面向助听设备的在线神经网络支持去混响的可定制端到端优化
音频与语音处理
2022-05-06 v1 机器学习
声音
摘要
本工作关注使用加权预测误差(WPE)算法的助听设备在线去混响。WPE 滤波需要目标语音功率谱密度(PSD)的估计。近期深度神经网络(DNNs)已被用于此任务。然而,这些方法优化 PSD 估计,其仅间接影响 WPE 输出,从而可能导致去混响有限。本文中,我们提出一种专用于在线处理的端到端方法,直接优化去混响输出信号。此外,我们提出通过修改优化目标以及训练中使用的 WPE 算法特性,使其适应不同类型助听设备用户的需求。我们表明所提端到端方法在 WHAMR! 数据集的无噪声版本上优于传统及常规 DNN 支持的 WPEs。
引用
@article{arxiv.2204.02694,
title = {Customizable End-to-end Optimization of Online Neural Network-supported Dereverberation for Hearing Devices},
author = {Jean-Marie Lemercier and Joachim Thiemann and Raphael Koning and Timo Gerkmann},
journal= {arXiv preprint arXiv:2204.02694},
year = {2022}
}
备注
\copyright 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works