摘要
为了解决低速声码器合成语音的偶发性嘶哑或变调问题,对参数提取进行改善,采用有监督学习的Fisher判决法,利用多个特征值组成的特征向量为判据;基音周期平滑的准确度在利用了更准确的清浊音信息后大有提高。测试结果表明:该算法能够大大降低清浊音误判率,减少严重基音周期错误数;应用该算法的SELP(sinuous excitationlinear prediction)2.4 kb/s的PESQ-MOS分优于2.4 kb/s的MELPe(mixed excitation linear prediction)和AMBE+(advanced multi-band excitation)算法,DRT(diagnosticrhythm test)分数达95%,具有良好的可懂度和自然度。
Many kinds of 2. 4 kb/s low bit rate vocoders have occasionally hoarseness or out-of-tone speech. Hence voiced-unvoiced classification method is improved using several parameters based on Fisher method. The pitch track precision is then improved by more precise voiced-unvoiced information. Tests results show that the Fisher classification method greatly reduces the voiced-unvoiced classification error rate and number of severe half or double pitch errors. The improved 2.4 kb/s SELP （sinuous excitation linear prediction） vocoder then get a higher PESQ- MOS score, even outperforming the US government＇s MELPe and DVSI＇s AMBE ＋ algorithm at the same rate. Additionally, the improved 2. 4 kb/s SELP vocoder has diagnostic rhythm test （DRT） scores of up to 95%, which produces excellent natural and intelligible speech.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第7期1119-1122,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(60572081)
关键词
语音编码
清浊音判决
MELPe算法
AMBE+算法
speech coding
voiced-unvoiced classification
MELPe （mixed excitation linear prediction） algorithm
AMBE ＋ （advanced multi-band excitation plus） algorithm
