Page 29 - Special Issue
P. 29

Forensic Acoustics
Sciences pointed out many areas of concern (National Acad- emy of Sciences, 2009).
For example, voice comparison is a common and important forensic request. An audio recording contains an utterance that the prosecution claims is the voice of the defendant while the defense denies that allegation and claims that the recorded speech was not the voice of the accused but was uttered by someone else. For many years, the common ap- proach was for an expert in speech analysis to employ the aural-spectrographic method, which entails recording the defendant reading a script version of the words spoken in the evidentiary recording. The expert then compares the sound and the visual spectrogram of the evidentiary record- ing and the defendant’s “exemplar” recordings and renders an opinion regarding the degree to which the defendant’s ex- emplar recordings match the evidence. The problem is that the method of “matching” the recordings has traditionally been highly subjective and therefore subject to mistakes and unconscious bias (Bolt et al., 1970; National Academy of Sci- ences, 1979; Poza and Begault, 2005).
Due in large part to the increasing use of statistically strong DNA evidence in crime investigation, the acoustical foren- sics field has motivation to move toward the use of likeli- hood ratio (LR) calculations as a way to deal systematically with the uncertainties present in acoustical comparisons (Morrison, 2011). The LR is a way to consider two compet- ing hypotheses pertaining to a particular event by assessing the probability of an observation of the event given each of the hypotheses (Perlin, 2010; Lindley, 2014). Specifically, the LR is the probability of an observation being made given that the suspect is, in fact, the perpetrator (the prosecution hypothesis) divided by the probability of the same observa- tion being made but that the perpetrator is, in fact, someone other than the suspect (the defense hypothesis).
Consider the hypothetical situation in which a particular bootprint is observed at a crime scene and a suspect is found who is wearing boots that "match" the observed print to a reasonable degree of scientific certainty. The prosecutor’s hy- pothesis is that the suspect is the perpetrator, so the numera- tor of the LR has probability 1. In other words, if the prose- cution hypothesis is correct that the suspect was at the scene while wearing the boots, the probability is 1 that the particu- lar bootprint would be observed. Meanwhile, the denomi- nator addresses the defense hypothesis that the bootprint
28 | Acoustics Today | Summer 2015
was left by someone other than the suspect, which would depend on the probability that someone else with the same size and type of boots could have left the print at the crime scene. If only 1 person in 100 owned that particular size and type of boot in the local population, the naive probability of observing the bootprint if someone other than the suspect was at the crime scene would be approximately 1/100, and so the LR in this example is 1/(1/100) or 100. Note that the LR is not in and of itself an objective way to establish guilt. The LR formulation needs to be interpreted as the statisti- cal increase in belief in a match to the suspect based on the particular evidence. Thus, the LR must be multiplied by the “odds of guilt” that are established based on other evidence and testimony in the case (Perlin, 2010).
A common difficulty for forensic examination is that the evi- dence is distorted, noisy, smeared, or otherwise incomplete. Determining the degree to which a smudged fingerprint matches a particular record in a fingerprint database or the degree to which a recorded utterance matches an exemplar recording ultimately requires discussion of the scientific models and assumptions used.
Forensic acoustics is an interesting specialty field that lies between the scientific world and the legal world. Although the world of science is accustomed to the process of develop- ing new theories and practical techniques that are gradually tested, refined, and modified incrementally through new experiments, publications, and peer review by fellow scien- tists and engineers, the legal world revolves largely around precedent and the need for judges and juries to make final decisions in a reasonable and quick manner. Members of the ASA are encouraged to become involved in improving the reliability and flexibility of acoustic forensic science, thereby enabling law enforcement, judicial, and accident investiga- tion professionals to work with increased scientific assur- ance and confidence.
  Rob Maher is Professor and Depart- ment Head of Electrical and Computer Engineering at Montana State Univer- sity in Bozeman. He holds a BS degree from Washington University in St. Lou- is, a MS degree from the University of Wisconsin-Madison, and a PhD from
Spring 2020, Special Issue | Acoustics Today | 29 Reprinted from volume 11, issue 3

   27   28   29   30   31