MAMDF Pitch Estimation Research And Its Application In Waveform Interpolation Speech Coding

Pitch period is an important parameter in speech coding, speech recognition,speech synthesis and other fields in speech signal process. The precision androbustness of pitch detection play an important role in these fields. Present pitchestimation algorithms have a disadvantage of giving rise to double or halve pitchperiod and other error. Especially in noisy speech signal, their performance decreasesobviously. With the development of communication technology, research in speechcoding has a great significance. At present, waveform interpolation speech codingalgorithm has become a hot in low rate speech coding.As the pitch estimation algorithm average magnitude difference function (AMDF)has very low complexity and fine real-time performance, many methods in pitchestimation are proposed based on AMDF. However, these algorithms are still not verysatisfying and are susceptible to noisy signal. The thesis proposes a new pitchestimation algorithm MAMDF (Modified Average Magnitude Difference Function)which can not only strengthen periodic character of voice speech but also reduce theinfluence of noisy signal. Experiments implemented in CSTR database show that thenew algorithm has a remarkable performance improvement in clean speechenvironment and noisy speech environment in comparison to traditional algorithms.In waveform interpolation speech coding which has a low bit rate, pitchestimation is a significant parameter that can decide the length of characteristicwaveform which is crucial to speech coding quality. The thesis applies the proposedpitch estimation algorithm to waveform interpolation speech coding to determinepitch period and realizes pitch interpolation. According to interpolated pitch periodthe thesis obtains characteristic waveform and aligned the neighboring characteristicwaveform in different situation. Finally the thesis calculates the power ofcharacteristic waveform. Therefore, the needed parameters are acquired in speechcoding. After appling the proposed pitch estimation algorithm, more accuratecharacteristic waveform and parameters in waveform interpolation speech coding isobtained which is beneficial to speech quality.