基于深度机器学习的英语语音识别研究

Research on English speech recognition based on deep machine learning

为了提高英语语音识别性能,采用深度机器学习算法用于语音识别。设置合理的帧采样频率获得英语语音信号,然后通过线性预测系数(LPC)获得表示该语音信号的特征参数,并生成语音特征向量,接着采用卷积神经网络(CNN)算法对语音特征进行训练,设置卷积核尺寸和语音识别准确率阈值,通过多次运算获得稳定的英语语音识别模型。差异化设置帧采样频率和卷积核尺寸,验证不同参数下的LPC+CNN算法的语音识别准确率性能,获得合适的帧采样频率和卷积核尺寸,通过和常用语音识别算法对比,经过了LPC优化的CNN算法英语语音识别准确率更高,且收敛快。

In order to improve the performance of English speech recognition,deep machine learning algorithm was used for speech recognition.Set a reasonable frame sampling frequency to get the English speech signal,and then obtained the characteristic parameters of the speech signal through the linear prediction coefficient(LPC),and generated the speech feature vector.Then used the convolution neural network(CNN)algorithm to train the speech features,set the convolution core size and speech recognition accuracy threshold,and obtained a stable English speech recognition results through multiple operations.The frame sampling frequency and convolution core size were set differently to verify the speech recognition accuracy performance of LPC and CNN algorithm under different parameters,and the appropriate frame sampling frequency and convolution core size were obtained.Compared with common speech recognition algorithms,the CNN algorithm optimized by LPC had higher accuracy and faster convergence.