南疆维吾尔语语音识别研究与实现

Research and Implementation of Speech Recognition about Uyghur in Southern Xinjiang

该研究以南疆维吾尔语语音为研究对象,分析了维吾尔语语音的发音特点,采集语音后对其进行预处理,提取Mel倒谱系数作为特征参数,实现了基于动态时间规整算法的南疆维吾尔语非特定人语音的识别。该识别方法在MATLAB上进行了仿真实验,测试样本420个,识别正确394个,识别正确率为93.8%,实验结果表明该识别方法对南疆维吾尔语语音识别效果较好,这为进一步研究大词汇量连续语音的维吾尔语语音识别提供了借鉴意义。

Using the speech of Uyghur in southern Xinjiang as object of study, we analyze the characteristics of Uyghur speech pronun- ciation, pretreatment it after collecting, extract Mel cepstrum coefficients as the feature parameters, and realize the speech recognition based on the dynamic time warping algorithm in Uyghur non - specific language. The identification method using experimental tool is MATLAB, testing a sample of 420 among which 394 is correctly identified the recognition rate is 93.8%. The experimental results show that the effect of recognition method about the Southern Xinjiang Uyghur speech recognition is very good, and it provides the basis for the further research on large - vocabulary continuous speech Uighur speech recognition.