Integration of speech and language processing in Chinese continuous speech recognition
【摘要】：正We propose a method of Chinese continuous speech recognition, which combines speech and language processing with frame-synchronous parsing algorithm. The context free grammar (CFG) and top down sentence analyzer are applied in the involved language processing, and the one pass Viterbi algorithm, based on the finite state automation (FSA) is used in the involved speech processing. The experiments, which use 60 phoneme and 8 tone HMMs, show that the average sentence recognition rate for 1070 utterance of ten speakers is 94.4%. Compared with conventional hierarchical system based on word spotting and lattice parsing algorithm, an improvement of 8% is obtained in the same task of perplexity 27.3.