Research On Chinese Speech Synthesis Based On Pitch Synchronization Superposition Method

Speech synthesis is an important branch of speech signal processing, after a long period of development, speech synthesis method is increasing. Among them, the pitch synchronous overlap algorithm (Pitch Synchronous OverLap-and-Add) since being raised has a relatively wide range of applications. Different from Simple waveform concatenation, the algorithm can be adjusted rhythm, such as the fundamental frequency, duration, etc. Such synthetic speech has improved the natural degrees.Currently, What the pitch synchronous overlap algorithms used are some natural voice speech database. In speech synthesis, it allows the synthesis of the results retaining some features of the original pronunciation, and makes the synthetic speech no longer stiff, with individual tone characteristics. However, this speech database contains a large number of natural speech, the capacity of the speech database is large. Meanwhile, In the speech synthesis, the accuracy of pitch period estimation is very important, affecting the synthesis results. The traditional methods to estimate are the autocorrelation function method and the average magnitude difference function method, they will be some false positives in the judgment of the pitch. In this paper, Against the problem of voice storage capacity, we propose a consonant vowel synthesis, thus reducing the storage capacity. In terms of pitch period estimate, we combine the auto-correlation function method with the average magnitude difference function method to make the estimates more obvious, reducing errors in judgment. The paper also processes the prosodic of a voice, to achieve the adjustment of fundamental frequency and duration, reflecting the PSOLA processing capabilities of the rhythm.The results of this paper show that the results of consonant vowel synthesis is very well, people are able to hear the synthetic syllable clearly. In estimating the pitch period, From the combination of two algorithms, we can see the peak performance is more obvious,and it reduces the misjudgment. In the processing of a voice, From the experimental results, we can see the algorithm in terms of rhythm regulation has been more skilled, it can achieve better results.