基於模糊系統的混合式濾波強化系統

An Adaptive Hybrid Filter Based on Fuzzy System for Speech Enhancement

張園田(Yuan-tain Chang)

指導教授 ： 姚志佳

朝陽科技大學/資訊學院/資訊工程系/碩士(2010年)

摘要

現實生活中，語音常常伴隨著不同的背景噪音。由於噪音會影響語音記錄的品質，造成訓練的聲學模型與實際應用環境不匹配，使得語音辨認系統的辨認正確率大大的降低。因此，本篇論文的主題就是針對上述的問題，對最常見的加成性噪音，提出一個有效的語音強化方法，針對輸入的語音訊號分類為穩態與非穩態類型，分析其中模型訊號之特性，並將頻譜刪減法與小波濾波以及維納濾波三個濾波方法做結合，以模糊推論系統重建上述三個濾波器的分配比率，以期待還原最理想的語音訊號，將環境噪音對與語音的影響降至最低，並且彌補使用單一濾波方式的缺點影響，提升SNR值。在實驗中，本論文使用Aurora-2雜訊語料庫做語音訊號的強化，實驗評估後經過混合式濾波強化系統能有效提升SNR值。

關鍵字

語音辨識 ； 小波濾波 ； 語音強化 ； 頻譜刪減法 ； 維納濾波

並列摘要

n this paper an adaptive hybrid filter, based on fuzzy inference system, is proposed for speech signal enhancement. As we know, noise causes the drastic reduction of the accuracy ratio of the speech recognition system. In this paper an effective voice enhancement scheme is proposed to filter the additive noise. In the hybrid filter spectral subtraction, wavelet filtering and wiener filtering are cooperated to filter the signal and complementary to each other. And, the input voice signal is classified into non-stationary and stationary type. The former signal is processed by the wavelet filter and the latter signal is processed by spectral subtraction and wiener filter. Within the part of spectral subtraction and Winner filter, the signal is classified by the probability of residual noise. Besides, the mixed ratio of these three filters is provided by eight fuzzy rules. Moreover, a compensation algorithm was proposed to eliminate the unwanted spur and compensate the filtering signals. Through the experimental results, it can be concluded that our proposal can improve the SNR effectively.

並列關鍵字

Wiener Filtering ； Wavelet Filtering ； Spectral Subtraction ； Speech Recognition ； Speech Enhancement

參考文獻

[1]C.-T. Lu, “Reduction of musical residual noise for speech enhancement using masking properties and optimal smoothing,” Pattern Recognition, vol.28, pp. 1300-1306, March. 2007.

[2]C.-T. Lu, K.-F. Tseng, “A gain factor adapted by masking property and SNR variation for speech enhancement in colored-noise corruptions,” Computer Speech and Language, vol.24, pp. 632-647, September. 2009.

[3]S. Rangachari, P.C. Lizou, “A noise-estimation algorithm for highly non-stationary environments,” Speech Communication, vol.48, pp. 220-231, 2006.

[4]C.-T. Lu, K.-F. Tseng, “Non-stationary signal processing using time-frequency filter banks with applications, Signal Processing, vol.86, pp. 3021-3030, February. 2006.

[5]I. Cohen, B. Berdugo, “Speech enhancement for non-stationary noise environments,” Signal Processing, vol.81, pp. 2403-2418, June. 2001.