基于粒子群优化的哼唱语音端点检测算法

Endpoint detection of humming music speech based on particle swarm optimization

针对传统方法难以处理连音分割的问题,提出基于信息熵的粒子群优化分割方法.利用信息熵对语音进行粗分割,在长语音段计算时间序列复杂度的基础上,通过粒子群优化算法,寻找以长语音段的信息熵均值为适应值的分割点.实验结果表明,与基于短时能量与过零率、信息熵等分割算法相比较,提出的粒子群优化分割方法在分割准确率上有较大提高.

Because of the limitation of traditional speech segmentation to continuous music speech, this paper presents a segmentation method based on optimized particle swarm optimization and informa-tion entropy. Firstly we roughly segment the speech by using information entropy. Then we count the estimated complexity of the long speech segments. And next we find the right split points by fitness value of the long speech segments＇ average information entropy. The results show that the proposed method could improve segmentation accuracy largely compared with the methods by using short energy and zero crossing rate or single information entropy.