Abstract: Three methods, the spectral subtraction, the autocorrelation and the linear filter for extraction of Chinese speech in noise background, are described and used to process vowels, words and sentences for signal to noise ratio in the range from -6dB to +6dB. The results indicate that the S/N ratio is increased by about 6 dB, although the intelligibility of the processed speech is not much improved. A comparison of several approaches for extracting fundamental frequency of the speech singal in such background is also made.