摘要
提出了一种适用于飞机座舱噪声环境下的鲁棒语音端点检测方法。在分析噪声特征的基础上,首先直接针对带噪语音谱的离散傅里叶变换系数建立复拉普拉斯分布模型;然后进行基于二元假设检验的似然比测试;最后将信号相邻帧的相关性与基于最大后验概率的判决规则相结合。定义两种门限值,根据前一帧的状态与当前帧的观测值共同决策当前帧的状态,从而搜索出语音起止点。实验表明:与目前典型语音端点检测算法对比,该方法在飞机座舱噪声环境中具有较好的鲁棒性。
A robust speech endpoint detection method for detecting the aircraft cockpit noise is presented. Firstly, based on the analysis of cockpit noise characteristics, the complex Laplacian distribution model is applied for the discrete Fourier transform coefficients of noisy speech spectra. Then, the likelihood ratio test based on binary hypothesis detecting is carried out. Finally, two separate thresholds are achieved based on the decision rule of conventional maximum a posterior probability (MAP) incorporating the inter-frame correlation, the state is depending based on the previous frame and the observation, so the speech endpoint is searched. Experiments show that the method has good robustness on the aircraft cockpit noise.
出处
《数据采集与处理》
CSCD
北大核心
2010年第2期223-227,共5页
Journal of Data Acquisition and Processing
基金
总装预研基金资助项目
国家"八六三"高技术研究发展计划(2007BJ131)资助项目
关键词
语音端点检测
舱音信息
复拉普拉斯模型
最大后验概率
speech endpoint detection
aircraft cockpit voice recorder information
complex Laplacian model
maximum a posterior probability (MAP)