摘要
针对驾驶室的低信噪比和非平稳复杂噪声环境,根据浊语音谐波结构特性,提出一种鲁棒的浊语音检测算法。将输入信号分帧,检测每帧信号中与谐波对应的峰,对连续多帧的峰进行谱跟踪,得到一簇谱线,这簇谱线能很好地表征浊语音的谐波特性,并提取谱线簇的谐波特征,进行浊语音判别。船舶驾驶环境下的浊语音检测实验表明,该方法能够在低信噪比条件下,从突发噪声、非平稳噪声中可靠地检测出浊语音,表现出较强的抗噪能力和较高的准确率。
Since low signal-to-noise ratio(SNR) and non-stationary noises in the environment are hard to discriminate, this pa- per proposed a robust voiced speech detection method based on harmonic structure of voiced speech. Firstly it cut the input sound signal into frames, and picked out spectral peaks, which corresponded with the harmonic of the voiced speech. Secondly, it tracked a set Of lines which represented the harmonic structure of voiced speech from the peaks in several adjacent frames. Last it picked up the frequency and energy characters as foundation to speech diction from the spectral lines. Simulation results show that the method can detect the voiced speech reliably from non-stationary and burst noises in low SNR environments, and has ~ood anti-noise performance and high accuracy in experiments on the database acquired in ships' steer house.
出处
《计算机应用研究》
CSCD
北大核心
2013年第8期2461-2463,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(60572160)
关键词
浊语音检测
谐波结构
谱跟踪
峰值检测
动态规整
voiced speech detection
harmonic structure
spectral tracking
peak picking
dynamic programming