| 摘要: | 本研究目的是利用德州儀器TMS320C6713開發板(Texas Instruments, Dallas, Texas, USA)以主動式噪音控制(Active Noise Cancellation, ANC)中的反饋式(Feedback)系統為基本架構，應用於耳罩式耳機。本研究分為軟體模擬與硬體實現兩部分，而此兩部分實驗的噪音語料及語音語料都使用相同的，而噪音分為單頻、複頻、以及多頻，語音則是使用之前實驗室學長錄製的語料。第一部分利用MATLAB軟體比較三種演算法，分別為濾波X最小均方演算法(FXLMS)、濾波X正規化最小均方演算法(FXNLMS)，以及洩漏濾波X最小均方演算法(Leaky-FXLMS)，模擬耳機內有無語音狀態下的消噪效果，比較出除噪效果最佳的演算法。而第二部分利用TMS320C6713開發板結合耳機與其他相關電路實現軟體模擬中最佳的運算方法。硬體實現中分為無語音和有語音兩部分實驗，兩者實驗差別在於，耳機的喇叭放出的聲音有無語音。無語音的實驗中，單純消除外界噪音，在單頻和複頻噪音平均消噪達15 dB以上，在多頻噪音下，可降噪的頻率範圍內有除噪效果的。而有語音的實驗中，單頻噪音訊噪比平均10dB以上，複頻噪音訊噪比平均5dB左右，但多頻噪音效果就不是很明顯。結論是，無論在何種噪音環境下，大都可以降低外在的噪音干擾，同時耳機的音訊輸入都沒有失真，保有原先的語音解析度。;This thesis presents the implementation of an adaptive feedback active noise cancellation (ANC) system for headphone application utilizing TMS320C6713 DSP Starter Kit.
The experiment in this study was divided into two sections: the software simulation and the headphone implementation. In this research, the speech and noise were used the same in two parts. Noise was divided into three sections: single frequency noise, dual frequency noise, and multi-frequency noise. And the speech was recoded from the laboratory senior. In the first part, this research used MATLAB to compare three ANC algorithms: Filtered-X Least Mean Square (FXLMS), Filter-X Normalized Least Mean Square (FXNLMS), and Leaky Filtered-X Least Mean Square (Leaky-FXLMS). The denoising performance of the three algorithms with and without speech output from the headphone were simulated. The most effective method of algorithm in reducing unwanted noise was chosen. In the second part, the optimal method in previous simulation was implemented utilizing TMS320C6713 board combined with headphone and the circuit related to the experiment. There were two parts of experiments in hardware, they were different from whether having speech from headphone or not. In the experiment without speech, the results showed the noise reduction more than 15dB in average in single frequency and dual frequency noise, and the result showed the effectiveness of reducing noise in possible reduction frequency range in multi-frequency noise ; the experiment with speech, the results showed more than 10dB in average noise reduction in single frequency and noise reduction around 5dB in average in dual frequency noise, but the result of attenuating noise was not obvious in multi-frequency noise. In conclusion, in all kinds of noise environment, most results showed the algorithm using in hardware not only could reduce unwanted noise but also allowed the desired speech signal to pass through without cancellation and maintain the original speech quality at the same time. |
