Integration of acoustic information in a large vocabulary word recognizer

摘要: This paper proposes a new way of using vector quantization for improving recognition performance 60,000 word vocabulary speaker-trained isolated recognizer phonemic Markov model approach to speech recognition. We show that we can effectively increase the codebook size by dividing feature into two vectors lower dimensionality, and then quantizing training each separately. For small size, integration results parameter provides significant improvement in as compared entire set together. Even 64, obtained when procedure are quite close those Gaussian distribution vectors.