利用零參數濾波器重構自己的聲音

Abstract

眾所周知，人們在聽到自己的錄音時常常會有不熟悉甚至是不舒服的感覺。這種對自己錄製聲音的怪異感其實是聲音的傳輸機制所導致。當我們在錄製聲音時，麥克風只有錄製到透過空氣傳導的聲音；但我們自己所聽到的聲音，是透過空氣傳導和骨頭傳導這兩者聲音一起組合而成的。有鑒於透過骨頭傳導的聲音無法被錄製，我們希望能找到一個快速且簡單的轉換方法，將自己錄音的聲音轉換成自己熟悉的聲音。 在本篇論文中，我們設計並實現了一個不需要調整參數的零參數濾波器，能將使用者的錄音轉換成使用者自己的聲音。我們將此濾波器和其他的轉換方法（雙參數模型和八參數等化器）一起做盲測驗證實驗。實驗結果顯示，零參數濾波器在使用上最為方便快速，受試者們對轉換結果都能給出正向回饋，並且在整體表現上也不遜於雙參數模型及八參數等化器。最後，我們進一步構思出獨特的潛在應用。

As we know, when people hear their own voices, they often feel unfamiliar or even uncomfortable. This strange feeling about recorded voice is actually caused by the transmission mechanism of the voice. When we record a voice, the microphone only records the sound transmitted through the air; however, the sound we hear is a combination of air-conducted and bone-conducted sounds. Since bone-conducted sounds cannot be recorded, we wanted to find a quick and easy way to convert our own recorded voices into our own familiar voices. In this paper, we design and implement a zero-parameter-filter approach that does not require parameter adjustment and converts one's recording into one's own voice. This filter was tested together with other conversion methods (two-parameter model and eight-parameter equalizer) in a blind validation experiment. The results showed that the zero-parameter filter was the easiest and fastest to use, and the subjects gave positive feedback on the conversion results, and the overall performance was no worse than that of the two-parameter model and the eight-parameter equalizer. Finally, we further conceive unique potential applications.