Abstract: Similarity is measured between incoming speech and a plurality of candidate strings of filler and keyword templates to test the hypothesis of a keyword being present against the hypothesis of a keyword not being present in the input speech. As alternatives to a candidate string containing a keyword template, other candidate strings are assembled with filler templates of short speech sounds that may be similar to a keyword. A keyword is "recognized" only when an optimal candidate string containing the corresponding keyword template is determined to match the input speech more closely than all other candidate strings. A concatenation penalty is added to the partial string score each time a new template is added to a candidate string, in order to bias the score in favor of a candidate string containing a longer, keyword template.

Abstract: An automatic speech recognition system has a multi-mode training capability using a set of previously stored templates of a limited number of predetermined seed words to train the templates for a vocabulary of words. The training speech samples each includes a vocabulary word juxtaposed with a seed word. An averager module maintains an active average template for each of the word units of the training speech samples including the seed word units, and the active average templates are used to continuously update the seed template set as they are used in the training speech samples. The preferred training procedure employs training phrases each having a vocabulary word embedded between two seed words, and two seed template sets are used in succession, the first being composed of single-digit words, and the second composed of carrier words.

Abstract: An apparatus for automatic speech recognition includes a plurality of processors, including template processors in which there are stored templates representative of both speech and non-speech sounds. Incoming sounds are continuously converted into digital signals in respective frames representative of speech and non-speech sounds, respectively. Sequences of such frames are compared with both the speech and non-speech templates to determine the closest matches. Endpoints of respective speech utterances are determined in response to the detection of respective non-speech-speech-non-speech sequences, whereupon such speech utterances are processed to recognize the same.

Abstract: The invention provides a method and system for speaker enrollment, as well as for speaker recognition. Speaker enrollment creates for each candidate speaker a set of short acoustic segments, or templates, of phonemic duration. An equal number of templates is derived from every candidate speaker's training utterance. A speaker's template set serves as a model for that speaker. Recognition is accomplished by employing a continuous speech recognition (CSR) system to match the recognition utterance with each speaker's template set in turn. The system selects the speaker whose templates match the recognition utterance most closely, that is, the speaker whose CSR match score is lowest. The method of the invention incorporates the entire training utterance in each speaker model, and explains the entire test utterance. The method of the invention models individual short segments of the speech utterances as well as their long-term statistics.