As a key technique, the pitch detection has been a hot spot in the field of speech processing. The following three disadvantages of power spectrum reprocessing (PSR) method for pitch detection are found: half pitch error and double pitch error with transition sound; easy distortion in noise speech; the complex method of judging voiceless and voiced speech. A series of improvement methods are proposed: nonlinear processing at timedomain; windowed filtering at frequencydomain; simplification method of judging voiceless and voiced speech. The experimental results based on MATLAT show that the improved method detects pitch trajectory more clearly and accurately than the AMDF method and the PSR method.