QQ客服 官方微博 反馈留言

题名	电子耳蜗语音处理策略研究及DSP实现
作者	孟丽
学位类别	博士
答辩日期	2009-05-13
授予单位	中国科学院声学研究所
授予地点	声学研究所
关键词	电子耳蜗 定点DSP CIS语音处理策略 调制深度 SII
其他题名	Study on the Speech Processing Strategy for Cochlear Implants
学位专业	信号与信息处理
中文摘要	电子耳蜗 (cochlear implant) 是唯一能够通过电信号直接刺激听觉神经以使重度耳聋患者恢复部分听力的装置。语音处理策略将语音信号转换为刺激电极所需要的电信号，在电子耳蜗技术中起着重要作用。其中CIS (Continuous Interleaved Sampling，连续交错采样) 策略因可达到较高的刺激速率和语音识别的准确率，被所有电子耳蜗产品采用。本文开展了CIS语音处理策略DSP实现和策略改进两个课题的研究。主要研究工作和相关结果如下： 本文第一部分研究CIS策略在所设计的基于定点DSP体外语音处理器上的实现。讨论了DSP系统工作主频确定方法；保证系统稳定最小字长的确定方法；溢出抑制等问题。实验数据表明实现CIS策略时，确定的工作主频可满足系统实时性要求；确定的最小字长可以保证系统的稳定性；采用改进的直接I型滤波器结构实现策略中高阶带通滤波器组，可以有效的抑制溢出，保证系统运算正确性。 本文第二部分对CIS策略进行改进研究，并选用客观指标SII对改进CIS策略进行定性研究。改进CIS策略将输入信号分为m个通道后，在一个刺激周期内选择调制深度最大的n (n < m) 个通道刺激相应电极。作为改进策略客观评价指标，计算改进策略在①. 存在背景噪声时，随着噪声级改变的SII值；②. 不存在背景噪声，随着说话者和收听者距离增大SII值；③. 噪声级和距离同时变化时的SII值。并将改进CIS策略与未改进CIS策略和SMSP策略进行对比研究。结果表明，改进后CIS策略在存在背景噪声、随着谈话者距离增大、背景噪声和谈话距离同时增大时，比未改进CIS策略和SMSP策略具有更好的言语可懂度。本文对现有语音处理策略改进研究，提供了提高电子耳蜗植入者语音识别准确率的理论基础，对电子耳蜗信号处理系统实现具有验证和指导意义。
英文摘要	Cochlear implants are the only prosthetic device that can restore partial hearing to a totally deafened person via electric stimulation of the residual auditory nerve. Signal processing strategy plays an important role in the development of cochlear implants techniques for transforming speech signal to electrical stimuli. This thesis addresses two subjects, the realization of CIS strategy on DSP system and the modified strategy of CIS. The main work and the results are as follows: In the first part of the thesis, a speech processor is developed based on DSP which provides a hardware system for the realization of the CIS speech processing strategy. The thesis reports on the clock rate, Q format which will make sure the system can work steadily. In order to avoiding the overflow, a modified direct form I structure is designed to implement the high-order band-pass filter used in CIS strategy. The preliminary data shows that the CIS strategy can obtain real-time working on the designed speech processor and works steadily and accurately. In the second part of the thesis, a modified CIS speech processing strategy is explored according to the “n – of – m” strategy. The speech signal is filtered into m contiguous frequency bands with band-pass filter bank and then is calculated the modulation depth of the each band. In conventional “n – of – m” strategy, the signal is filtered into m frequency bands, and processor selects n (n < m) envelope outputs with largest energy. But in the modified CIS speech processing strategy, the processor selects n (n < m) envelope outputs with the largest modulation depth from m frequency bands at each stimulation cycle to stimuli. Pulses are then delivered in an interleaved manner to the corresponding n electrodes. An objective index of SII is selected to estimate the modified CIS strategy. The results show that the modified CIS stratety can obtain higher SII value in noise and with distance than the conventional CIS strategy and SMSP stratey. The modified CIS of this thesis improves the intelligibility of speech recognition.
语种	中文
公开日期	2011-05-07
页码	109
内容类型	学位论文
源URL	[http://159.226.59.140/handle/311008/526]
专题	声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文
推荐引用方式 GB/T 7714	孟丽. 电子耳蜗语音处理策略研究及DSP实现[D]. 声学研究所. 中国科学院声学研究所. 2009.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。