关键词: 语音识别, 分形维, 语音去噪, 音节分割

Abstract: In order to enhance the effect of existing wavelet denoising and determine beginning-ending points of each syllable in continuous speech，the thesis proposes an algorithm based on fractal theory.The algorithm first uses dynamic threshold algorithm which combines fractal dimension with wavelet transform to denoise the speech signal，it can extract pure speech as far as possible;on this basis，the paper designs the algorithm which is based on the mean of fractal dimension trajectory to carry out syllable segmentation.The experimental results show that the algorithms not only achieves speech denoising and syllable segmentation but also has good robustness.In the case of low SNR，the algorithm is still able to maintain high accuracy rate.It has better prospect in speech recognition field.

Key words: speech recognition, fractal dimension, speech denoising, syllable segmentation