It is useful to decide whether the microphone signal includes a target speech or not at a temporal moment because the process called the voice activity detection (VAD) can reduce any redundant efforts made for the speech coding or the speech recognition, or it can help provide more accurate noise estimation for the speech enhancement. The detection of speech or non-speech in a frame has been simply done by observing the variance of its energy level, zero crossing rate or periodicity. In this occasion, however, the detection error increases exponentially as much as the background noise is added up. Unvoiced fricative sounds which have low energy with being distributed over widebands are more vulnerable to the background noise than any other phonemes are. It is proposed in this literature that voice activity can be detected more robustly in noisy environment by observing the subband power ratio of the noisy speech and its beamformed signal. Also, it is shown to be effective in the fricatives than in the vowels. Whatsoever, this method guarantees much better performance than single microphone VADs when the noise is obviously reduced by beamforming.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.