本文為美國德克薩斯大學達拉斯分校（作者：SUNDARRAJAN RANGACHARI）的碩士論文，共73頁。

在存在噪聲背景的情況下，語音增強演算法可以提高語音的質量和可懂度。本文針對語音增強應用中的噪聲譜估計問題進行了研究。針對高度非平穩的噪聲環境，提出了兩種噪聲估計演算法。在方法1中，首先使用語音活動檢測器將每幀連續的語音訊號分類為語音存在/不存在幀，並使用語音不存在幀的常數平滑因子和語音存在幀的頻率依賴平滑因子更新噪聲譜估計。在方法2中，使用頻率依賴平滑因子來更新噪聲頻譜估計，而不需要判斷每一幀中是否有語音存在。在這兩種方法中，頻率依賴平滑因子都是基於子帶估計的語音存在概率來計算的。語音存在性是通過計算有噪語音功率譜與其區域性最小值的比值來確定的，該比值是通過利用前瞻因子對有噪語音功率譜的歷史值進行平均來計算的。採用區域性最小估計演算法可以很快適應高度非平穩的噪聲環境。經過正式的聽力測試證實，本文所提出的噪聲估計演算法在用於語音增強時優於其它噪聲估計演算法。

The quality and intelligibility of the speech in the presence ofbackground noise can be improved by speech enhancement algorithms. This thesisaddresses the issue of estimating the noise spectrum for speech enhancementapplications. Two noise estimation algorithms are proposed for highlynon-stationary noise environments. In method-1 a voice activity detector is first used to classifyeach frame of speech continuously into the speech present/absent frames, andthe noise spectrum estimate is updated using a constant smoothing factor forspeech absent frames and a frequency dependent smoothing factor for speechpresent frames. In method-2 the noise spectrum estimate is updated using afrequency dependent smoothing factor irrespective of speech present/absentframes. In both methods, the frequency dependent smoothing factor is calculatedbased on estimated speech presence probabilities in subbands. Speech presenceis determined by computing the ratio of the noisy speech power spectrum to itslocal minimum,which is computed by averaging past values of the noisy speech powerspectra with a look-ahead factor. The local minimum estimation algorithm adaptsvery quickly to highly non-stationary noise environments. This was confirmedwith formal listening tests that indicated that the proposed noise estimationalgorithms when integrated in speech enhancement were preferred over othernoise estimation algorithms.

1 引言

2 分析已有的噪聲估計演算法

3 本文提出的噪聲估計演算法

4 總結與結論

下載英文原文地址：

更多精彩文章請關注微訊號：