摘要
提出基于短时能量和过零率的简化语音信号双门限端点检测算法,搭建Matlab的算法仿真平台,实验结果表明,基于短时能量和过零率的双门限端点检测算法在保证检测率的前提下,运算复杂度和运算量均优于倒谱、分形、加权门限端点检测方法。采用Verilog语言完成了该模块的设计和仿真,并成功应用于孤立词语音识别系统中。该语音识别系统采用定点数设计方式,语音信号的采样频率为8kHz,每次采样的数据为8bits,晶片内部稳定工作频率为20MHz。实验结果表明,在200个词源的条件下,平均可以达到90%以上的识别效果。
This paper puts forward a simplified two-threshold speech endpoint detection algorithm for speech signal based on shortterm energy and zero-crossing rate. Building a simulation platform of Matlab algorithm, and the experimental results show that the two-threshold speech endpoint detection algorithm for speech signal based on short-term energy and zero-crossing rate is better than cepstrum, fraetal, and spectrum entropy method of endpoint detection, under the premise of computing complexity and computing quantity.
This paper completes the various modules of the design, simulation and systems integration with the Verilog language, using fixed-point design approach, the voice signal sampling frequency of 8kHz, the data for each sample 8bits, the internal stability of the chip operates at 20MHz, the coefficient of linear prediction obtained by the circuit compares with simulation results on Matlab platform, the error rate less than 0.2 percent, on conditions of 200 source words, it can achieve an average of more than 90% of the resuh of recognition.
出处
《计算机与现代化》
2011年第12期57-59,70,共4页
Computer and Modernization
