Pitch Detection Algorithm And Its Application In Speech Synthesis

It’s extremely important to detect pitch period accurately and effectively in all fields of speech signal processing(such as speech recognition, speaker recognition, speech analysis, speech synthesis, low bit rate speech coding, diagnostic system for pronunciation diseases, language instruction for the hearing handicapped, etc.), due to its direct influence on the performance of the whole system. However, there is still no effective pitch detection algorithm that can be applied to any situation and any speaker because the detection of pitch period is affected by various factors. To this end, through researching traditional pitch detection algorithms and other classical modified algorithms in-depth, we strive to find a pitch detection algorithm with relatively better accuracy and robustness in this paper.As a pitch estimation algorithm average magnitude difference function (AMDF) has very low complexity and short calculation time, many methods in pitch detection are proposed based on AMDF. But the performances of these improved methods are still not satisfactory. This paper presents two new pitch estimation modified algorithms:modified AMDF pitch detection algorithm based on DWT (DW-AMDF) and modified AMDF pitch detection algorithm based on polynomial fitting (PF-AMDF). The two proposed methods respectively use DWT and polynomial fitting to eliminate the falling trend of AMDF so that the problem of false detection, which emerges as AMDF is susceptible to noise, can be fundamentally solved. Lots of experiment shows that the performance of the proposed DW-AMDF and PF-AMDF outperforms AMDF and other AMDF-based modified algorithms.In the time-domain pitch synchronous overlap-add method, pitch period is a vital parameter that can decide the correctness of pitch pulse labeling which is crucial to speech synthesis quality. After applying the proposed pitch estimation algorithm, pitch pulse labeling can be more accurate, which is beneficial to the naturalness of synthesized speech.