Method and system for generating advanced feature discrimination vectors for use in speech recognition
A method of renormalizing high-resolution oscillator peaks, extracted from windowed samples of an audio signal, is disclosed. Feature vectors are generated for which variations in both fundamental frequency and time duration of speech are substantially mitigated. The feature vectors may be aligned within a common coordinate space, free of those variations in frequency and time duration that occurs between speakers, and even over speech by a single speaker, to facilitate a simple and accurate determination of matches between those AFDVs generated from a sample of the audio signal and corpus AFDVs generated for known speech at the phoneme and sub-phoneme level. The renormalized feature vectors can be combined with traditional feature vectors such as MFCCs, or they can be used exclusively to identify voiced, semi-voiced and unvoiced sounds.
1. A method of generating advanced feature discrimination vectors (AFDVs) representing sounds forming at least part of an input audio signal, the method comprising:
taking a plurality of samples of the input audio signal, the plurality of samples being a portion of the input audio signal as it evolves over a window of predetermined time;
for each portion of the input audio signal taken:
performing a signal analysis on the portion to extract one or more high resolution oscillator peaks therefrom, the extracted oscillator peaks forming a spectral representation of the portion;
renormalizing the extracted oscillator peaks to eliminate variations in a fundamental frequency and a time duration for each portion occurring over the window;
normalizing a power of the renormalized extracted oscillator peaks;
forming the renormalized and power normalized extracted oscillator peaks into an AFDV for the sample;
collecting a set of audio samples from a specific individual to form a model of a voice of the individual comprising a database in an AFDV format; and
creating an audio fingerprint in the AFDV format of an individual comprising an aggregation of the collected audio samples.
2. The method of claim 1 , where the creating comprises transforming the aggregation of collected audio samples into at least one of a common frequency scale and a common time scale resulting in a comparison coordinate representation of the audio fingerprint.
3. The method of claim 2 , wherein the comparison coordinate representation is stored in a database.
4. The method of claim 3 , wherein the database is comprised of a harmonic portion and a non-harmonic portion.
5. The method of claim 3 , wherein the database is comprised of at least one of a phoneme and a category of phoneme.
6. The method of claim 5 , further comprising capturing an utterance from an unidentified person, and identifying one or more types of phonemes uttered by the person via a speech recognition algorithm for comparison to the database.
7. The method of claim 3 , further comprising receiving an unknown audio input comprising a speaker of interest and an interfering sound, matching at least one of the speaker and the interfering sound to the database, and at least one of preferentially enhancing and preferentially suppressing one of the speaker of interest and an interfering sound.
8. The method of claim 2 , wherein the comparison coordinate representation comprises a predetermined number of a plurality of excitation periods.
9. The method of claim 1 , wherein renormalizing the extracted oscillator peaks further comprises estimating at least one of a fundamental frequency and a pitch of the audio sample.
10. The method of claim 1 , further comprising storing a plurality of parameters defining the renormalized and power normalized extracted oscillator peaks.