摘要
文章研究并实现了汉语连续语音中的音节自动切分算法─—基于归并的音节切分自动机（merging－basedSyllabledeteCtionautomaton,简称MBSDA）算法．MBSDA算法利用了包括语音的短时能量、过零率和基音周期在内的多种特征参数，把特征参数高度相似的相邻帧（1帧或若干帧）的语音信号进行“归并（merging）”，形成“归并类似段（mergedsimilarsegment，简称MSS）”，它们被认定属于同一音节的相同状态．这些MSS经过一个包含若干状态的“音节切分自动机（syllabledetectionautomaton，简称SDA）”后，输出音节的切分点．每个确定的切分段中所包合音节个数的范围（rangeofsyllablenumber，简称RSN）也由MBSDA算法给出．
In this paper, an automatic syllable detection method namely merging-based syllable detectionautomaton (MBSDA) is studied and implemented. The MBSDA uses a variety of features including the frameenergy, the zero crossing rate and the fundamental frequency to merge similar consecutive frames (one or sever-al frames) into one merged similar segment (MSS). The frames in the same MSS are treated as frames of thesame state of a phonetic. These MSSs are passed into a syllable detection automaton (SDA) to give the syllabledetection results. In addition, the MBSDA gives the range of syllable number (RNS) of each definite detectionsegment.
出处
《软件学报》
EI
CSCD
北大核心
1999年第11期1212-1215,共4页
Journal of Software
关键词
语音识别
归并
音节切分自动机
连续汉语语音
Syllable detection, merging, syllable detection automaton, vowel feature segment, range of syllable number
