论文标题
使用干燥语音信号和声学系统的联合估计进行验证
Dereverberation using joint estimation of dry speech signal and acoustic system
论文作者
论文摘要
语音消失的目的是从信号中消除随时间不变的脉冲响应过滤器的质量降解效应。在本报告中，我们描述了一种言语编织方法，涉及对干燥语音信号和房间冲动反应的联合估计。我们探索分别适用于每个任务的深度学习模型，以及如何将它们组合成具有共享参数的联合模型。
The purpose of speech dereverberation is to remove quality-degrading effects of a time-invariant impulse response filter from the signal. In this report, we describe an approach to speech dereverberation that involves joint estimation of the dry speech signal and of the room impulse response. We explore deep learning models that apply to each task separately, and how these can be combined in a joint model with shared parameters.
