STUDY ON A NEW COCHLEAR IMPLANT SYSTEM BASED ON PHYSIOLOGICAL MODEL OF AUDITORY PERIPHERY SYSTEM

这个为期两年的项目由六个部分组成。(1)在多台数字信号处理器上实现了听觉滤波和毛细胞模型的实时处理器。(2)利用听觉模型设计了一种新的语音处理器，以提高辅音的质量。(3)设计了一种新的声学模拟，根据测量的t水平产生感知等效的响度感觉。c级电流。(4)对低频FIR滤波器进行了重新设计，提高了特征频率的分辨率。(5)采用听觉模型对不发音区和LPC模型对发声区进行复合语音处理。(6)基于语音编码技术提取触发脉冲。该项目的结果如下：在微处理器上实现了足够精度的定点信号处理。新的语音处理器从听觉模型的低通道中提取基频。在一个音高间隔内，选择四个最常见的放电通道依次刺激。通过声学模拟的感知测试显示，发音的顿音、不发音的摩擦音、不发音的非洲音、不发音的非洲音和鼻音的音质有所提高。元音质量不如目前实用的语音处理器。采用LPC模型设计复合处理，改善元音词性。LPC模型可以提高元音的语音质量，但辅音有退化。问题是语音的辅音部分不一定是由听觉模型处理的，而是由LPC处理的。研究了基于波编码技术提取驱动脉冲序列以最小化合成波平方误差的不同方法。提取算法考虑了人工耳蜗系统的约束条件。通过分析和综合的语音感知测试，元音质量良好，辅音质量一般。这似乎值得继续进一步研究。

This two year project consists of six parts. (1) A real-time processor of the auditory filter and the hair-cell model had been implemented on multiple digital signal processors. (2) A new speech processor was designed using an auditory model to improve the quality of consonants. (3) A new acoustic simulation was designed to cause perceptually equivalent sensation of loudness in accordance with the measured T-level. C-level of electric current. (4) The FIR filters of low frequency were redesigned to have higher resolution for the characteristic frequencies. (5) Compound speech processing was examined with the auditory model for the voiceless region and the LPC model for the voiced region. (6) Trigger pulse was extracted based on speech coding technology.Results for this project were as follows. Fixed point signal processing realized with sufficient precision on a microprocessor. The new speech processor extracts the fundamental frequency from lower channels of the auditory model. Within a pitch interval, four most firing channels were selected to stimulate in sequence. Perceptual test by acoustic simulation had shown improved quality of voiced stops, voiceless fricatives, voiceless africatives, voiced africatives, and nasals. Vowel quality was not as good as the current practical speech processor. The compound processing was designed to improve vowel part of speech by LPC model. The LPC model could improve voice quality of vowels, however, had degraded consonants. Problem remained is that consonant part of speech was not necessarily processed with the auditory model but LPC instead. We examined different approach based on wave coding technology that extracts sequence of driving impulse to minimize squared error of synthesized wave. Extraction algorithm took into account the constraints of cochlear implant system. Speech perception test by analysis and synthesis had shown good quality of vowels and fair quality of consonants. This seems worth continuing further study.