论文部分内容阅读
本文提出了汉语语音导引特征的概念，讨论了语音导引特征在自动语音识别中用于导引匹配搜索的运用方式和重要作用；提出并设计了汉语塞音选择性特征自动萃取的小波变换方法和数字信号处理算法。本文方法和算法从声学信号处理和检测的角度，将汉语清辅音声波输入信号自动分为塞音子集ＢＤＧ：｛ｂ，ｄ，ｇ｝、塞音子集ＺＺＨＪＧＰＴｃＣＨＱＫ：｛ｚ，ｚｈ，ｊ，ｇ，ｐ，ｔ，ｃ，ｃｈ，ｑ，ｋ｝和擦音集ＦｓＳＨｈＸ：｛ｆ，ｓ，ｓｈ，ｘ，ｈ｝；对输入的合清辅音的音节，计算检测并输出汉语自动语音识别系统可以利用的清辅音类属标记ｂ．ｄ．ｇ、ＳＴＯＰ／ＢＤ和ｆ．ｓ．ｓｈ．ｘ．ｈ以及它们的音段起始时标；从声学信息计算检测的角度为汉语自动语音识别系统提供一种新的“从粗到细”的辅助匹配结构。算法可用性模拟实验采用实际语音的数据库数据，以手工标注信息作为自动检测分类正确与否的对比标准。对１２６７个汉语全音节中，总数９１３个待分类清辅音的初步分类结果表明：正确分类率分别为ｂ．ｄ．ｇ：９６．１％，ＳＴＯＰ／ＢＤ：９５．１％和ｆ．ｓ．ｓｈ．ｘ．ｈ：８９．０％，总体平均正确分类率为９３．６％。
In this paper, the concept of Chinese speech guidance feature is proposed. The application of speech guidance feature in automatic speech recognition to guide matching search and its important role are discussed. A wavelet transform method is proposed and designed to automatically extract Chinese Selective Selective feature And digital signal processing algorithms. In this paper, the method and algorithm of this paper are used to automatically divide the input signal of Chinese consonant consonant sound into BDG: {b, d, g} from the perspective of acoustic signal processing and detection, and the subsets of ZZHJGPTcCHQK: {z, zh, j, fsSHhX: {f, s, sh, x, h} for the syllable of the consonant consonant to be calculated and detected and output. The automatic speech recognition system of Chinese can be used Clear consonants belong to mark b. d. g, STOP / BD and f. s. sh. x. h and their start timestamps of the sound segments. A new “matching from coarse to fine” structure is provided for Chinese automatic speech recognition system from the perspective of acoustic information calculation and detection. The algorithm usability simulation experiment uses the database data of the actual speech to mark the information by hand as the contrast standard for automatically detecting whether the classification is correct or not. For the 1267 Chinese syllables, the preliminary classification results of 913 clear consonants to be classified show that the correct classification rates are b. d. g: 96.1%, STOP / BD: 95.1% and f. s. sh. x. h: 89.0%, the overall average correct classification rate was 93.6%.