應用灰色系統理論於語音訊號處理雜訊抑制之研究與設計

A Method and Design for Speech Signal Processing to Suppress Noise by Using Grey System Theory

胡永柟(Yung-Nan Hu) ； 吳仲琪(Chung-Chi Wu) ； 董志強(Chih-Chiang Tung)

《科學與工程技術期刊》 5卷3期 (2009/09) Pp. 43-50

摘要

本論文研究是利用灰色理論(grey theory)中的灰預測(grey prediction)方法來進行語音分析，並提出利用Levinson-Durbin 方法取出極點建立語音模型，這種方法可以讓語音分析的方法加入創新的思維。本文之設計在模擬的過程中共分爲兩大部分：(1)使用麥克風所取得的語音檔案，應用自動回歸方程式擷取出其極點(poles)資料並存於資料庫。(2)考慮具有白雜訊(white noise)之環境中透過自動回歸(auto regression, AR)模型濾波器及灰預測之方法使其語音更加清晰。本論文所模擬之結果最重要的是可以有效抑制雜訊及製作模型來增加語音處理的速度。期望能落實理論於實務設計中，以另一種方法提供給語音辯識領域可以有其它選擇。

關鍵字

灰色理論 ； 灰預測 ； AR模型 ； 語音處理

並列摘要

In this research the prediction component of grey theory is applied to analyze speech. For the purpose of innovation, the Levinson-Durbin method is adopted for extracting poles to construct a speech model. The two most critical factors are as follows: (1) MATLAB software is used to extract the pole data, which is established by using a microphone to compile speech files and concurrently construct a data-base for storage. (2) To ascertain the quantity of the speech data, an auto regression (AR) filter model is considered in the AWGN (additive white Gaussian noise, AWGN) environments. In this experiment, not only is a method for suppressing speech noise proposed, but also a speech model established for increasing the speed of speech processing. We expect the results of this research to provide another choice in promoting performance in speech recognition.

並列關鍵字

grey theory ； grey prediction ； AR model ； speech processing