Abstract: Methods and apparatus for speech recognition based on a hidden Markov model are disclosed. A disclosed method of speech recognition is based on a hidden Markov model in which words to be recognized are modeled as chains of states and trained using predefined speech data material. Known vocabulary is divided into first and second partial vocabularies where the first partial vocabulary is trained and transcribed using a whole word model and the second partial vocabulary is trained and transcribed using a phoneme-based model in order to obtain a mixed hidden Markov model. The transcriptions from the two models are stored in a single pronunciation lexicon and the mixed hidden Markov model stored in a singe search space. Apparatus are disclosed that also employ a hidden Markov model.

Abstract: The invention expands a feature extraction unit of a type known in the art with an adaptive channel compensation. The filter coefficients for a high-pass filter for eliminating the chronologically sliding average are set time-dependently with the adaptive channel compensation. As a result, a faster channel adaptation, and a better error compensation in the speech feature extraction are achieved, leading to a more reliable speech recognition. The parameters for the individual function blocks of the feature extraction unit are determined and stored in advance for specific speech transmission channels such as in mobile radio telephone, ISDN and analog speech transmission. The speech recognition with the assistance of the feature extraction unit can subsequently then ensue in channel-associated fashion with the stored parameters values.