摘要
以VC++6.0为开发平台,实现一个基于隐马尔可夫模型(Hidden Markov Model,简称HMM)非特定人的安多藏语孤立词语音识别系统。对有声段语音进行MFCC参数的提取,对提取后的MFCC参数进行矢量量化后训练HMM模型,形成特征模板库,最后进行识别。根据安多藏语的特点,改进端点检测的方法,提高了孤立词语音信号检测的准确性,并进一步提高了识别率。
Based on the VC + +6.0, this article realize an isolated human non-specific word speech recognition system with a hidden Markov model-based (Hidden Markov Model), use the real-time endpoint detection method, extract and train MFCC paragraph parameter for the audio voice in order to reduce the voice frames, and finally recognize the isolated words. According to the characteristics of Amdo Tibetan, improved endpoint detection methods to improve the isolated word speech signal detection accuracy, and further improve the recognition rate.
出处
《软件导刊》
2010年第7期173-175,共3页
Software Guide
基金
西北民族大学研究生科研创新基金项目(YCX09070)
关键词
端点检测
MFCC参数
语音识别
Endpoint Detection
MFCC Parameters
Speech Recognition