Large-vocabulary isolated word recognition with fast coarse time alignment

摘要: An isolated word recognition system for large vocabularies (1000 words and up) with an average rate of more than 98 per cent is presented. Each utterance characterized by a sequence feature vectors which are obtained autocorrelation analysis. The resulting coefficients quantized in such way, that entire vector can be stored single data word. A distance measure adapted to this representation has been developed. classification performed hierarchically two steps. In the preselection stage, divided into 16 segments hardware employed coarse nonlinear mapping. short ranked list candidates processed following final classifier performs time alignment fully resolved patterns using Dynamic Programming. Thus response high performance achieved. Without full use parallelism overall vocabulary less one second on signal processor.