Speech recognition on MPEG/Audio encoded files

Abstract

A technique to perform speech recognition directly from audio files encoded using the MPEG/Audio coding standard is described. The technique works in the compressed domain and does not require the MPEG/Audio file to be decompressed. Only the encoded subband samples are extracted and processed for training and recognition. The underlying speech recognition engine used is based on the hidden Markov model. The technique is applicable to layers I and II of MPEG/Audio, and training under one layer can be used to recognize the other. Results based on the recognition of a speaker-dependent, small vocabulary, and continuously spoken sentences shows accuracy as high as 99% using this technique.