Abstract: With the random signal caused by breathing airflow in snore, the double linear prediction residual methods are used to get the snore source signal, and then by calculating its cepstrum, the pitch of snore can be extracted. The snores are classified into several types based on the periodicity of the snore waveforms. The extracted pitches and the statistics of their distribution are compared with the severity of sleep apnea hypopnea syndrome (SAHS). The snores from the 93 subjects are analyzed in the experiments. The results show that with the aggravation of SAHS, the quasi-periodicity of snoring decreases and the jitter of the pitch contour increases. The pitch and the features of Mel-Frequency Cepstral Coefficients (MFCC) are used to judge the severity of SAHS, and the accuracy is 85.5%. The statistical characteristics of the snore pitch correlation could be used as one of the parameters to judge the severity of SAHS. This study would play a positive role in design and implementation of the portable SAHS diagnosis instruments.