基于BP神经网络的非特定人语音识别算法研究

当前非定特人语音识别算法大多只适于连续语音,且识别精度和速度均较低。为此,提出一种新的基于BP神经网络的非特定人语音识别算法,介绍了标准BP神经网络,针对其收敛速度慢的弊端,通过变化的自适应学习速率,令网络训练针对各种阶段自行设置学习速率值,利用变学习速率构建对应的改进BP神经网络模型,将改进的BP神经网络模型看作识别非特定语音的识别器,输入待识别语音,令累计预测残差达到最小,实现非特定人语音识别。将改进模型应用于非特定人语音识别中进行验证,结果表明所提算法识别率更高、识别速度更快,不仅适于连续语音的识别,也适于不连续语音的识别。

Research on the algorithm of speaker independent speech recognition based on BP neural network

At present, most of the speaker independent speech recognition algorithms are only suitable for continuous speech. To this end, proposes a new speech recognition algorithm based on BP neural network, introduces the standard BP neural network, aiming at the drawbacks of the slow convergence rate, the adaptive learning rate by changing the network for various stages of training, set up their own learning rate, using the improved BP neural network model to construct the corresponding variable learning rate the identification device, the improved BP neural network model as the input speech recognition, the speech to be recognized, the cumulative prediction error is minimized, realize speaker independent speech recognition. The improved model is applied to verify the speech recognition, results show that the proposed algorithm has higher identification rate and faster recognition speed, not only suitable for recognition of continuous speech, also suitable for recognition of continuous speech.