基于电话语料的维吾尔连续音素识别

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。

Telephone Speech Corpus-based Uyghur Continuous Phoneme Recognition

Combined the characteristics and semantic information of Uyghur language, and based on the large number of telephone speech corpus, an continuous phoneme recognition platform of Uyghur language is established, and the recognition algorithm of Uyghur continuous phoneme is implemented by using the HTK tool： first, according to the specific technical indicators, the recording and labeling of large-scale telephone speech corpus is done and with phoneme as the primitive and through training, the HMMmodel of each phoneme is achieved, then the input speech is recognized, and the different recognition rates of the acoustic model under different number of Gaussian mixtures are obtained. And the statistics and analysis on the recognition rates of 32 phonemes is done, thus laying a foundation for further improvement of the recognition rate.