基于WSOLA 算法的語音時(shí)長調(diào)整研究

目前的PSOLA 算法進(jìn)行時(shí)長調(diào)整時(shí), 其計(jì)算量大, 很難實(shí)現(xiàn)實(shí)時(shí)變速不變調(diào)處理。利用波形相似性來解決語音時(shí)長調(diào)整的WSOLA 算法, 實(shí)驗(yàn)證明能生成高品質(zhì)的語音; 在算法上高效、魯棒, 可以通過修改調(diào)整因子α實(shí)現(xiàn)連續(xù)范圍內(nèi)的在線語音時(shí)長調(diào)整。此算法已應(yīng)用在數(shù)字語音教學(xué)系統(tǒng)中, 具有良好的實(shí)時(shí)性和高品質(zhì)語音。

PSOLA, a popular algorithm for time-scale modification, has heavy computation, which is very difficult for onling operation. Waveform similarity is proposed for tackling the problem of time-scale modification of speech, which is called WSOLA algorithm, through experiments produces the high quality speech output, is algorithmically and computationally efficient and robust, and allows for on-ling processing with arbitrary time-scaling actors chosen over a wide continuous range of values. The algorithm has already been used in the tutoring system of the digital speech, having good real-time quality and high quality speech.