forensic use of voice prints

[Donald M. Lance]

AAllan at AOL.COM wrote:

> The late Henry M. Truby of Florida, an ADS member among other things, was a
> strong advocate and practitioner of voiceprints. I don't have references to
> any publications of his at my fingertips, but he'd be a name to look for.
> - Allan Metcalf

Truby was much maligned and even more misunderstood.  He had a "strong personality."  He
did NOT like the term voiceprint.  He explained to me how he did his "identification," but
he may never have published a description of his procedure.  If some police department had
a suspect and had a recording of the suspect saying something (e.g. "There's a bomb on the
second floor"), Truby would ask the police to get a line-up of 8 or 10 people saying the
phrase spoken by the suspect.  The suspect had to be in the line-up so that there would be
two iterations from the same vocal tract.  By examining all features of wide-band and
narrow-band spectrograms of all 9 or 11 "voiceprints," including nonlinguistic data such
as the higher formants, he claimed to be able to teach someone to achieve >99% accuracy in
identifying the two spectrogram of the phrase spoken by the same person.  It was a matter
of comparison rather than identity.  If the suspect tried to use a different "accent" or
otherwise tried to disguise his voice, he was still not able to change nonlinguistic
aspects of the speech signal that are dependent on the architecture of the vocal tract.

DMLance