论文部分内容阅读
结合维吾尔语的语音特征和语义信息,在大量电话语音语料库的基础上,以建立维吾尔语连续音素识别平台为目标,通过构建隐马尔科夫模型工具HTK(Hidden Markov Model Toolkit)工具实现了维吾尔语连续音素识别算法:首先根据具体技术指标完成了较大规模电话语音语料库的录制和标注工作;确定音素为基元,通过训练获得了每个音素的HMM(Hidden Markov Model)声学模型,随后对输入的语音进行识别,声学模型在不同的高斯混合数目下,得出了识别结果;统计了32个音素的识别率并对它进行分析,为了进一步提高识别率奠定了基础。
Combining with Uyghur phonetic features and semantic information, based on a large number of telephone phonetic corpus and establishing a Uyghur continuous phonetic recognition platform, the Uyghur language is implemented by constructing Hidden Markov Model Toolkit (HTK) Continuous phoneme recognition algorithm: Firstly, according to specific technical indicators, the recording and annotation of large phone phonetic corpus was completed. The phoneme was determined as the primitive, and the HMM (Hidden Markov Model) acoustic model of each phoneme was obtained through training. Of the voice recognition, the acoustic model in different Gaussian mixture number, the recognition results; statistics of the 32 phoneme recognition rate and its analysis, in order to further improve the recognition rate of the foundation.