華語韻律轉換之研究與實作

Research and Implementation of Prosody Conversion for Mandarin Chinese

江克敬(Ke-Ching Chiang)

指導教授 ： 張智星

國立清華大學/電機資訊學院/資訊工程學系/碩士(2008年)

摘要

本論文建置一個華語韻律轉換的系統，可達成語音的：（1）聲調轉換，（2）長度轉換與（3）音量轉換。
我們先對整句語句使用UPDUDP （unbroken pitch determination using dynamic programming）進行音高追蹤，對整個語句取出平均幅度差函數法的（average magnitude difference function, AMDF）時域特徵，再以動態程式最佳化（DP, dynamic programming）擷取出連續而不中斷的音高特徵曲線。同時我們也用維特比譯碼器（viterbi decoding）進行強制對位（forced alignment）切出音素。
接下來取出非氣音的部分進行韻律轉換，合成方式是在時域上作處理，利用基週同步疊加法（PSOLA, pitch synchronous overlap and add）調整基週頻率以達成聲調轉換。用波形相似性疊加法（WSOLA, waveform similarity overlap and add）調整音長。再將原始語音與目標語音的每個基週標位到下一個基週標位的間隔當成音框大小（frame size），求取其音量，兩者進行線性轉換（linear mapping）之後，調整原始語音的音量以達成音量轉換。
我們還嘗試了另一種基頻擷取與合成的方法STRAIGHT（speech transformation and representation using adaptive interpolation of weighted spectrum）進行聲調轉換，並比較STRAIGHT與PSOLA兩種不同方法的合成結果。STRIGHT可以達到比較好的合成效果，但是會花費較長的程式執行時間。

關鍵字

韻律轉換 ； 語音合成

並列摘要

In this thesis, a prosody conversion system for Mandarin Chinese was developed in order to achieve (1) pitch conversion, (2) duration conversion, (3) energy conversion.
UPDUDP (unbroken pitch determination using dynamic programming) is first applied to each utterance for pitch tracking which extracts an unbroken pitch contour from a given utterance based on time-domain acoustic feature of AMDF (average magnitude difference function) and DP (dynamic programming). Utterances are then segmented into phonemes using Viterbi forced alignment.
The voiced parts of each utterance are then extracted to perform prosody conversion. Voice conversion is achieved by using PSOLA (pitch synchronous overlap and add) in time domain to adjust the fundamental frequency. WSOLA (waveform similarity overlap and add) is employed to adjust duration. Frame size is set to be the interval between one pitch mark and the next pitch mark of source wave and target wave. The volume for each frame is thus computed and linear mapping is performed. Energy conversion is then achieved by adjusting the source energy to the target energy.
We have also made an attempt of using another approach of pitch tracking and speech synthesize called STRAIGHT (speech transformation and representation using adaptive interpolation of weighted spectrum) for voice conversion. The synthesized result of STRAIGHT is compared with that of PSOLA, and it is found that STRAIGHT can achieve better performance than PSOLA but requires more computation time.

並列關鍵字

voice conversion ； speech synthesis

參考文獻

[3] 詹詩涵, 基於音高調節之歌聲合成系統., 國立清華大學資訊系統及應用研 究所, 碩士論文, 2006.

[2] 石文俐, 中文語音合成之韻律產生器的改良與研究., 國立清華大學資訊工

[11] Jiang-Chun Chen, J.-S. Roger Jang, "TRUES: Tone Recognition Using Extended Segments", ACM Transactions on Asian Language Information Processing, 2008.

[14] Cheng-Yuan Lin and J.-S. Roger Jang, "A Two-Phase Pitch Marking Method for TD-PSOLA Synthesis", GESTS International Transaction on Speech Science and Engineering, No. 2, Vol. 1, PP. 211-221, Dec 2004.

[4] T. Styger and E. Keller, “Formant synthesis”, In E. Keller (Ed.), Fundamentals in Speech Synthesis and Speech Recognition, pp. 109–128. Wiley, 1994.