Method and system for bio-metric voice print authentication

A system for generating a biometric voice print, comprising:

• a voice processor for receiving a spoken utterance and at least one repetition of the spoken utterance from a user;
• a biometric voice analyzer for:
• calculating one or more vocal tract shapes from the spoken utterance and the at least one repetition, and
• calculating a vocal tract configuration difference between the one or more vocal tract shapes based on a varying pronunciation of the spoken utterance and the at least one repetition;
• wherein the voice processor:
• generates one or more feature vectors from the one or more vocalized frames by:
• segmenting a spoken utterance into one or more vocalized frames;
• performing a perceptual filter bank analysis on the one or more vocalized frames by estimating speech energy and noise energy in one or more frequency bands along a Bark frequency scale, and suppressing background noise by discarding filterbanks having a ratio of speech energy to noise energy that do not exceed a threshold of vocalization;
• calculating Linear Prediction Coefficients (LPC) from the perceptual filter bank analysis;
• converting the LPC's to Line Spectral Pair coefficients (LSP's);
• calculating formants and anti-formants from the LPS's;
• and
• creating a feature vector from the formants and anti-formants;
• calculates a feature matrix from the one or more feature vectors;
• and
• normalizes the feature matrix over the one or more vocalized frames.

A method for voice authentication, comprising:

• determining, by a voice processor, two or more vocal tract shapes from one or more received spoken utterances from a user;
• evaluating, by the voice processor, a vocal tract difference between the two or more vocal tract shapes;
• comparing, by the voice processor, the vocal tract difference against a stored representation of a reference vocal tract shape of the user's voice;
• determining, by the voice processor, whether the vocal tract configuration difference is indicative of natural changes to the reference vocal tract shape, wherein natural changes are variations in the vocal tract configuration which can be physically articulated by the user;
• determining a source of the spoken utterance, wherein the source is one of the user speaking the spoken utterance into a microphone or a device playing back a recording of the spoken utterance into the microphone;
• and
• granting access if the source is the user, and not granting access if the source is the device;
• wherein the determining the source further comprises:
• identifying whether an acoustic signal representing the spoken utterance is characteristic of a waveform produced by a digital recording device,
• wherein the identifying includes recognizing a spectral tilt imparted by the digital recording device.