1Lockheed Martin Aculight, Bothell, Washington, USA; and 2Department of Electrical Engineering, University of Texas at Dallas, Richardson, Texas, USA.
Ear Hear. 2014 Mar-Apr;35(2):262-70. doi: 10.1097/AUD.0b013e3182a768e8.
There is recent interest in focused stimulation of the cochlea via modalities such as tripolar electrical and infrared neural stimulation to improve speech in noise comprehension and music perception. The purpose of this work was to use vocoder-based simulations to investigate speech recognition for broad stimulation (standard monopolar paradigm) versus more focused stimulation under a variety of signal-to-noise ratios, dynamic ranges, and numbers of discriminable loudness steps.
Vocoder simulations were used to assess the intelligibility of sentences, consonants, and vowels that were noise vocoded and presented to 7 normal-hearing listeners for identification. A novel aspect of the simulations presented here was the use of nonuniform quantization steps within the dynamic range to more closely simulate the Weber functions observed in cochlear implant users. Intelligibility was assessed for the different filter slopes under a variety of signal-to-noise ratio levels, dynamic ranges, and numbers of discriminable steps.
Speech processed via vocoder simulations representing focused stimulation was found to be substantially more intelligible than speech processed via a monopolar electric vocoder simulation, with differences of up to 60 percentage points. There were no significant differences, however, seen between the two focused approaches (signal attenuations of 10 and 17 dB/mm) for the conditions investigated. Speech processed via the highly focused vocoder (17 dB/mm) was robust to constraints on small envelope dynamic range and small number of discriminable steps within the dynamic range, as high performance was maintained with at least a 5 dB dynamic range and eight or more discriminable steps. Significant drops in intelligibility were noted when the number of steps fell below eight.
Highly focused stimulation-tripolar electrical and infrared neural stimulation-has potential for increased performance in noise compared with monopolar stimulation, but much work remains to bear this potential out and to take full advantage of each modality's strengths.
最近人们对通过诸如三极电和红外神经刺激等方式来集中刺激耳蜗以提高言语在噪声中的理解和音乐感知能力产生了兴趣。本研究的目的是使用基于声码器的模拟来研究在各种信噪比、动态范围和可分辨响度级数量下，宽刺激（标准单极范式）与更集中刺激的语音识别。
使用声码器模拟来评估噪声声码化并呈现给 7 名正常听力听众进行识别的句子、辅音和元音的可懂度。这里模拟的一个新颖方面是在动态范围内使用不均匀的量化步长，以更接近地模拟在人工耳蜗使用者中观察到的韦伯函数。在各种信噪比水平、动态范围和可分辨步骤数量下，评估了不同滤波器斜率下的可懂度。
通过代表集中刺激的声码器模拟处理的语音被发现比通过单极电声码器模拟处理的语音更具可懂度，差异高达 60 个百分点。然而，在所研究的条件下，两种集中方法（信号衰减 10 和 17 dB/mm）之间没有观察到显著差异。通过高度集中的声码器（17 dB/mm）处理的语音对小包络动态范围和动态范围内可分辨步骤数量的限制具有鲁棒性，因为在至少 5 dB 的动态范围和 8 个或更多可分辨步骤的情况下保持了高性能。当步骤数量低于 8 时，可懂度显著下降。
与单极刺激相比，高度集中的刺激——三极电和红外神经刺激——具有在噪声中提高性能的潜力，但仍有大量工作需要做，以验证这种潜力并充分利用每种模式的优势。
