12 June 2013
An ongoing murder case in Florida, USA is discussing the vexed issue of whether it is possible to identify a speaker from a tiny, barely intelligible ‘grab’ of poor quality audio. One of the issues is the extent to which speaker comparison depends on prior decisions about what is being said (i.e. forensic transcription). The USA is very open about their court proceedings, so we are able to follow along with the debate. In this post you can hear the audio, and then (preferably in that order) read about case and listen to the expert testimony.
Before reading about the case, you might like to listen to the crucial part of the audio without context. (Warning: it is potentially distressing.)
And with context (an excerpt from the emergency 911 phone call – you will hear how most of the background speech is drowned out by the caller’s voice, making the above excerpt the only usable part).
And here are samples of the two voices with which it is being compared.
First the voice of Trayvon Martin from an earlier (and completely unrelated) recording.
Then the voice of George Zimmerman from his phone call to police.
Can you tell which of these two speakers was screaming in the first excerpt? Do you think it is possible for anyone to state with certainty which it is (even knowing it has to be one of these two)? As you have probably guessed by now, that is what a witness in the trial claims to be able to do.
Now you might like to go ahead and read the story. A lengthy wikipedia article gives very detailed coverage of the whole case, including (towards the end) several transcription issues other than those discussed here.
A few shorter news articles focused on the voice evidence are linked below. You may notice that one of the expert witnesses is world renowned expert on forensic phonetics and transcription, Prof Peter French. As you might expect, he opposes the witness who claims the voices in the two audio extracts above come from one speaker (you can watch his evidence below). Thanks to another forensic phonetics expert Prof Paul Foulkes, for bringing this important case to my attention.
- ABC (America) Digital News 7 June (this is the video from which the audio above was extracted)
- Orlando Sentinel 8 June
- ABC (America) News 7 June
- Christian Science Monitor 8 June
- CNN 9 June
- ABC (America) Digital News 10 June
- ABC (America) Digital News 17 June more experts called but still no decision on whether voice evidence will be allowed
- 21 June: still no ruling on the voice evidence but some interesting new developments in the Orlando Sentinel and on Reuters
The voice evidence
Below is Prof French’s evidence. The first 15+ mins is Prof French giving his credentials. At about 19 mins he goes into general methodological issues. From about 25 mins he talks about the problems of comparing screaming and modal (normal) speech. About 40 mins he moves onto the speaker comparison issues in this case (note especially his comments at around 50-55 mins in which he discusses voiceprints) and at 57 mins he talks about how the opposing expert witness’s analysis depends on prior transcription of the screaming in the above audio extract as ‘stop’, which Prof French argues is unjustified. Around 1 hour 2 mins he goes into some transcription issues similar to those discussed on this site.
More on the voice evidence
Other voice witnesses can be accessed from the same YouTube page; the evidence Prof French is opposing is that of a Mr Reich, who first gives his credentials, then from about 15 mins discusses his analysis in this case (well worth a listen after you have heard Dr French’s comments).
If you want to pursue the voice issues for yourself, you can hear all the audio from the case, and read a lot of public commentary, at this site supporting the Martin case.