CAJ | 학술논문

为提高在强背景噪声条件下语音端点检测的性能，提出一种将改进的谱减法和频谱方差相结合的方法。先利用谱减法对带噪语音信号进行增强处理，然后使用基于频谱方差的方法进行检测，并以一男一女对10个数字和26个英文字母各发音2次的录音作为样本进行仿真验证。实验结果表明：该方法计算简单，可靠性高，能有效减少背景噪声，提高语音信号的信噪比，在较低的信噪比下仍能比较准确地检测到语言信号的端点。
위제고재강배경조성조건하어음단점검측적성능，제출일충장개진적보감법화빈보방차상결합적방법。선이용보감법대대조어음신호진행증강처리，연후사용기우빈보방차적방법진행검측，병이일남일녀대10개수자화26개영문자모각발음2차적록음작위양본진행방진험증。실험결과표명：해방법계산간단，가고성고，능유효감소배경조성，제고어음신호적신조비，재교저적신조비하잉능비교준학지검측도어언신호적단점。
For improving performance of voice activity detection (VAD) under big background noise, proposes an improved method by combining modified spectral subtraction and spectral variance. The proposed scheme employs modified spectral subtraction to enhance the speech firstly, and then detect based on spectral variance. And at the same time, a large number of simulation experiments have been done to confirm the validity of the scheme which take the pronunciations of 10 digits and 26 letters recorded each one twice by a man and an woman as testing samples. Experiment result shows that the method simple calculation and high reliability, it can reduce background noise effectively, increase signal-to-noise ratio of voice signal, can accurately detection voice signal endpoint under low signal-to-noise ratio.