Pitch detection based on time-frequency analysis

A method for pitch detection from speech signals was presented. The method employs the facts that peaks on the spectrogram of a voiced speech sound appear periodically in the range of low frequencies. In order to find robustly the pitch value for every frame, a matching algorithm based on the ratio between the above peak values was introduced. This algorithm can eliminate the interferer, which come from the peaks adjacent to the pitch and give correct pitch value. Compared with the most known algorithms, it has a good performance. It is more robust against noise, as well as more accurate in measurement precision.