Texas Instruments Corporation

Enrico L Bocchieri

George R Doddington

Michael L McMahan

William M Fisher

Recognition of sound units is improved by comparing frame-pair feature vectors which helps compensate for context variations in the pronunciation of sound units. A plurality of reference frames are stored of reference feature vectors representing reference words. A linear predictive coder (10) generates a plurality of spectral feature vectors for each frame of the speech signals. A filter bank system (12) transforms the spectral feature vectors to filter bank representations. A principal feature vector transformer (14) transforms the filter bank representations to an identity matrix of transformed input feature vectors. A concatenate frame system (16) concatenates the input feature vectors of adjacent frames to form the feature vector of a frame-pair. A transformer (18) and a comparator (20) compute the likelihood that each input feature vector for a frame-pair was produced by each reference frame. This computation is performed individually and independently for each reference frame-pairs. A dynamic time warper (22) constructs an optimum time path through the input speech signals for each of the computed likelihoods. A high level decision logic (24) recognizes the input speech signals as one of the reference words in response to the computed likelihoods and the optimum time paths.

Speaker independent speech recognition method and system

Speech recognition is improved using prestored words and prestored sentences which provide grammar or context rules. Multiple word hypotheses and plural sentence hypotheses are used. A speech recognition system uses a word hypothesizer to generate hypothesized words for a sentence recognizer. The sentence recognizer receives information containing the hypothesized word and its beginning and end points. The sentence recognizer compares each incoming hypothesis with words allowed by a grammar, and generates a partial or complete sentence hypothesis for each state in which the hypothesis is allowed by the grammar. A dynamic programming technique using backpointer is used to retain all currently valid partial and complete sentence hypotheses. When a complete sentence satisfies suitable criteria, it is output as a recognized sentence. Gaps and overlaps between hypothesized words are used in calculation of a sentence error for each partial and complete sentence hypothesis.

Growing community of inventors

Dallas, TX, United States of America

Enrico L Bocchieri