基于谱减法的藏语语音减噪研究

语音信息的采集过程中难免会有各种噪声的干扰,噪声降低了语音的可懂度及清晰度,甚至严重影响语音处理的准确性和可靠性。谱减法是语音减噪中最常用的方法,该方法基于人的感觉特性,对语音短时幅度谱进行估计,适用于受加性噪声污染的语音。本文通过统计分析谱减法的增益补偿因子和过减因子参数,研究基于谱减法的藏语语音减噪处理,并取得较为理想的实验结果。

Study on Noise Reduction of Tibetan Speech based on Spectral Subtraction

In the process of the acquisition of speech information, there will inevitably be all kinds of noise interference. The noise reduces the intelligibility and quality of speech and even seriously affects the accuracy of speech processing. Spectral subtraction is the most commonly used method for the noise reduction of speech, which estimates the short-time amplitude spectrum of the speech based on the characteristics of human consciousness. It is suitable for speech that is contaminated by additive noise pollution. In this paper, statistical analysis of spectral subtraction gain compensation factor and reduction factor parameter is used to study the noise reduction processing of Tibetan language speech based on spectral subtraction, and an ideal experimental result is obtained.