RECENT ADVANCES IN SPEECH RECOGNITION TECHNOLOGY AT NTT LABORATORIES

This paper introduces recent research activities on speech recognition, ranging from acoustic processing to linguistic processing, at NTT (Nippon Telegraph and Telephone Corporation) Laboratories. These include the proposal of DELTA-LSP parameters, hierarchical DELTA-cepstral parameters, a new method of utilizing pitch information, automatic speaker adaptation techniques, robust HMM phoneme models, new training algorithms for neural networks, linguistic processing that uses syntactic and semantic knowledge, implementation of prototype continuous speech recognition systems, and an efficient text-independent speaker recognition algorithm.