基于嵌入式平台的实用语音识别研究

摘要

从实用角度出发,提出了一种基于时域和频域的特征参数提取算法,利用感知哈希函数和线性预测参数得到特征参数.感知哈希函数是语音数据到数字摘要的一类单向映射,具有相同感知的语音数据表示唯一地映射为一段数字摘要,这样使得参数匹配起来更加准确容易.针对这种特征参数,采用隐马尔可夫模型对语音进行识别,在仿真实验的基础上,将该算法移植到树莓派嵌入式平台上,最后通过实际测试,表明该算法能达到较高的准确率,具有一定的可行性.%From the practical point of view,a new feature extraction algorithm from the time-frequency domain is proposed based on perceptual hash function and line spectrum pair. Perceptual hash function and the linear predic-tion parameters is used to achieve characteristic parameter. Perceptualhash function is a kind of one-way mapping data from voice data to digital voice.The same perception maps to a digital abstract uniquely,it makes the process of pattern match more accurate and easier. In view of the characteristic parameters,the hidden Markov model is used to identify the speech. According to the simulation experiment,the algorithm is transplanted to the embedded platform of raspberry pie. Finally,through the actual test,it shows that the algorithm can achieve high accuracy and be feasibility.