Abstract: This paper compares and analyzes two types of noise estimate algorithms which based on traditional endpoint detection and based on minimum statistics and optimal smoothing，then applies them to spectral subtraction algorithm.Spectral subtraction based on minimum statistics is more effective for enhancement non-stationary noise，but it has“music noise”problem.A modified method is proposed to resolve this problem.Simulation results show that this method enhances the Signal to Noise Ratio（SNR） and improves the speech quality.

Key words: endpoint detection, minimum statistics, noise estimate, spectral subtraction

摘要： 对基于传统端点检测技术和基于最小统计和平滑滤波的两类噪声估计方法进行了对比分析，并应用于谱相减算法中。针对基于最小统计和平滑滤波的谱减算法中出现的“音乐噪声”问题，提出了一种改进方案，实验表明，使用改进谱减法增强后的语音信号在信噪比和语音质量上均有提高。

关键词: 端点检测, 最小统计, 噪声估计, 谱相减