耳蝸物理模型為基礎的音高辨識方法

Abstract

本論文提出了一個新的音高辨識方法：利用簡單的耳蝸物理模型，可以讓電腦模擬人耳聽覺的產生，來辨識音高。傳統音高辨識方法主要分為兩大類：第一類型方法利用簡單的自相關函式，將一段聲音波形輸入函式後，經由簡單的運算，可以迅速得到主要頻率音高；第二類型方法則是將一段時域資料的聲音波形，經過傅立葉轉換後，轉成頻域資料得到頻譜，再分析頻譜得到音高。本論文藉由一個簡單的耳蝸物理模型，可以直接利用時域資料去振動耳蝸中的基底膜，再藉由分析基底膜的振動情形抓出音高。不同於第一類方法只能抓到一個主要的頻率，我們的方法，因為整條基底膜的彈性並不一致，所以可以同時抓出各個頻率的組成大小。另外，少了頻譜轉換的步驟，因此我們的方法，運算速度比起第二類方法快速許多。

In this paper, an algorithm for pitch recognition is designed. This algorithm is based on a simplified cochlear model. The traditional methods are mainly divided into two categories: one is to utilize and analyze the amplitude of sound in time domain directly; the other is to transform the sound into the frequency domain first, and then do some analysis to recognize the pitch. The operation amount in time domain is relatively small, but mostly it can only detect a single frequency. The second type of methods needs to do the transform first, so the speed is relatively slow. After getting the frequency spectrum, we can apply some algorithm to do the pitch recognition. My algorithm, which is called CM (Cochlear Model), combines the advantages of above-mentioned two kinds of methods. CM utilizes the amplitude of sound directly. Through the simple cochlea physical model, the vibration situation of the BM(basement membrane) in the cochlea can tell the pitch. For the elasticity in the BM is not uniform, we can tell more than one single frequency at the same time.