基于深度学习语音分离技术的研究现状与进展

发布时间：2018-05-16 17:29

本文选题：神经网络 + 语音分离　；参考：《自动化学报》2016年06期

【摘要】：现阶段,语音交互技术日益在现实生活中得到广泛的应用,然而,由于干扰的存在,现实环境中的语音交互技术远没有达到令人满意的程度.针对加性噪音的语音分离技术是提高语音交互性能的有效途径,几十年来,全世界范围内的许多研究者为此投入了巨大的努力,提出了很多实用的方法.特别是近年来,由于深度学习研究的兴起,基于深度学习的语音分离技术日益得到了广泛关注和重视,显露出了相当光明的应用前景,逐渐成为语音分离中一个新的研究趋势.目前已有很多基于深度学习的语音分离方法被提出,但是,对于深度学习语音分离技术一直以来都缺乏一个系统的分析和总结,不同方法之间的联系和区分也很少被研究.针对这个问题,本文试图对语音分离的主要流程和整体框架进行细致的分析和总结,从特征、模型以及目标三个方面对现有的前沿研究进展进行全面而深入的综述,最后对语音分离技术进行展望.
[Abstract]:At present, the technology of speech interaction is widely used in real life. However, due to the existence of interference, the technology of voice interaction in real environment is far from satisfactory. The technology of speech separation based on additive noise is an effective way to improve the performance of speech interaction. In recent decades, many researchers all over the world have put in great efforts and put forward many practical methods. Especially in recent years, due to the rise of deep learning research, the technology of speech separation based on deep learning has been paid more and more attention, showing a bright prospect of application, and gradually becoming a new research trend in speech separation. At present, many speech separation methods based on deep learning have been proposed. However, there is a lack of systematic analysis and summary of the deep learning speech separation technology, and the relationship and distinction between different methods are seldom studied. In order to solve this problem, this paper attempts to analyze and summarize the main flow and the whole framework of speech separation in detail, and make a comprehensive and in-depth review of the current research progress in the three aspects of features, models and objectives. Finally, the speech separation technology is prospected.
【作者单位】：中国科学院自动化研究所模式识别国家重点实验室;内蒙古大学计算机系;
【基金】：国家自然科学基金(61573357,61503382,61403370,61273267,91120303,61365006)资助~~
【分类号】：TP183;TP181

【相似文献】