学术论文

      融合Mel系数和kNN的语音端点检测

      Voice activity detection based on Mel-scale frequency cepstral coefficients and kNN

      摘要:
      语音端点检测是语音处理过程中的重要环节.为了提高在不同噪声环境下语音端点检测的准确率和鲁棒性,提出了融合语音Mel频率倒普系数和kNN分类算法相的语音端点检测方法.该方法提取语音信号的Mel频率倒普系数作为语音特征参数,然后将特征向量作为kNN分类的输入进行训练学习,建立语音端点检测模型,并进行仿真实验,结果表明该方法是一种准确率高、鲁棒性强的语音端点检测方法.
      Abstract:
      The important part of speech processing is voice activity detection.This paper proposes the integration voice activity detection method by Mel-scale frequency cepstral coefficients and kNN classfication algorithm to improve the accuracy and robustness of endpoint detection in different noise.First,it extracts speech signal Mel-scale frequency cepstral coefficients.Second,it takes the feature vectors as input of kNN classification training and learning,establishes voice activity detection model,and then it carries out a simulation,the results show that this method is a high accuracy,strong robustness voice activity detection method.
      Author: HAN Yun-fei ZHANG Tai-hong BAI Tao
      作者单位: 新疆农业大学计算机与信息工程学院,乌鲁木齐,830052
      刊 名: 信息技术 ISTIC
      年,卷(期): 2017, (3)
      分类号: TN912.34
      在线出版日期: 2017年4月18日
      基金项目: 新疆维吾尔自治区高技术研究发展计划项目资助