語音時間調變之研究

The Research on Time-Scale Modification of Audio Signals.

林亦欣(Yi-Hsin Lin)

指導教授 ： 虞台文

大同大學/電機資訊學院/資訊工程學系所/碩士(2005年)

摘要

現今，語音/聲音訊號時間調變的技術被普遍地應用，例如：語言學習、語音信箱和醫學治療。
在這篇論文中，我們會先複習一些和時間調變相關的方法，包括了 overlap add technique (OLA)，the Sinusoidal-based model以及 Phase Vocoder。接著，我們會分別地改進 WSOLA 及 Phase Vocoder。對於改進WSOLA 而言，根據不同的調變係數來選擇適當的合成訊框(synthesis window)大小可以降低 WSOLA 的缺點。對於改進Phase Vocoder 而言，相位(phase)應該當不規則的高峰(peak)因素被考慮時而被鎖住，即達到更佳的垂直相位統一性(vertical phase coherence)。

關鍵字

時間調變

並列摘要

In the thesis, we will review some related methods of time-scale modification including the emph{overlap-add technique (OLA)}, the emph{Sinusoidal-based model}, and the emph{Phase Vocoder}. Then, we improve WSOLA and Phase Vocoder separately.
For WSOLA, it is shown that picking proper size of synthesis window according to the scaling factor $alpha$ is able to reduce artifacts of WSOLA and, hence, to render the synthesized signal to sound more natural. For Phase Vocoder, we find that phases should be locked with peak irregularity being considered. i.e., to achieve better vertical phase coherence.

並列關鍵字

time-scale modification

參考文獻

[1] M. Dolson, ``The Phase Vocoder: A Tutorial', Computer Music Journal Vol. 10, NO. 4, pages 14-27, Winter

[2] O. Erogul, and I. Karagoz,

``Multiresolutional Modification of Speech Signals for

Rehabilitation Research and Development Vol. 36, NO.3, pages

230-6, July 1999