71
Olympus DM-620 WAV PCM 936652 974200 37548 -5.46 -6.73 -39.77 -41.2 -10.62 -11.72 -28.41 -29.63 -20.37 -21.5
Olympus WS-550M MP3 914080 929416 15336 -5.7 -5.5 -40 49 -40.6 -10.32 -10.3 -21.67 -21.74 -16.26 -16.51
Olympus WS-560M MP3 914061 929354 15293 -7.66 -6.01 -41.23 -40.17 -11.57 -10.5 -22.95 -21.77 -17.49 -16.48
Olympus WS-700M WAV PCM 935072 987130 52058 -13.88 -13.78 -40.68 -41.16 -18.33 -18.17 -34.52 -34.56 -29.12 -28.91
Olympus WS-700M WAV PCM 937467 989361 51894 -7.23 -9.05 -43.42 -46.65 -10.94 -12.58 -31.4 -33.51 -22.19 -23.85
Olympus WS-750M WAV PCM 935113 988918 53805 -5.58 -5.65 -40.4 -41.74 -10.18 -10.48 -29.17 -30.11 -21 42 -21.76
Olympus WS-760M WAV PCM 934812 988650 53838 -5.72 -4.93 -40.92 -37.77 -10.68 -10.37 -30.22 -28.77 -21.9 -21.54
Olympus WS-802 WAV PCM 958100 1011825 53725 -3.89 -4.47 -36.74 -36.09 -10.35 -11.52 -28.7 -29.06 -21.61 -22.76
Olympus WS-822 WAV PCM 961200 1017061 55861 -4.46 -3.79 -35.84 -34.96 -10.92 -10.21 -29 -28.24 -22.3 -21.58
Olympus WS-823 WAV PCM 965000 1020886 55886 -4.48 -4.19 -37.96 -35.85 -10.67 -10.51 -29.65 -28.61 -22.09 -21.89
Philips Voice Tracer WAV PCM 1039143 1170716 131573 -14.3 -14.36 -51.54 -51.82 -18.12 -18.28 -42.59 -42.96 -33.38 -33.56
61


BREAK #3
Break #3
Recorder Format Samples Peak Amplitude (dB) Minimum RMS Power (dB) Maximum RMS Power (dB) Average RMS Power (dB) Total RMS Power (dB)
Brand Model WAV PCM /MP3/WMA Begin End Length Left Right Left Right Left Right Left Right Left Right
Olympus DM-520 WMA 741000 767947 26947 -8.9 -6.46 -45.37 -44.8 -14.15 -12.14 -28.94 -27.06 -22.34 -20.41
Olympus DM-620 WAV PCM 1415200 1450579 35379 -4.77 -5.25 -39.89 -41.2 -10.2 -10.75 -27.44 -28.25 -19.62 -20.18
Olympus WS-550M MP3 1373000 1383000 10000 -28.7 -27.62 -40.12 -40.56 -38.88 -39.7 -39.29 -40.04 -39.35 -39.95
Olympus WS-560M MP3 1373000 1383000 10000 -29.01 -28.37 -41.5 -40.08 -40.38 -39.09 -40.82 -39.54 -40.79 -39.52
Olympus WS-700M WAV PCM 1427881 1481726 53845 -15.01 -14.1 -40.96 -41.78 -22.05 -20.15 -36.01 -35.61 -32.33 -30.74
Olympus WS-700M WAV PCM 1430300 1484200 53900 -4.92 -9.89 -42.79 -46.28 -10.29 -15.93 -30.86 -35.7 -21.55 -27.16
Olympus WS-750M WAV PCM 1440000 1475000 35000 -23.9 -25.37 -37.26 -38.76 -33.64 -35.84 -35.61 -37.44 -35.54 -37.4
Olympus WS-760M WAV PCM 1440000 1475000 35000 -25.85 -23.42 -39.12 -36.54 -35.94 -32.33 -37.57 -34.42 -37.6 -34.43
Olympus WS-802 WAV PCM 1460000 1500000 40000 -22.11 -22.47 -36.61 -35.95 -31.74 -31.09 -34.01 -33.42 -33.91 -33.34
Olympus WS-822 WAV PCM 1465000 1505000 40000 -22.52 -20.93 -35.57 -34.72 -31.35 -30.56 -33.12 -32.45 -33.1 -32.4
Olympus WS-823 WAV PCM 1466000 1512000 46000 -23.31 -21.09 -37.14 -35.75 -33.5 -31.29 -35.32 -33.35 -35.31 -33.32
Philips Voice Tracer WAV PCM 1612650 1744191 131541 -14.37 -14.67 -50.73 -51.54 -17.99 -18.28 -42.66 -43.11 -33.1 -33.44
62


NUMBER OF SAMPLES PER BREAK
Number of Samples per Break
Format Recorder Make/Model Serial Number Break 1 Break 2 Break 3 Mean Standard Deviation
Olympus DM-620 100115567 35376 37548 35379 36101 1253.14
Olympus WS-700M 100124078 51755 52058 53845 52552.67 1129.40
Olympus WS-700M 100126397 53853 51894 53900 53215.67 1144.84
Olympus WS-750M 200104369 35000 53805 35000 41268.33 10857.07
WAV PCM Olympus W5-760M 200110592 35000 53838 35000 41279.33 10876.12
Olympus WS-802 100137893 40000 53725 40000 44575 7924.13
Olympus WS-822 100169832 45000 55861 40000 46953.67 8108.97
Olympus WS-823 100258938 40000 55886 46000 47295.33 8021.82
Philips Voice Tracer LFH0882 131601 131573 131541 131571.67 30.02

MP3 Olympus WS-550M 200137081 10000 15336 10000 11778.67 3080.74
Olympus WS-560M 200126197 10000 15293 10000 11764.33 3055.91

WMA Olympus DM-520 100104915 26942 25953 26947 26614 572.45
63


LEFT POWER PER BREAK
Left Power ser Break
Format Recorder Make/Model Break 1 Break 2 Break 3 Mean Standard Deviation
Olympus DM-620 -19.88 -20.37 -19.62 -19.96 0.38
Olympus WS-7C0M -32.51 -29.12 -32.33 -31.32 1.91
Olympus WS-700M -21.53 -22.19 -21.55 -21.76 0.38
WAV PCM Olympus WS-750M -35.38 -21.42 -35.54 -30.78 8.11
Olympus WS-760M -37.72 -21.9 -37.6 -32.41 9.10
Olympus WS-802 -34.47 -21.61 -33.91 -30.00 7.27
Olympus WS-822 -33.54 -22.3 -33.1 -29.65 6.37
Olympus WS-823 -35.35 -22.09 -35.31 -30.92 7.64
Philips Voice Tracer -33.76 -33.38 -33.1 -33.41 0.33

MP3 Olympus WS-550M -39.29 -16.26 -39.35 -31.63 13.31
Olympus WS-560M -40.66 -17.49 -40.79 -32.98 13.41

WMA Olympus DM-520 -22.27 -19.59 -22.34 -21.4 1.57
64


RIGHT POWER PER BREAK
Right Power per Break
Format Recorder Make/Model Break 1 Break 2 Break 3 Mean Standard Deviation
Olympus DM-620 -20.36 -21.5 -20.18 -20.68 0.72
Olympus WS-7C0M -30.72 -28.91 -30.74 -30.12 1.05
Olympus WS-700M -27.08 -23.85 -27.16 -26.03 1.89
WAV PCM Olympus WS-750M -37.26 -21.76 -37.4 -32.14 8.99
Olympus WS-760M -34.71 -21.54 -34.43 -30.23 7.52
Olympus WS-802 -34,03 -22.76 -33.34 -30.04 6.32
Olympus WS-822 -32.73 -21.58 -32.4 -28.90 6.34
Olympus WS-823 -33.32 -21,89 -33.32 -29.51 6.60
Philips Voice Tracer -33.85 -33.56 -33.44 -33.62 0.21

MP3 Olympus WS-550M -39.85 -16.51 -39.95 -32.10 13.50
Olympus WS-560M -39.78 -16.48 -39.52 -31.93 13,38

WMA Olympus DM-520 -20.36 -18.71 -20.41 -19.83 0.97
65


Full Text