An Algorithm Research on Speech Endpoints Detection and Its Implementation on DSP
Mei Zhang · International Journal of Digital Content Technology and its Applications · 2013
In order to improve the accuracy and real-time performance of speech endpoints detection, this paper puts forward speech endpoints detection algorithm based on wavelet analysis and fuzzy neural networks. The algorithm firstly extracts the speech signal characteristic quantity based on wavelet analysis. Then, as an input, the characteristic quantity is sent into the fuzzy neural network for endpoint detection operations. Finally it will determine the category of the signal. This algorithm is achieved by TMS320VC5416 DSP as the core circuit. This paper emphatically introduces the extraction of the signal characteristic quantity, the fuzzy neural network model and the learning algorithm. More over, the hardware modules and software designed to realize this algorithm are also given in this paper. Experimental results show that the system’s accuracy rate of endpoints detection is high, and it can correctly judge the speech signal’s endpoint even in low signal noise ratio (SNR).