Phase Variance Probability Signatures for Classification of Voiced Speech

摘要

A technique is presented to extract a new feature dubbed phase variance probability signature from the spectral phase of speech. The speech is divided into glottal pulses and spectral phase is computed for each glottal pulse. The variance of the phase across each glottal pulse is used to construct a signature for the utterance. By averaging several signatures from utterances of the same sound, the target probability signature is computed. By using a comparison function and the target probability signatures of several known sounds, unknown utterances can be classified.