People use voice identification everyday. For instance, we sometimes recognize people based on their voices,
when we pick up the phone. Also, dictation software is available so a person just speaks to type.
Current Uses of the Voice in Biometrics:
Voice recognition is considered the least intrusive, which makes it popular.
It previously had some problems since the voice changes with a human's mood and illness (ie-stuffy noses from colds) and
background noise poses some problems.
However, new technology is evading that problem and becoming more accurate. Though
it is more accurate, it is still used for mainly speaker verification and/or in conjunction with another form of biometrics.
For instance, QVoice sells Who is It! which combines voice recognition
with face and fingerprint recognition. Star Trek version is exactly like Who is It! except it comes with fun Star Trek
themes and sounds.
Systems plot frequencies that the voice produce every 1/100th of a second to form a "waterfall." The "waterfall" is
a three-dimensional image that is a map of the voice for one second. A line graph of curved and straight arrows note
the transitions between phonemes, the smallest units of speech. A database stores this information until
it is needed. Voiceprint analysis is used for both verification and identification.
The information can be stored on a local database. However, a different method requires using a telephone. A user calls
a database system using the telephones. After he/she is verified using voiceprint analysis, the system will give the user
a "volatile password." The password expires in 5 minutes. The Citadel system by
iNTELiTRAK uses this method.
- Developed by IBM Research. Unlike the other methods, VoicePrint determines
resonance of the cavities in the throat with vocal-tract geometry from a voice sample. This method can not be fooled by
different languages, different voice samples, etc. It is also not fooled by stuffy noses, since that effects the
nasal cavities, not the throat ones.