METHOD AND SYSTEM OF AUTOMATIC LANGUAGE RECOGNITION USING A-POSTERIORI TRUST POINTS

摘要

Systems, articles and methods include automatic speech recognition techniques using a posteriori confidence scores. The method includes receiving a plurality of frames of audio sequence with human-language audio data, wherein individual frames of the plurality of frames each have a frame term that is a segment of human speech, wherein one or more frame terms form a single phoneme; Receiving frame values, wherein individual frame values are associated with a different frame term and used to form acoustic scores for a phoneme; Determining which frame terms are associated with a candidate utterance of a language model; Generating a plurality of probabilities, each probability being an individual acoustic model statement, wherein one of the frame terms is associated with the utterance, each frame term being provided separately to an acoustic model having multiple acoustic model statements; and Determining a confidence score of the utterance to be used to determine whether the utterance is likely to match the actual speech recorded on the frames, comprising combining the probabilities of the frame terms associated with the utterance.