无线电工程
無線電工程
무선전공정
RADIO ENGINEERING OF CHINA
2015年
8期
35-38,86
,共5页
非特定人%语音识别%隐形马尔科夫模型%DSP实现
非特定人%語音識彆%隱形馬爾科伕模型%DSP實現
비특정인%어음식별%은형마이과부모형%DSP실현
speech-independent%speech recognition%DHMM%DSP implementation
针对非特定人孤立词的快速高精度语音识别问题，在分析语音信号特征和离散隐形马尔科夫模型（ DHMM）原理的基础上，提出了一种基于DHMM快速语音识别算法。算法首先对语音信号进行预处理和端点检测以提取有用信号，进而完成语音特征参数提取和矢量量化，最后采用DHMM模型利用Viterbi搜索算法实现语音识别；在训练阶段需要完成码本设计和DHMM模型参数的生成；采用ADI公司Blackfin系列BF533芯片对识别算法进行DSP实现。理论分析和仿真结果表明，在日常环境条件下算法的识别正确率达97?5％，且DSP运行时间小于10 ms。
針對非特定人孤立詞的快速高精度語音識彆問題，在分析語音信號特徵和離散隱形馬爾科伕模型（ DHMM）原理的基礎上，提齣瞭一種基于DHMM快速語音識彆算法。算法首先對語音信號進行預處理和耑點檢測以提取有用信號，進而完成語音特徵參數提取和矢量量化，最後採用DHMM模型利用Viterbi搜索算法實現語音識彆；在訓練階段需要完成碼本設計和DHMM模型參數的生成；採用ADI公司Blackfin繫列BF533芯片對識彆算法進行DSP實現。理論分析和倣真結果錶明，在日常環境條件下算法的識彆正確率達97?5％，且DSP運行時間小于10 ms。
침대비특정인고립사적쾌속고정도어음식별문제，재분석어음신호특정화리산은형마이과부모형（ DHMM）원리적기출상，제출료일충기우DHMM쾌속어음식별산법。산법수선대어음신호진행예처리화단점검측이제취유용신호，진이완성어음특정삼수제취화시량양화，최후채용DHMM모형이용Viterbi수색산법실현어음식별；재훈련계단수요완성마본설계화DHMM모형삼수적생성；채용ADI공사Blackfin계렬BF533심편대식별산법진행DSP실현。이론분석화방진결과표명，재일상배경조건하산법적식별정학솔체97?5％，차DSP운행시간소우10 ms。
The paper focuses on the fast and high?accuracy speech recognition of speaker?independent isolated words.Based on an analysis of the characteristics of speech signal and the principles of Discrete Hidden Markov Model( DHMM) ,a fast speech recognition algorithm using DHMM is presented.Firstly,the useful speech signals are separated from the background noise by using preprocessing and endpoint detection.Secondly,speech feature parameters are extracted and vector quantized.Finally,speech recognition is realized by using Viterbi search algorithm based on DHMM parameters.During the training phase,the codebook design and generation of DHMM parameters are completed.The algorithm is DSP implemented by using the BF533 chip of Analog Devices Inc.Theoretical analysis and simulations show that the recognition accuracy is as high as 97.5% under daily environment conditions and its DSP runtime is less than 10 ms.
