基于改进RBF神经网络的ADPCM语音编码算法研究

Improvement in ADPCM Speech Coding Algorithm Based on RBF Neural Network

以G.721ADPCM语音编码算法为研究对象,在语音编码的预测中引入神经网络模型来克服传统线性滤波方法中存在的不足,研究了基于RBF神经网络的ADPCM语音编码系统的结构。通过k均值聚类算法来确定RBF神经网络的中心和宽度,用最小二乘法确定RBF网络权值的方法改进了ADPCM语音编码算法。实验证明,其平均信噪比较原ADPCM编码算法有1.2dB的提高。

The speech coding algorithm based on G.721 ADPCM is researched. To overcome the shortage of linear prediction, the module of ANN is used to replace the traditional LP technology, and the structure of ADPCM speech coding based on ANN is discussed. With this method, the ADPCM speech coding with RBF neural network is improved. The k-means clustering algorithm is employed to determine the RBF neural network＇s centre and spread, and the OLS method is used to determine the weight. The experiment results indicate that the speech SNR based on RBF has increased 1.2 dB compared with the G.721 by ITU.