CIRCUIT FOR SPEECH RECOGNITION

摘要

To recognise a speech signal, it is necessary to carry out an analysis of the speech signal with the aim of extracting characteristic features. The extracted features are represented by so-called spectral feature vectors which are compared with the reference feature vectors stored for the speech signal to be recognised. The reference feature vectors are determined during a training phase in which a speech signal is recorded several times. The result of the recognition essentially depends on the quality of the spectral feature vectors or reference feature vectors. It is therefore proposed to provide recursive high-pass filtering of the spectral feature vectors. This reduces the influence of interference variables on the result of the recognition and provides for a high degree of speaker-independence of the recognition. This makes it possible to use the circuit arrangement for speech recognition even in systems which presuppose speaker-independent speech recognition.