聲訊音調改變之研究設計

Abstract

本文主要在探討與研究聲音訊號之音調改變等問題，進而設計出能作即時 處理、低複雜度、高傳真之可調音調與速率之演算法。其中主要探討問題 為：如何改變輸入聲音訊號之音調。此問題可應用在聆聽者可隨意地依個 人喜好調整樂音或聲樂訊號之音高。另一用途為樂音或語音的壓縮。我們 可減慢輸入聲訊之播放速率降低其音高，如此聲訊資料量減少而達到壓縮 的目的，若要解壓縮，只需以音調改變的演算法調回原來之音高。本文嘗 試多種方法解決音調改變之問題，最快速及能保持高品質之方法為：將聲 音訊號切割成一段段的音區，每段音區再根據音調改變的程度，將音區壓 縮或增展，下一段音區選取時，必須符合其起始處與處理後聲訊的尾端在 時間軸上相對應。每一音區與音區銜接處，用MAE之方法尋找最適當之銜 接處，如此其聲音將更為傳真，而且計算量低。再運用次取樣法 (Subsample) 及區塊二元搜尋法 (Block Binary Search) 可減少相當多 的計算量。最後提出一個硬體實現的架構，以求做出即時改變音調的晶片 。 The thesis work is looking for real-time, low-complexity, and high-quality pitch shifting algorithms. The main problem is how to shift input audio pitch. It can be applied to shift the pitch of music or speech according to anyone's likes. The other application is to compress music or speech. We can reduce the playing rate to shift pitch down by pitch shifting algorithm, so the audio data will decrease to achieve the purpose of compression. If you want to decompress audio data, you could use the algorithm to shift pitch up. The thesis tested many methods to solve the problem of pitch shifting. The fastest and high-quality algorithm is to divide audio signals into frames. Every frame is compressed or expanded by pitch shifting algorithm. The choice of next frame must according to the rule that the begin of current frame and the end of previous frame must be at the same time, and it uses MAE to find the best connection of two frames so that the audio can get higher quality and the complexity of computation can lower. Subsample and Block Binary Search (BBS) can reduce very large amount of computation. Finally, we propose an architecture to perform pitch shifting in real time.