关键词: 支持向量机, Gaussian核, 语音识别, 低信噪比

Abstract: A low SNR speech recognition system for isolated words and non-specific speakers is constructed in this paper. Improved MFCC speech features （Mel-Frequency Discrete Wavelet Cepstral Coefficients, MFDWCs） are adopted and Support Vector Machines（SVM） is utilized as classification algorithm. The system obtains higher recognition accuracy, comparing to the results based on RBF Artificial Neural Network（ANN）. The experimental results show SVM possesses better robustness than RBF ANN, especially in low SNRs.

Key words: support vector machines, Gaussian kernel, speech recognition, low Signal Noise Ratio（SNR）