一种抑制方向性噪声的双耳语音增强算法  

A Binaural speech enhancement algorithm that suppressing the directional noises

作　　者：方义[1] 冯海泓[1,2] 陈友元[1] 胡晓城[2] FANG Yi FENG Haihong CHEN Youyuan HU Xiaocheng(Shanghai Acoustics Laboratoryrm , Chinese Academy of Science Shanghai 200032 Jiaxing Engineering Center, Institute of Acoustics, Chinese Academy of Science Jiaxing 314006)

机构地区：[1]中国科学院声学研究所东海研究站,上海200032 [2]中国科学院声学研究所嘉兴工程中心,嘉兴314006

出　　处：《声学学报》2016年第6期897-904,共8页Acta Acustica

基　　金：国家自然科学基金青年科学基金项目(11104316);中国科学院声学研究所所长择优基金项目(Y154221701)资助

摘　　要：为了给双耳听力设备佩戴者带来更好的语音可懂度,提出了一种利用双耳时间差与声级差的近场语音增强算法,该方法首先利用这两种差异来估计语音的功率谱和语音的相干函数,然后计算干扰噪声在左右耳间的头相关传输函数的比值,最后构造两个维纳滤波器。客观评价的参数显示该算法去噪效果优于对比算法而目标语音的时间差误差和声级差误差低于对比算法。主观的言语接受阈测试表明该方法能有效提高语音可懂度。结果表明,该算法在能够有效去除干扰噪声的同时,保留了目标语音的空间信息。In order to bring better speech intelligibility for the wearer of binaural hearing aids, we presented a near-field binaural noise reduction algorithm based on the interaural time difference and interaural level difference. We used these two differences to estimate the power spectral density and the coherence function of the speech. And then we calculated the ratio between the left and right Head Related Transfer Functions for the directional noise. Finally, we constructed two Wiener filters. According to the objective parameters, we can find that this algorithm showed a better performance of noise reduction than compared algorithms while the interaural time difference error and interaural level difference error of target speech were lower than the compared algorithms. The subjective speech reception thresholds test showed that using this method we can effectively improve the speech intelligibility. Then we can come to the conclusion that this method can significantly suppress the interference noise while preserving the spatial information of the target speech.

关 键 词：算法处理 耳语音 语音段 功率谱 维纳滤波器 可懂度 语音增强 佩戴者 言语接受阈 传输函数 

分 类 号：TN912.35[电子电信—通信与信息系统]