Saunders G H, Kates J M
Center for Research in Speech and Hearing Sciences, City University of New York, New York 10036, USA.
J Acoust Soc Am. 1997 Sep;102(3):1827-37. doi: 10.1121/1.420107.
Microphone arrays can improve speech recognition in the noise for hearing-impaired listeners by suppressing interference coming from other than desired signal direction. In a previous paper [J. M. Kates and M. R. Weiss, J. Acoust. Soc. Am. 99, 3138-3148 (1996)], several array-processing techniques were evaluated in two rooms using the AI-weighted array gain as the performance metric. The array consisted of five omnidirectional microphones having uniform 2.5-cm spacing, oriented in the endfire direction. In this paper, the speech intelligibility for two of the array processing techniques, delay-and-sum beamforming and superdirective processing, is evaluated for a group of hearing-impaired subjects. Speech intelligibility was measured using the speech reception threshold (SRT) for spondees and speech intelligibility rating (SIR) for sentence materials. The array performance is compared with that for a single omnidirectional microphone and a single directional microphone having a cardioid response pattern. The SRT and SIR results show that the superdirective array processing was the most effective, followed by the cardioid microphone, the array using delay-and-sum beamforming, and the single omnidirectional microphone. The relative processing ratings do not appear to be strongly affected by the size of the room, and the SRT values determined using isolated spondees are similar to the SIR values produced from continuous discourse.
麦克风阵列可以通过抑制来自期望信号方向以外的干扰，来提高听力受损听众在噪声环境中的语音识别能力。在之前的一篇论文[J. M. 凯茨和M. R. 韦斯，《美国声学学会杂志》99, 3138 - 3148 (1996)]中，在两个房间里使用AI加权阵列增益作为性能指标，对几种阵列处理技术进行了评估。该阵列由五个全向麦克风组成，间距均匀为2.5厘米，沿端射方向排列。在本文中，针对一组听力受损受试者，评估了其中两种阵列处理技术——延迟求和波束形成和超指向性处理的语音可懂度。使用单音节词的语音接收阈值(SRT)和句子材料的语音可懂度评级(SIR)来测量语音可懂度。将该阵列的性能与单个全向麦克风和具有心形响应模式的单个指向性麦克风的性能进行了比较。SRT和SIR结果表明，超指向性阵列处理最为有效，其次是心形麦克风、使用延迟求和波束形成的阵列和单个全向麦克风。相对处理评级似乎不受房间大小的强烈影响，并且使用孤立单音节词确定的SRT值与连续话语产生的SIR值相似。
