Voice Activity Detection Based on Harmonic Intensity under Complicated Noise Environment

The robustness of VAD(Voice Activity Detection) is crucial to the construction of the practical speech recognition system. Most means of VAD is done in laboratory-scale environment, it requires steady background noise and high Signal Noise Ratio(SNR). But in fact, these conditions above can´t be satisfied usually. To adapt the complicated noise environment, A new VAD algorithm is put forward based on voice harmonic intensity ,which serves as a basic feature of speech can be used to replace the traditional energy feature. Use harmonic intensity as an indicator, can demarcate rough position of the voice in noise, and then, using two-stage Wiener filter and traditional dual-threshold detection method, VAD can be carried out under noise environment and lay a good foundation for Robust Speech Recognition. The results of experiment under complicated noise environment show that VAD based on the harmonic intensity feature has higher accuracy and better resistance to noise than traditional methods.