使用目標與干擾比之語音活動偵測建構雜訊能量估測器

Abstract

本論文提出一套可適用於非穩態雜訊環境的雜訊能量估測器，並使用此估測結果進行語音純化。雜訊能量估測器使用目標阻斷器和方向性補償器來估測雜訊能量，並且為了調整適應性的方向補償器，採用目標與干擾比進行語音活動偵測，之後根據語音活動偵測結果結合頻譜遮罩進行語音純化。為了確認此演算法的效能，在各種不同的雜訊環境下進行實驗，並透過訊雜比與知覺語音評價兩項客觀標準進行評量，由實驗證明，此法在非穩態雜訊的環境中，可以有效的壓抑干擾雜訊，目標聲源也能同時保有不錯的語音品質。

This thesis proposes a noise power estimator for nonstationary noisy environment, and uses the result for speech purification. The noise power estimator combines target blocker and directivity compensator to estimate noise power. For updating the adaptive directivity compensator, we use target to jammer ratio (TJR) based voice activity detection (VAD). The VAD result is also combined with spectrum mask for speech purification. To verify the effectiveness of the proposed algorithm, we experimented in different noise environments, and verified it by signal-to-noise ratio (SNR) and perceptual evaluation of speech quality (PESQ). The experimental results show that the algorithm is able to reduce interference and also save the speech quality.