论文部分内容阅读
在大词表孤立词语音识别中,Viterbi搜索是时间消耗的主要因素。为改善基线系统性能,根据汉语孤立词识别的特点,提出了一种基于音节切分的束搜索算法,在音节层和词条层进行剪枝。该算法不增加内存开销。实验结果表明:在词表规模为10 000时,该算法以0.23%的识别率下降率为代价,将Viterbi搜索的时间消耗降低为基线系统的26.73%;相对于小词表,该算法在大词表情况下对系统性能的改善尤为明显。
Viterbi search is a major contributor to time consumption in large-vocabulary isolated word speech recognition. In order to improve the performance of baseline system, a syllable search algorithm based on syllable segmentation is proposed based on the characteristics of isolated Chinese word recognition. The algorithm does not increase the memory overhead. The experimental results show that the proposed algorithm reduces the time consumption of Viterbi search to 26.73% of the baseline system at the rate of 0.23% recognition rate drop at the scale of 10 000 words. Compared with the small word list, The improvement of system performance is particularly evident in the case of the vocabulary.