论文部分内容阅读
词图的高效生成算法是语音识别领域的重要研究课题。该文提出了一种基于词格的词图生成算法(trellis-based lattice-generating algorithm,TBLG),该算法在正向Viterbi解码生成的词格(trellis)基础上,进行反向A*解码生成词图。实验结果表明,与经典的解码器HDecode相比,TBLG生成的词图最优备选效果优于Hdecode。生成高密度词图时,TBLG在解码速度上远远快于HDecode。同时在相同识别率下,TBLG算法生成的词图更加简洁。
The efficient generation of word map algorithm is an important research topic in the field of speech recognition. This paper presents a trellis-based lattice-generating algorithm (TBLG) based on the trellis, which is based on the Trellis generated by the forward Viterbi decoding. The trellis-based lattice-generating algorithm Word map. Experimental results show that, compared with the classical decoder HDecode, the best candidate word graph generated by TBLG is better than Hdecode. When generating high-density word maps, TBLG is much faster in decoding speed than HDecode. At the same time, under the same recognition rate, the word map generated by TBLG algorithm is more concise.