以雙耳時間差為基礎之語音增強技術初探

Investigation on Interaural Time Difference Based Speech Enhancement

廖憲正(Hsien-Cheng Liao)

《電腦與通訊》 139期 (2011/06) Pp. 34-38

https://doi.org/10.29917/CCLTJ.201106.0008

摘要

使用人耳聽覺特性來進行噪音消除以增強語音之技術，因其相較於其他語音增強之方法具有較佳之效能以及較低硬體成本之優勢，最近漸漸地受到重視。本論文探討以雙耳時間差來進行噪音遮蔽之方法，並嘗試與語音辨識器作結合，減少多餘之運算。我們以一語音命令實驗來進行測試，結果顯示結合雙耳時間差與辨識器以及使用平滑式噪音遮罩，在低訊噪比以及近噪音源的情況下，辨識率都有相當明顯的提昇。

關鍵字

麥克風陣列 ； 語音辨識 ； 雙耳時間差

並列摘要

Dual-microphone based noise filtering speech enhancement which is motivated by human binaural system, is becoming a popular front-end for automatic speech recognition recently due to its good performance and low cost. This paper discusses the method for speech enhancement using Interaural Time Differences. Moreover, we try to integrate it with a speech recognizer in order to reduce computational complexity. A voice command experiment was conducted to evaluate the method. Results show that integration of Interaural Time Differences and the speech recognizer with a sigmoid-type noise mask could improve recognition rate significantly, even under low signal-to-noise ratios and in the conditions with noise interferences close to the signal sources.

並列關鍵字

Microphone Array ； Speech Recognition ； Interaural Time Difference ； ITD