SIGNAL PROCESSING STRATEGIES FOR COCHLEAR PROSTHESIS

本研究的长期目标是改善当前人工耳蜗信号处理策略的语音性能。目前的研究将提出一系列策略，这些策略将使用感知上与输入信号“最接近”的信息来刺激电极。在这些策略中，通道振幅的选择使得刺激（电）模式和声学模式之间的差异最小化，也就是说，通道振幅是通过优化感知驱动的误差准则来选择的。本文将研究两种类型的误差准则，一种是强调包络信息重要性的时间准则，另一种是强调光谱信息重要性的光谱准则。提出的实验有三个具体目的。第一个目的是研究为更好的频谱匹配而设计的策略是否比传统的连续交错采样（CIS）策略在语音识别任务上表现得更好。第二个目的是了解光谱信息在Med-El和Nucleus人工耳蜗设备的单词识别中的重要性。为了实现这一目标，我们将向患者植入来自TIMIT语音语料库的句子，其中元音或辅音将被白噪声取代。第三个目的是研究为更好的波形匹配而设计的策略是否比CIS策略在语音识别任务上表现得更好。所提出的策略将在便携式语音处理器上实施，并将在人工耳蜗患者身上进行测试。这些研究的结果将为更好地理解哪个信息，时间（包络）或频谱，在感知上对人工耳蜗植入更重要奠定基础。最重要的是，这将为未来人工耳蜗信号处理策略的发展开辟新的途径。

The long term objective of the proposed research is to improve the speech performance of current signal processing strategies for cochlear prosthesis. The present study will propose a series of strategies which will stimulate the electrodes using information that is "closest", perceptually, to the input signal. In these strategies, the channel amplitudes are selected such that the difference between the stimulating (electrical) pattern and the acoustic pattern is minimized, that is, channel amplitudes are selected by optimizing a perceptually motivated error criterion. Two types of error criteria will be investigated, a temporal criterion, which emphasizes the importance of envelope information, and a spectral criterion which emphasizes the importance of spectral information. The proposed experiments have three specific aims. The first aim is to investigate whether a strategy designed for better spectral matching performs better on speech recognition tasks than the conventional Continuous Interleaved Sampling (CIS) strategy. The second aim is to understand the importance of spectral information on word recognition with the Med-El and Nucleus cochlear implant devices. To accomplish this aim, we will present to implant patients sentences from the TIMIT speech corpus in which the vowels or the consonants will be replaced with white noise. The third aim is to investigate whether a strategy designed for better waveform matching performs better on speech recognition tasks than the CIS strategy. The proposed strategies will be implemented on a portable speech processor and will be tested with cochlear implant patients. The results of these investigations will lay the groundwork for a better understanding of which information, temporal (envelope) or spectral, is perceptually more important for cochlear implants. Most importantly, it will open new avenues for the development of future signal processing strategies for cochlear prosthesis.