forensic use of voice prints

[Mark_Mandel at DRAGONSYS.COM]

My co-workers at Dragon Systems have offered the following replies:

>>>>>
1:
There is a whole subsection of the speech community that is concerned with
speaker ID. See, e.g., the Proceedings of RLA2C, La Reconnaissance du Locuteur
et ses Applications Commerciales et Criminalistiques / Speaker Recognition and
its Commercial and Forensic Applications, Avignon 20-23 April 1998. ESCA and
IEEE were among the sponsors. See
     http://www.lia.univ-avignon.fr/RLA2C/

2:
I do not know of specific cases, but am aware that there is considerable
skepticism in the speech community on the validity of them as evidence.  Certain
"experts" in the forensic community claim to be able to reliably identify
individuals using spectrogrms.  Many of us in the scientific community do not
subscribe to that view.  Those who call them "voiceprints" are especially
suspect, since spectrograms are definitely NOT analogous to fingerprints.

I believe that Kenneth Stevens of MIT has been active in this discussion.

3:
Try asking on the newsgroup comp.speech.research.

4:
I recall having worked on a project (back in Rhode Island when I was a graduate
student) wherein the problem being addressed (by me) was that of enhancing quiet
speech (whisper to be specific) of interest to the police and a National TV
network station. The problem was to make the whisper intelligible (using
reference background speech data and adaptive noise canceling techniques). The
goal was to gather evidence towards a "murder trial". Since the project was
classified, I won't be able to provide more specifics. In general though, my
impression is that if an individual (or a group) with sufficient professional
credibility testifies about forensic voiceprints, then it may be used as one of
the evidences (but not a sufficient evidence) in the trial.

Hope that ,s of some use.

5:
There was a huge fight over Voiceprints at the National Research Council in the
80's, and they found that there was no evidence that a voiceprint assisted
operator could better identify a speaker than the operator alone without the
help of the voiceprint.

The most vigorous pursuers of voiceprints have been the Michigan State Police -
I'd look there for court cases.  The courts have been back and forth, and I
think that while the scientific evidence is that voiceprints do not work for
identification, they apparently are quite good for assuring non-matches if the
speaker's characteristics are far apart from the sample.  The courts have been
all over the place.

Sandra Disner (at UCLA) is an expert in these matters, as is Peter Ladefoged.
I'd ask them for help.

6:
One of my closest friends is State Medical Examiner for the State of New
Hampshire and a world authority in forensics. I will forward her the email and
see if she has any insight.
<<<<<

Hope this helps.

   Mark A. Mandel : Senior Linguist and Manager of Acoustic Data
         Mark_Mandel at dragonsys.com : Dragon Systems, Inc.
 320 Nevada St., Newton, MA 02460, USA : http://www.dragonsys.com