Durand Begault, Ph.D., FAES, Charles Salter and Associates
On Tuesday, June 27, 2017 at 8:00 p.m., the Audio Engineering Society, Los Angeles Section heard a talk entitled Forensic Analysis and Audio Engineering: Digging Through Trash for Treasure, a presentation from distinguished guest Durand Begault Ph.D, FAES of the Charles M. Salter Associates Audio Forensic Center in San Francisco. Dr. Begault covered a range of topics in audio forensics, including the authentication and enhancement of voice recordings; unique considerations concerning cellular telephone voicemail; the reliability of gunshot analyses; comparisons of music in copyright infringement cases; and finally, the overall stands of science and how well do current audio forensic practices conform to them.
Dr. Begault began by describing audio forensics as the "'Scientific' analysis of audio and acoustics pertaining to a legal matter." He is frequently asked to enhance a noisy recording to make speech more intelligible, enabling transcription. In some cases he is even asked to make an event detectable. This naturally requires that there is sufficient signal in the noise to begin with, which is frequently not the case. He describes how the process of enhancement requires a great deal of judgment, from how to prevent the addition of new audio artifacts, as well as how to control one's own biases. He notes that one must be willing to label parts of a recording as unintelligible if need be, and that one must resist loosening appropriate standards in order to please a client. Furthermore, one must achieve a standard of clarity and intelligibility that would be available to a jury, not just to a pair of "golden ears." He explains that as an audio forensics consultant, his determinations are technical and not evidential¬-a judge determines whether the evidence can be admitted into a case, but the forensic audiologist provides technical expertise on matters such as provenance (where did it come from?) and continuity (has it been edited?). The tools a forensic audiologist uses are critical listening, waveform analysis (particularly with respect to discontinuities), spectral analysis, metadata analysis (an increasingly rich area for data gathering), hashing to compare digital identity, and Electrical Network Frequency Analysis (using background electrical grid noise and frequency variation to narrow down locations). Dr. Begault noted that the Audio Engineering Society was a pioneer of forensic audiology with its 1972 report on the Nixon tapes and established many of the practices still followed today.
Dr. Begault described how vocal analysis can be quite challenging, and despite tools such as voiceprints and voice biometrics, the intra-speaker variability of the human voice is quite large and often confounds inter-speaker variability. Dialects, emotion, different languages and recording quality all have a deleterious effect on establishing whether a particular person is the source of a recorded voice. A notorious recent case involved the Trayvon Martin/George Zimmerman case, in which the opinions of two forensic audiologists were excluded from evidence -one asserted that a scream heard on the 911 recording was "not George Zimmerman" while the other offered that it was "probably Mr. Martin." Beyond the variability and capability of the human voice, the circumstances under which recordings are made create further uncertainty-non-optimal microphone placement, multiple microphones in a cell phone and noise cancellation, room acoustics, and audio codecs all limit the "Speech Transmission Index," a "single-value quantifier of the effects of signal-noise ratio and reverberation on modulation depth of the speech envelope." For instance, a narrow-band cutoff of 3400 Hz can make it impossible to distinguish seed from feed or thin from fin. As a forensic audiologist, Dr. Begault has visited locations and recreated the conditions under which the original recordings were obtained, in order to determine whether the verbal testimony of victim and accused are consistent with what is available on cell phone and 911 recordings.
Moving on to the topic of gunshot analysis, Dr. Begault described gunfire as an "extremely brief and powerful impulse event, loud enough to cause hearing damage and capable of overloading a recorder." He is generally asked to answer whether recorded events are indeed gunfire, to verify the type of weapon, the location with respect to the microphone(s), and the timing and number of shots. The sources of these recordings are quite varied, and can include dashcams, bodycams, taser-cams, dispatch center recordings, security cameras with audio, and commercial gunshot detection systems. These recordings can be subject to multiple distortions, including but not limited to background noise, overlapping gunshots, inadequate signal-noise, codec distortion, time variance distortion caused by moving sources or recording devices, and environmental context effects of reflection and absorption. These confounding effects can cause experts to achieve different conclusions, Dr. Begault showing a graph noting the discrepancies in the timing of gunshots found by four different forensic audiologists.
After discussing the topic in greater depth, Dr. Begault segued into an overview of forensic musicology, breaking it into compositional, recording, and production analyses. Compositional analysis involves copyright infringement irrespective of the recording media, and re-quires expertise in music theory and compositional analysis. This type of infringement is difficult to prove, and requires that the defendant had prior access to a song, that the song under examination has substantial similarity in melody, rhythm and structure to the prior work, and that the elements themselves are copyrightable. In the case of Gaye vs. Thicke, two experts reached opposite conclusions, Judith Finnell claiming that the songs were "the same," where Sandy Wilbur asserted that they were not "meaningfully similar," and were "really different songs."
Dr. Begault concluded his discussion by discussing the issues of reliability of the evidence evinced by forensic audiologist. He noted that at the current time, it's very difficult to assign measures of the accuracy of the inferences, nor even a measurement of the uncertainty of the judgments made by forensic audiologists. Replication of results is difficult, statistical analysis can rarely be performed, contexts vary greatly, all of which make standardization and consistent results challenging.
The Audio Engineering Society, Los Angeles Section wishes to thank Dr. Durand Bergault and the Audio Forensics Center of San Francisco for his fascinating and stimulating talk.
