Last week I received a phone call from Guy Gordon, a WDIV TVR NBC affiliate here in Detroit, Michigan. He came to the Primeau Forensics office, bringing with him a recording of an unknown voice making certain remarks. Mr. Gordon asked us to perform a voice identification analysis on the recording. The assignment was to determine if Warren Mayor Jim Fouts was the voice speaking in this recording.
Mr. Gordon also brought some recordings with known samples of Mayor Jim Fouts speaking, which I compared with the unknown recording. I performed two forms of voice identification testing (also known as speaker recognition) with a second audio forensic expert employed with Primeau Forensics. We performed both aural spectral voice ID as well as biometric voice identification testing. After analyzing the results of the testing, we concluded that the voices were a ‘probable match’. The biometric software came back with a high probability of a match between the voices across the majority of the tests. The aural spectral method also showed high consistencies between the voices with very little difference in their pitch. I was asked to interpret the results in percentage of probability and I reported an 80% chance that the unknown voice in the recording was that of Mayor Jim Fouts.
During a voice identification analysis, it is important that I stay unbiased and focus on the science of the investigation. I have no stake in the outcome of the testing and I make sure that anything I report is based on my data and analysis. This ensures that any other expert could perform the same tests and provide the same results.
It’s important to note that the test I performed would be considered a preliminary or ‘Emergency Voice Identification Test.’ The reason I call it an ‘Emergency’ test is because an immediate analysis was required and we therefore had to use any available sample of the suspected person’s voice to perform the comparison. In an ideal voice identification analysis, we would create what is called an exact exemplar. This is a recording of the suspected person reading the same dialogue as spoken in the unknown recording. In this test, we used both an in-person interview with Mayor Jim Fouts, as well as a radio interview found online. For the purposes of this test, we were able to obtain accurate voice information from the found samples. Creating an exact exemplar would lend further credibility to the analysis and results.
In many cases, an Emergency Voice Identification Test is the best option after the event in question occurs. Sometimes the suspected party is not willing to create an exact exemplar, and in some cases the client does not want to alert the suspected party of the testing until some results have been found. Threatening calls are especially difficult to deal with, as alerting the suspected person prior to any conclusive results can be dangerous. Providing these initial results can provide the company or client the support they need to take the proper steps forward.
In this case, the emergency voice identification test was the fastest way to get results so that other city officials could move forward. We performed all the available testing to compare the voice directly with known samples. My conclusions were based on the data and results that were calculated from the voice samples. By basing my conclusions on the data, I avoid as much bias as possible so that the results are objective. High consistencies were observed between the unknown evidence recording in question and the two sample recordings we had available. The pitch comparison, formant comparison, and biometric comparison all showed the same results which led me to my conclusion of Mayor Jim Fouts being a probable match to the voice in question.
In 1991, recordings thought to include the voice of President Donald Trump was thought to have posed as a publicist to brag about himself. Full story can be seen here: https://www.apnews.com/5f3ad25e150b1c20074abfa9160d30a5
According to forensic testing using Biometric Voice Identification testing, the voice recordings thought to be Donald Trump are not the voice of Donald Trump.
A lead story from Washington Post this morning is regarding a recording that was thought to be Donald Trump. Trump denied the recording was his voice. Primeau Forensics was asked by The Washington Post to perform a forensic voice identification test to determine if the unknown voice in the Washington Post story features the voice of Donald Trump.
Voice Identification is the process of comparing ‘known’ voice samples in a recording against a recording of an ‘unknown’ voice to determine if the speakers are consistent. In this case, the ‘unknown’ sample recording is the publicist recording. Audio Forensic Expert makes use of critical listening, frequency and waveform analysis, and employs voice biometric software to ensure that the analysis is accurate.
Our senior expert, Ed Primeau has been practicing audio forensics for over 35 years. He has specific training in Voice Identification Analysis and has extensive experience testifying in court.
Primeau Forensics located a C-Span interview from 1991 titled ‘Donald Trump on Economic Recovery’. We chose this recording as the ‘known’ voice of President Donald Trump voice for forensic comparison. Therefor, we chose this older voice sample because it was closer in time to the ‘unknown’ recording.
The biometric software program that we used is a Speech Pro Product titled ‘SIS 2’. We formatted each speech sample and loaded them into the biometric software.
The result was a 97.5% mismatch meaning the ‘unknown’ voice recording that surfaced in the Washington Post today is NOT the voice of Donald Trump. The image above is a screen shot of ‘SIS 2’ with the test results highlighted.
As an audio forensic expert, one of my day-to-day activities is voice identification and speaker recognition. Primeau Forensics receives calls on a weekly basis from concerned businesses and citizens for help identifying a derogatory voicemail message or threat to an employee or supervisor.
Also common are requests to enhance and clarify fraudulent calls placed to insurance or credit card companies during which a bogus transaction takes place by people impersonating account holders.
In cases that I have testified on successfully this year, voice identification has assisted the courts in settling disputes from individuals who have falsified their voice identification.
So how do you know if you have a recording that could successfully be identified? Here are three things to consider before proceeding with the voice identification test.
1. Does the recording have at least 20 words?
In order to perform a successful voice identification analysis, there must be enough of the ‘unknown’ voice to analyze. If a recording has between 20 and 50 words, it is an ideal sample of the unknown voice. More than 50 words can be excessive for an accurate test, though extra voice samples can’t hurt.
2. Can an exemplar of the suspected person be created?
In order to complete a voice identification, a sample of the suspected person or ‘known’ voice must be created under the supervision of the forensic expert. This sample, or exemplar, of the known voice should be recorded in the same manner that the unknown voice was recorded in and the suspected person should read the exact same words spoken in the unknown voice sample. Without this exemplar recording, a voice identification test may not be possible.
3. How has the court handled voice identification in the past?
There have been both positive and negative rulings on voice identification over the years. As technology grows, voice identification has become much more accurate and is becoming more and more accepted. The oral spectral method is now being accompanied by speech biometrics as a means of confirming results. When both of these methods are used, there is a strong amount of scientific evidence that is produced. With objective scientific evidence, the court is much more likely to accept these analyses into evidence.
Part of the reason oral spectral voice identification has been under criticism in the past is because much of the testing procedure is subjective to the voice identification examiner. Electronic measurement and critical listening are the two primary testing procedures.
Using voice biometrics in addition to the oral spectral method adds to the credibility of the voice identification test because the software has several algorithms that measure the known and unknown voices scientifically to arrive at a percentage of certainty that the known and unknown voices are either identical, might be the same or are not the same.
The five levels of voice identification results are:
possible identification
probable identification
possible elimination
probable elimination
inconclusive
Based on the increasing technology and increasing need for voice analysis, it is our opinion that voice identification is here to stay. Scientific methods for voice identification are constantly being tested and improved by researchers and software development companies. It is a powerful tool that more and more litigations are making use of.
So, you may have heard the rumor that Sir Paul McCartney is dead. In fact, the evidence that has been presented over the years is quite entertaining. The ‘Abbey Road’ album cover with Paul barefoot and songs like Strawberry FieldsForever or Revolution 9, that are purported to state messages regarding Paul passing, have all been linked to McCartney’s alleged death.
The rumor is that Sir Paul McCartney died in a car crash and was replaced with a man named Billy Shears who won the Paul McCartney lookalike contest. I am writing this blog post today to end the years of rumors with actual science: voice identification testing. The full voice identification report can be viewed below.
This assignment began when I received a call from Paul DuBay, a Beatles fan from San Antonio, Texas. He retained me to conduct this voice identification test because he wanted to know the truth.
The goal of voice identification testing and speaker recognition is to compare the known and unknown voices using critical listening, electronic measurement, and visual inspection of sound wave formation and spectrogram. The software programs I used for this voice identification test include Adobe Audition, Sony Sound Forge and Easy Voice Biometrics. Biometric technology is used as a secondary voice identification and speech recognition tool.
I would like to add that this is considered an exception to voice identification testing on many levels. Voice identification testing must be performed using an exact exemplar of spoken words. This is voice identification testing that include science in the form and structure of known and unknown vocals instead of spoken word. To further perform testing on Paul McCartney’s voice spoken words must be used before 1966 and after 1966 that are exact. Perhaps someday that test will be performed in the meantime enjoy the following post
For this untraditional voice identification test I used various songs performed and recorded before and after 1966 that feature the voice of Sir Paul McCartney. I was asked to compare songs from both of these time frames (pre and post 1966) to determine if the current Sir Paul McCartney is the same person as the pre-‘Paul is Dead’ (PID) Paul McCartney. Below is a report that I prepared for Mr. Dubay as a result of my testing.
26 January 2015
Dear Mr. DuBay,
I am an audio and video forensic expert and have been practicing for over 30 years. I have testified in several courts (See updated CV attached) throughout the United States and worked on various international cases. My forensic practices for audio investigation include digital and analogue audio authentication, restoration and voice identification. As a video forensic expert, my practices include video authentication, restoration and identification.
I received from you the following digital audio music files:
Song Title/Music File Name: LP Title:
‘1963 I Saw Her Standing There.m4a’ Please Please Me
‘2002 I Saw Her Standing There(Live).m4a’ Back In the U.S
‘1965 I’ve Just Seen a Face.m4a’ Help!
‘1976 I’ve Just Seen a Face (Live).m4a’ Wings Over America
‘1964 Kansas City.m4a’ Beatles For Sale
‘1988 Kansas City.m4a’ CHOBA
‘1967 Sgt Pepper’s Lonely HCB.mp3’ Sgt Pepper
‘1964 Long Tall Sally.m4a’ Beatles Second Album
‘Too Many People.m4a’ Ram
These songs were “before and after” samples from either the pre-publicity of Paul McCartney’s alleged death in 1966 or the post-‘Paul is Dead’ (PID) news story.
I am familiar with the theories that Paul McCartney was killed in a car accident in 1966. This is why you contacted me and asked that I perform voice identification and speaker recognition testing on various songs performed and recorded before and after 1966. You asked that I compare these songs from both of these time frames (pre and post 1966) to determine if the current Sir Paul McCartney is the same person as the pre-‘Paul is Dead’ (PID) Paul McCartney.
The goal of a voice identification test and speaker recognition is to compare the known and unknown voices using critical listening, electronic measurement, and visual inspection of sound wave formation and spectrogram. The software programs I used for this voice identification test include Adobe Audition, Sony Sound Forge and Easy Voice Biometrics. Biometric technology is used as a secondary voice identification and speech recognition tool.
I help my clients understand voice identification testing and speech recognition by using familiar voice examples. If you are in your office and a fellow employee comes into the office that you have worked with for many years and says ‘hello’, you recognize that voice without making eye contact. The same is true when you are at home and your spouse or even a relative comes in and says hello and begins talking to you. You know who the voice is before making eye contact because you are familiar with the voice. This is how critical listening examination is conducted during voice identification tests. I have been performing voice identification testing most of my career as an audio forensic expert. In fact, throughout the course of my career I have performed dozens of successful voice identification tests including a test for CNN on the voice of Apple’s ‘Siri’.
When beginning a voice identification test, I first become extremely familiar with both the known and unknown voices and list all similarities as well as differences during this repeated listening. In the case of Sir Paul McCartney, I listened to all songs repeatedly during this critical listening phase. I also measured and viewed the sound spectrum and wave formation repeatedly to arrive at my professional conclusion.
The following report will include descriptions of the similarities observed during critical listening, electronic measurement, visual inspection as well as biometric testing. I have not observed any differences in any of the voice samples tested.
I began by downloading the digital audio files that you sent onto my forensic computer then opened all using Adobe Audition CS 5.5. Next, I began critical listening to all of the vocal samples multiple times to become extremely familiar with all voice samples presented.
I focused first on the two samples of ‘I Saw Her Standing There’ as they were superior audio samples that also include a vocal number count at the beginning. Next, I focused on ‘Long Tall Sally’ and Sergeant Pepper’s Lonely Hearts Club Band’ as their vocal delivery are extremely similar and unique.
The beginning count (1-2-3-4) of the ‘1963 I Saw Her Standing There.m4a’ version and the beginning count (1-2-3-4) from the ‘2002 Live Version of I Saw Her Standing There’ is identical which indicates the rhythm of Sir Paul’s internal metronome is the same. The vocal range and phrasing in both samples is also the same. The slight difference in vocal tone is attributed to the age difference of Sir Paul as he has matured over the years and so has his voice.
The spectrogram image below shows exact frequency spectrum in spite of the difference in years:
In the above image, the spectral frequency display is shown in the lower half of the image in color. The yellow, brighter colors indicate the stronger higher volume frequencies present in that portion of the audio while purple and black colors represent frequencies that are weaker or lower in volume. The audio on the left side of the image is the original recording of the song from 1963 and the portion on the right is the more recent recording from 2002. These can be heard in the comparison audio work product attached to this report.
When closely examining the formant frequencies shown in the spectral display above, it is noted that they are nearly identical. Formants are resonances or spectral created by a human voice. These are the frequencies that have the highest presence in a person’s voice and determine most of the tonal qualities of that individual voice. Because the formants in both recordings are almost identical, I conclude beyond a reasonable degree of scientific certainty that they are the same voice, Sir Paul McCartney. The slight variations can be explained by age difference of the voice between the recordings.
The next spectrogram image below shows the sample ‘Wooo’ from the chorus of ‘I Saw Her Standing There’. The original recording from 1963 is displayed on the left and the newer recording from 2002 is shown on the right. Note the fundamental frequencies that are circled in blue in both recordings. Through close examination (narrow band spectrum analysis), it is clear that the fundamental frequencies, harmonics and the range of the frequencies in the early recording and more recent recording are identical. Critical listening also revealed no differences between the two samples, which can also be heard in the comparison audio work product attached to this report. Therefore, through both visual inspection and critical listening, I have determined that the voice in each sample of ‘I Saw Her Standing There’ is the same voice and that of Sir Paul McCartney.
I continued investigating all songs submitted for voice identification testing and found similar results and arrived at the same conclusion. All vocals from all songs submitted are that of Sir Paul McCartney.
My next comparison was between the ‘1964 Kansas City’ Recording and the ‘1988 Kansas City’ recording. This is shown in the spectrogram image below:
The sections that I chose to test are circled. They are the words ‘Kansas City’. The recording to the left is the recording of the song from 1964 and the recording of the song on the right is from 1988. Through close visual inspection of the prominent frequencies in the words ‘Kansas City’, I found that both the fundamental frequencies and the frequency ranges are again nearly identical. I also used critical listening to further support my findings and have determined that the voice from the 1964 recording ‘Kansas City’ is identical to the voice in the 1988 recording. The vocal expression, pronunciation of the words and voice range are an exact match. I continue to conclude that all vocals from all songs submitted are that of Sir Paul McCartney.
The voice tones of all songs examined old and new, are extremely close and often identical when listening critically and viewing the narrow band waveform and frequency spectrum. Songs that were recorded farther apart in time have some small differences, which can be explained by Sir Paul’s difference in age when they were recorded. As people age their voice changes and so does their body. Vocal chords mature and usually grow deeper. Think of a boy going through puberty, the vocal chords mature and so does the voice. The same applies to people who enter their later years. Even though there are these slight differences, fundamental parts of the voice always remain the same.
I believe and will prove scientifically that a person’s singing voice is as unique if not more unique than their speaking voice. In the following paragraphs I will compare the Pre 1966 Paul singing style with the post 1966 Paul singing style by critical analysis of Sir Paul’s vocal timber and very loud and distinct voice. I have made observations while critically listening to upper register, near falsetto, voice signatures for the songs Long Tall Sally and Sgt Pepper’s Lonely Hearts Club Band. I will use the studio recorded versions of both songs, however, I have chosen the deconstructing Sgt Pepper isolated vocal to compare to the Long Tall Sally vocal. This back to back comparison file will be available in my audio work product attached to this report.
In Long Tall Sally, pre 1966, Sir Paul’s voice is very forceful and distinct. His vocal range and style of delivery exact. His O’s in the lyric ‘OOOUU Baby- some fun tonight’ match his vocal style in the first verse of Sgt Pepper’s Lonely Hearts Club Band.
Furthermore, listening to the 1973 studio recording of ‘Too Many People’ during the intro near falsetto adlib ‘piss off cake ay ay ay…ooooh’ at the beginning of the song, I clearly hear the exact same falsetto vocal Sir Paul delivered during his entire career with the Beatles and solo. It is an extremely unique style of singing that can only be produced by the real Sir Paul McCartney. See audio work product attached to this report. All vocals from all songs submitted are that of Sir Paul McCartney.
In the image below, Sgt Pepper is on the left and Long Tall Sally is on the right. Notice the sound spectrum ‘fingerprints’ between each vocal sample are nearly identical in display. Considering that these are different songs, this is a very significant identifier that both vocals are sung by Sir Paul McCartney.
I have also loaded isolated segments from ‘Sgt Pepper’ and ‘Long Tall Sally’ into a voice biometrics software program which is capable of taking unique but different voice samples and comparing them biometrically resulting in a percentage of certainty for identification.
I loaded isolated sections of Sir Paul from the beginning of ‘Long Tall Sally’ and the isolated first verse of Sir Paul’s vocal from ‘Sgt Pepper’s Lonely Hearts Club Band’ into Easy Voice Biometrics. The test resulted in a 53 % match. I believe the percentage rate is high enough to confirm a positive identification even though by biometric software standards we would like to see a higher percentage of certainty. In my opinion, this is due to the different words being sang and measured in two different songs. See screen shot of EVB test result below:
The biometric test was done as a secondary test to determine the voice similarities using another voice identification and speaker recognition testing process. Critical listening is the primary voice identification tool that I used to arrive at my conclusion (please see and hear audio work product attached to this report).
Conclusion
Listening to people like Dick Clark and Ringo Starr speaking as well as singing through the years, you can hear how their voices have matured yet are identifiable as being from the same person. This maturity fact is why I point out that Sir Paul McCartney’s voice has also matured. Sir Paul has an incredible voice that is extremely unique and, based on my 31 years experience as an audio forensic expert and scientific forensic testing; there is no other voice in the world that comes close to sounding the same or measures spectrographically the same as Paul McCartney.
Through careful analysis of the waveform and spectrogram as well as critical listening and biometric measurement, I conclude beyond a reasonable degree of scientific certainty that the voice heard in all of the song samples examined is of Sir Paul McCartney. This voice identification test confirms the rumor that Paul is Dead is not true.
This concludes this voice identification testing.
Respectfully submitted,
Edward J. Primeau, CCI, CFC
For more on Voice Identification, check out Ed Primeau’s latest book, “That’s Not My Voice!” available on Amazon.
One of the most important elements of Voice Identification is the ability to recognize the characteristics of the human voice. There are many elements to distinguish these characteristics, some audial, some visual.
Think about when you have your back to a person who enters the room and says hello. If it is your child, spouse or co-worker, I bet you recognize them immediately because you are familiar with their voice. This is the starting point for voice identification; becoming familiar with the characteristics of the unknown voice.
I began editing spoken word on reel-to-reel tape with razor blades and splicing tape. I had to learn to visualize the words in my mind’s eye in order to cut the tape in the right place. Today, we have software programs that display the waveform and sound spectrum of the spoken words, which make the editing process more accommodating. The editor can see the way the words look on the computer screen while deciding where to make the edit and connect the sentences, removing the stutters, coughs, gaps and mistakes.
During the editing process, you will learn to listen for voice characteristics almost subconsciously. These characteristics include the way the words are spoken, the word pronunciations, vowel and consonant pronunciations, the recording noise floor (unwanted background noise), the way the words flow together, and significant patterns of speech you may detect, like accent, dialect and impediments, nasal cavity resonance, voice tone and inflection and speech pacing.
Pay attention to both differences and similarities from recording to recording, and take notes on your observations building a speech database for when writing the report.
Exemplars are defined as expert supervised audio recordings of predetermined spoken word samples for the purpose of voice identification comparison. During the exemplar creation process it is important to coach the person (subject) speaking for the recording into the same level of energy as the evidence recording of the unknown voice. Listen to the energy and attitude of the voice you are examining (evidence or unknown recording). Do you hear a mood or psychological characteristic in the voice?
In some bomb threat recordings I have examined, the speakers have an angry, sad or depressed attitude in their voice while speaking the recorded words. It is important to note that at the time of creating an exemplar, the subject is often not in the same psychological state as the individual in the unknown recording. While making the exemplar, do your best to coach the person (suspect) to speak with the same energy as the voice on the evidence recording.
Your critical listening ear will help you complete this process to the best of your (and their) ability. You have to listen critically beyond the subject’s current mood, because it is often difficult to coach them into the mood of the person on the evidence recording. Listen for specific speech characteristics in the exemplar and evidence recordings. What do you notice about the unknown voice that is characteristic of the known voice?
To practice, spend some time listening to spoken word recordings. These can be in the form of talk radio, podcasts and audio books. Write down speaking characteristics of the voice recordings like this:
• English accent
• Southern accent
• Consistent sibilant “s”
• Consistent long “a”
• Medium pitch, low pitch, high pitch
• Emphasis on “al” as in “halp” instead of “help”
• Does the subject have a characteristic rhythm to his speech or a pattern of delivering words and pausing?
Listen to several spoken word recordings and make a list of speech characteristics. Take notes on your observations.
Only through practice and experience will you become familiar with voice identification. When creating a new audio comp or assembly file in Sony Sound Forge or Adobe Audition, you will be able to listen to the speech sections that you are comparing repeatedly and with easy access. Back-to-back critical listening is an extremely important tool for voice identification. It is the best way to develop your critical listening skills and begin to recognize the different speaking characteristics of each voice examined. The familiar and unfamiliar speaking samples can be identified and characteristics can be easily noted.
Learn more about Voice Identification and Critical Listening in Forensic Expert Ed Primeau’s new book, That’s Not My Voice! available now on Amazon.
Back at the end of October, I was contacted by Nicole Bogart from the Global News in Toronto who asked that I conduct a voice identification between known samples of Mayor Rob Ford and a male caller to a radio show named Ian. They believed that the caller who was defending the mayor was not a person named Ian but rather Rob Ford himself.
In the video below, I explain my process and play samples of the audio’s that were used to compare and arrive at my conclusion.
Imagine for a moment that you’re sitting at home after a long day at work, enjoying some time in your living room, watching TV or the like. All of a sudden, you hear the door open behind you. You sit and wonder who could have walked in, until the anonymous person speaks. The second you hear their voice, you recognize and register immediately who has entered the room. The same scenario can happen all day at work as well. You learn to recognize your co-workers voices after hearing them all day for several months, even years.
As these examples show, we recognize and interpret characteristics of voices very quickly. Why is this? Vocal characteristics are very distinct amongst all people regardless of race, gender, or ethnic background. If you spend a decent amount of time around someone, you become used to the vocal characteristics they express and can immediately correlate those characteristics with the identity of who is speaking to you.
Vocal Identification in audio forensics relies on the ability to recognize these characteristics in any unknown human voice. In the examples above, the repetition of exposure to those vocal characteristics make you respond without even thinking twice about who could be behind you. The difference in regards to audio forensics is that a forensic expert doesn’t always know anything about the voice in question. This requires the forensic expert to rely on their ‘critical listening skills.’
So, how is this done? What specific characteristics are forensic experts actually looking for? They look for the types of speaking characteristics that could be relevant and specific in identifying a person’s voice. Everyone has very distinct features to their voices, regardless of how slight or severe they might be. When compiling information for voice identification, the forensic expert must listen over and over to the unknown voice with pen and paper taking scrupulous notes of all speech characteristics. They focus specifically on things such as inflection, pronunciation of certain words, any form of an accent, stutters and lisps, amongst other variables. Make careful and precise notes about all of these variables. They try to be as SPECIFIC as possible. The forensic expert will then create a voice profile for this person.
If by chance the forensic expert has access to the suspect thought to be the unknown voice, this puts them at more of an advantage. The goal then is to create an exemplar of that person saying the same message in question with the same delivery heard in the recording. This is called an ‘exact exemplar.’ They will review and compare notes from both the original recording and the exemplar itself. An exemplar gives you a more neutral quality, and isn’t biased by background noise, feedback, or any other external features that may have affected the original recording.
By understanding the critical listening phase of voice identification, you will better understand the value and importance of voice identification as a tool for the audio forensic examiner and audio forensic expert.
For more on Voice Identification, check out Ed Primeau’s latest book, “That’s Not My Voice!” available on Amazon.
Historically, voice identification has had its ups and downs in the United States Courts. Forensic experts that practice voice identification follow voice identification processes that have been established in the scientific community. However, there are times the voice identification process is necessary with less than perfect circumstances.
In my opinion, especially under less than perfect circumstances, the voice identification process is not always black and white. Litigators need help building a case to either prosecute or defend and some of the main evidence is voice recordings. A majority of the time these voice recordings are threatening and disturbing to the recipients who are targets to the threats left in these messages. Some messages are longer than others – it is the short messages that pose the biggest challenge for voice identification experts.
According to the scientific community, in order to conduct voice identification, the following process is followed by the audio forensic expert:
Make sure the unknown recording has sufficient amount of words
Make sure recording is clear and free from noise
Sufficient frequency range of recording (telephone calls are leader in lack of frequency)
Exact exemplar creation using available exact electronics
Multiple exemplar recording samples of suspect’s voice
Test aural (listening-single and multiple word formation) and spectrographic (visual-electronic measurement)
List all similarities and differences for both aural and visual testing
Create comparison back to back sample file using evidence and exemplar samples
There are five standard conclusions:
Positive identification
Probable identification
Probable elimination
Positive elimination
Inconclusive
In order to arrive at a positive conclusion, the audio forensic expert must find a minimum of 20 speech sounds and/or electronic/spectrographic identifying or disqualifying characteristics. However in the case of a crisis, investigators ask voice identification experts to determine as exact as possible what percentage of certainty can be established with regard to what I consider nontraditional voice identification. In this case I follow as much of the established protocol as possible and provide a percentage of certainty in my findings.
Threatening calls in the workplace require the identity of the caller be revealed or at least narrow down the list of suspects so management can take appropriate actions to protect other employees. These threatening calls are most always short, which presents challenges to the forensic expert due to the number of words available to be compared to known speech samples. In this case the voice identification examiner needs as much exemplar (known speech) recording as possible to use for comparison purposes.
I consider these threatening phone calls and messages to fall under emergency ‘crisis’ situations and adjust the voice identification procedure to accommodate the client and attempt to identify two or three suspects.
Another tool that helps the audio forensic expert in these crisis situations is the voice identification line up. In this style of voice identification testing, it is necessary for the voice identification expert to create a comparison file with back to back samples taken from all suspect exemplar recordings. This comparison digital audio file can be closely examined very quickly noting all similarities and differences to determine a suspect.
Many companies have a database of absenteeism calls from employees. These voice mail recordings are great alternates to exact exemplars because they were recorded through the phone similar to many of the threatening calls. For this reason, I recommend that companies maintain a voice mail recording database for crisis management cases like these. Then when the voice identification expert is called on to provide their opinion about the suspects for investigators to consider, the voice identification expert can help strategize additional steps and strategies to take that will help solve the mystery.
I have found that once a suspect has been identified preliminarily and asked to cooperate in creating an exact exemplar, they either confess or investigators discover additional clues to help with the crisis investigation.
In one particular voice identification case, my preliminary testing helped reveal two suspects. We discovered that the threatening call was made from a company phone. There was a CCTV camera near the company phone which provided video of one of the suspects near the phone the day and time the threatening call was made. The identification could not have been made without the preliminary nontraditional voice identification test that helped identify two suspects. That helped me know who to look for in the CCTV video. The CCTV digital video recording was poor but I was able to identify clothing logos on the person in the video with the suspect who often wore similar clothing. Investigators were able to leverage a confession once the suspect was confronted with this preliminary evidence.
So, although formal voice identification is not always possible, an audio forensic expert can aid in the investigation of threatening voice mail messages and help keep the attacked workplace safe for other employees. When the case reaches a point when a formal charge is made, the voice identification testing can then be directed to traditional testing protocols.
For more on Voice Identification, check out Ed Primeau’s latest book, “That’s Not My Voice!” available on Amazon.
As you probably have read in previous blog postings, I’ve made some comments about the 911 and other police recordings that were involved in the Trayvon Martin shooting case. My opinions were published in the Orlando Sentinel and I also made some appearances on MSNBC and CNN as well as Fox News.
I was contacted by a producer at The Today Show shortly after they acquired a copy of an exemplar that the police made of George Zimmerman’s voice. They made the exemplar (sample of Mr. Zimmerman’s voice) so that it could be compared to the cry for help in the 911 police recording.
The Today Show wanted me to form an opinion and comment while on camera regarding the exemplar that was made by the police. I decided not to go on camera with a comment because the prosecution in this case named me as an expert witness. I did not feel a media update would be appropriate even though the prosecution has yet to contact me since naming me as an expert witness.
However, I do want to go on record and comment that a more appropriate or exact exemplar needs to be made of George Zimmerman’s voice in order to conduct a proper voice identification test.
I believe the police did the best that they could in order to recreate that sample or exemplar of George Zimmerman’s voice. They had him say the word “help” for comparison to the cry for help on the 911 recording. The word “help” was shouted the night this incident took place. The question is who shouted help?
However, what was not done properly when creating the exemplar was the delivery of the word “help” by Zimmerman. The reason being that, it was very different than the long, drawn out scream of help that is heard in the background of the 911 recording which is posted on several places on the Internet. I first learned about that recording from the Mother Jones Blog and, of course, I have discussed it on previous blog posts as well.
I originally formed an opinion that the cry for help was not George Zimmerman’s voice, which was the opinion I shared when asked by the major media. I formed my opinion by using my own critical listening skills that I’ve acquired after 28 plus years as an audio forensic expert who also performs voice identification. No scientific voice identification testing was done because I did not have the proper resources.
The bottom line here is that this has turned into a media frenzy. People in the United States have already formed an opinion as to whether or not Zimmerman is guilty. In some social circles, this case has even taken on a discriminatory slant due to the fact that people are choosing sides based on their own personal views pertaining to racial discrimination.
It’s very unfortunate that Trayvon Martin was shot and killed by George Zimmerman; this is a fact that we know for sure. But it is up to the court and the due process of the United States to determine whether or not George Zimmerman acted in self defense with regard to the shooting.
The way to properly record a voice identification exemplar or sample of the accused voice is to have the person recreate the distance of the cell phone call exactly like the police did. Additionally, the delivery of the words need to be as similar to the original as possible. In this case, I believe that the police could have turned to the FBI for help recording this exemplar.
As an audio forensic expert who has been named as an expert witness in this case, I would have been happy to help create this exemplar as well. And why it was released to the media before consulting with the expert witness team is unknown to me. Below is a link to the audio file that was recorded by the police of George Zimmerman’s voice which was considered an exemplar. Also, s the original 911 call is now. Listen to both of them and decided whether or not it is the voice of George Zimmerman.
[audio:http://WWW.AUDIOFORENSICEXPERT.COM/wp-content/uploads/2012/07/Female-911-Call-with-Gunshot.mp3|titles=Female 911 Call with Gunshot]
[audio:http://WWW.AUDIOFORENSICEXPERT.COM/wp-content/uploads/2012/07/VoiceExemplars_03222012_643pm.mp3|titles=VoiceExemplars_03222012_643pm]
For more on Voice Identification, check out Ed Primeau’s latest book, “That’s Not My Voice!” available on Amazon.
Based on my 28+ years working with the human voice for identification purposes, I have been asked on several occasions to determine a person’s identity based on a voice sample when an exact exemplar was not available. I call this nontraditional voice identification. I do not make this type of voice identification a day to day practice and I always let our clients know that my findings are preliminary based on experience rather than exact science.
One example is the Elvis Presley voice identification I did. Since Elvis is dead, I could not record an exact exemplar.
Because the family who produced the song was convinced it was Elvis and needed my opinion, I compared the unknown song to known Elvis Presley songs from the same timeframe. My opinion was that the new song is the voice of Elvis. Not everybody agreed with my opinion, however. Since science was not involved, these people had reason to base their opinion on their personal knowledge of Elvis Presley’s style.
The above Elvis story is a good example of nontraditional voice identification. An emergency situation is another example, such as when a threatening call was made with intent to do bodily harm to a person or persons. Emergency voice identification becomes necessary to provide the authorities with evidence so they can order the accused to cooperate and provide the voice identification expert an exact exemplar.
One scenario is when a bomb threat call is made to a place of business. Voice identification becomes necessary to determine if the threat was made by a previous employee. I have worked on cases like these many times. Management often has an idea of who made the threat. Their opinion is based on more information than a voice identification expert would use such as discharge date, personality characteristics and work history. I would conduct an emergency voice identification using known samples of the accused person’s speech, like voice mail recordings when the suspect employee called in sick.
Here are the steps I take for emergency voice identification:
1. Locate a known sample of the accused voice
2. Locate the threatening voice recording
3. Make a list of all speaking characteristics of the known voice. These include stutters, vowel pronunciation, any other speech observations and consistent speaking styles that can be observed.
4. Make a list of all speaking characteristics of the unknown voice. Are there any similarities or differences that can be assessed?
5. Use a spectrum analyzer to note voice spectrum of both samples. Are the samples close enough in spectrum to be a match or different enough to help arrive at a conclusion?
The voice identification expert’s task is to take all information from critical listening, spectrum analysis or electronic measurement, and visual inspection of the wav formation and decide if the voice in question is the same voice or not. The next step is to have the law enforcement officials use the voice identification expert’s report to encourage the suspect to cooperate and record an exact exemplar.
