实时语音增强人工耳蜗的技术研究

发布时间：2018-03-25 18:32

本文选题：人工耳蜗　切入点：语音增强　出处：《中国科学院大学(中国科学院沈阳计算技术研究所)》2017年硕士论文

【摘要】：人工耳蜗是现在市面上唯一能够有效地恢复中重度耳聋患者听力的设备,在周围环境噪声比较小的情况下,人工耳蜗已经能取得非常好的效果。但是如果周围环境比较嘈杂,人工耳蜗产品的效果就不尽人意了。目前国外的人工耳蜗产品占了中国市场的绝大部分份额并且价格十分的昂贵因此研究应用在人工耳蜗产品上的语音增强算法对耳聋患者及其家庭有十分重要的意义。本文首先介绍了人工耳蜗语音增强算法的研究背景和意义以及国内外现状。接着着重对人工耳蜗原理以及语音增强算法进行了详细的介绍。然后结合基于频域特征的语音端点检查算法,提出了一种新的语音增强算法。端点检测算法使用小波包ERB子带的谱熵和改进的频域能量的能熵比法。其中,小波包ERB子带的谱熵考虑了人耳听觉掩蔽模型和语音与噪声信号之间的频率分布之间的不同;频域能量利用了有语音帧和无语音帧的能量不同。维纳滤波算法实时采集语音数据并使用新的参数来区别无语音段和有语音段,并在无语音段平滑更新噪声谱。最后对改进的算法进行了大量的实验。实验结果表明,该端点检测算法能够很好的区分有语音段和无语音段,这就使得在低信噪比的情况下语音增强效果得到了提升,同时算法的鲁棒性和实时性也得到了保障。在与其他两种算法对比中,得到了更好的语音增强效果。
[Abstract]:Cochlear implant is the only device on the market that can effectively restore hearing in patients with moderate or severe deafness. Cochlear implants can achieve very good results when the ambient noise is low. But if the surrounding environment is relatively noisy, The effect of cochlear implant products is not satisfactory. At present, cochlear implant products account for most of the market share in China and the price is very expensive. Therefore, the research on speech enhancement algorithm applied in cochlear implant products is very expensive. It is very important for deaf patients and their families. Firstly, this paper introduces the background and significance of cochlear speech enhancement algorithm and the present situation at home and abroad, and then focuses on the principle of cochlear implant and speech enhancement algorithm. Then the speech endpoint checking algorithm based on frequency domain features is introduced in detail. A new speech enhancement algorithm is proposed. The endpoint detection algorithm uses the spectral entropy of the wavelet packet ERB subband and the energy entropy ratio of the improved frequency-domain energy. The spectral entropy of the wavelet packet ERB subband takes into account the difference between the auditory masking model and the frequency distribution between the speech and the noise signal. The energy in frequency domain takes advantage of the difference between speech frames and voice-free frames. Wiener filtering algorithm collects speech data in real time and uses new parameters to distinguish speech segments from speech segments. Finally, a lot of experiments are carried out on the improved algorithm. The experimental results show that the endpoint detection algorithm can distinguish the speech segment from the speech free segment. This improves the performance of speech enhancement in the case of low signal-to-noise ratio, and the robustness and real-time of the algorithm are guaranteed. Compared with the other two algorithms, a better speech enhancement effect is obtained.
【学位授予单位】：中国科学院大学(中国科学院沈阳计算技术研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【相似文献】