Voice Activity Detection (VAD) is important in speech processing. In the applications, the systems usually need to separate speech/non-speech parts, so that only the speech part can be dealt with. How to improve the performances of VAD in different noisy environments is an important issue in speech processing. Deep Neural network, which proves its efficiency in speech recognition, has been widely used in recent years. This paper studies the present typical VAD algorithms, and presents a new VAD algorithm based on deep neural networks and Viterbi algorithm. The result demonstrates the effectiveness of the deep neural network with Viterbi used in VAD. In addition, it shows the flexibility and the real-time performance of the algorithms.
展开▼