Robust speech/non-speech detection in adverse conditions using the fuzzy polarity correlation method

Speech/non-speech detection is important in many areas of speech processing technology. In real environments, the speech signal is usually corrupted by background noise, which greatly affects the performance of speech processing systems. It is well known that a major cause of the efficiency decrease in automatic speech recognition (ASR) is the inaccurate detection of the endpoints. Therefore, high performance speech recognition requires efficient speech detection, especially in noisy environments. The paper describes a robust speech/non-speech detection algorithm with high reliability in very noisy environments. The algorithm is based on the fuzzy polarity correlation method. In order to decide on the speech/non-speech section, we use a similar degree between the positive polarity correlation sequence and the negative polarity correlation sequence for the input speech signal. The evaluation results show that this detection algorithm can obtain higher rates of accuracy in noisy environments