Speaker Verification

Speaker verification is used to authenticate who someone says he or she is. The first step is based upon training.

A user of speaker verification says a phrase 1..N times. The N can vary as the speaker verification algorithm extracts information from the training sample.

Once enough information has been extracted, a voiceprint is generated and saved away. The next time that a user attempts authentication, the phrase spoken by the user is compared against the stored voiceprint.

How closely the spoken phrase must match the voiceprint before a positive
response is returned can be adjusted.

Voiceprints can be anywhere from 6K to 32K in size. The size is important when thinking about applying the technology to smart cards.

The advantages to speaker verification are many:

	User experience is improved. Instead of having to press buttons
	to play DTMF digits, the user simply says a phrase.

	The verification technology can be applied against multiple
	applications. Speaker verification is a very good candidate for a
	web service.

	Call processing time is reduced. It is much quicker to say a three to
	five second phrase than try to press ten or fifteen digits.
	Not only is the process accomplished in a shorter time span, the
	process is also less error prone.

	Speaker verification is also more secured than maintaining a pin that
	is punched into a system. If someone has access to the pin, they
	can access the system. Even if someone has a recording of an
	individual’s voice, they still not can access a system based upon
	speaker verification.