Universal Speech Enhancement with Score-based Diffusion

计算机科学 语音增强 可理解性(哲学) 语音识别 人工智能 降噪 哲学 认识论
作者
Joan Serrà,Santiago Pascual,Jordi Pons,Recep Oğuz Araz,Davide Scaini
出处
期刊:Cornell University - arXiv 被引量:43
标识
DOI:10.48550/arxiv.2206.03065
摘要

Removing background noise from speech audio has been the subject of considerable effort, especially in recent years due to the rise of virtual communication and amateur recordings. Yet background noise is not the only unpleasant disturbance that can prevent intelligibility: reverb, clipping, codec artifacts, problematic equalization, limited bandwidth, or inconsistent loudness are equally disturbing and ubiquitous. In this work, we propose to consider the task of speech enhancement as a holistic endeavor, and present a universal speech enhancement system that tackles 55 different distortions at the same time. Our approach consists of a generative model that employs score-based diffusion, together with a multi-resolution conditioning network that performs enhancement with mixture density networks. We show that this approach significantly outperforms the state of the art in a subjective test performed by expert listeners. We also show that it achieves competitive objective scores with just 4-8 diffusion steps, despite not considering any particular strategy for fast sampling. We hope that both our methodology and technical contributions encourage researchers and practitioners to adopt a universal approach to speech enhancement, possibly framing it as a generative task.

科研通智能强力驱动
Strongly Powered by AbleSci AI
更新
大幅提高文件上传限制,最高150M (2024-4-1)

科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
刚刚
1秒前
调研昵称发布了新的文献求助10
1秒前
FashionBoy应助高妖丽采纳,获得10
2秒前
2秒前
小研发布了新的文献求助10
3秒前
jincheng发布了新的文献求助10
4秒前
Atobe发布了新的文献求助10
5秒前
WQR发布了新的文献求助10
7秒前
8秒前
9秒前
深情安青应助乐观的天问采纳,获得10
9秒前
完美世界应助LJJ采纳,获得10
9秒前
自信书蕾完成签到,获得积分20
10秒前
qq完成签到,获得积分10
10秒前
10秒前
兜兜揣满糖完成签到 ,获得积分10
10秒前
11秒前
林大侠完成签到,获得积分10
11秒前
12秒前
852应助鲍鲍采纳,获得10
12秒前
bu应助感动期待采纳,获得10
12秒前
qq发布了新的文献求助10
13秒前
13秒前
13秒前
14秒前
14秒前
子默发布了新的文献求助10
15秒前
糊涂涂完成签到,获得积分10
15秒前
16秒前
17秒前
LL发布了新的文献求助10
18秒前
19秒前
外向以云完成签到 ,获得积分10
20秒前
20秒前
暗夜星辰完成签到,获得积分0
21秒前
高妖丽发布了新的文献求助10
21秒前
可爱半凡发布了新的文献求助10
22秒前
zeb完成签到,获得积分10
24秒前
高分求助中
Licensing Deals in Pharmaceuticals 2019-2024 3000
Cognitive Paradigms in Knowledge Organisation 2000
Effect of reactor temperature on FCC yield 2000
How Maoism Was Made: Reconstructing China, 1949-1965 800
Introduction to Spectroscopic Ellipsometry of Thin Film Materials Instrumentation, Data Analysis, and Applications 600
Promoting women's entrepreneurship in developing countries: the case of the world's largest women-owned community-based enterprise 500
Shining Light on the Dark Side of Personality 400
热门求助领域 (近24小时)
化学 医学 生物 材料科学 工程类 有机化学 生物化学 物理 内科学 纳米技术 计算机科学 化学工程 复合材料 基因 遗传学 催化作用 物理化学 免疫学 量子力学 细胞生物学
热门帖子
关注 科研通微信公众号,转发送积分 3310243
求助须知:如何正确求助?哪些是违规求助? 2943212
关于积分的说明 8513174
捐赠科研通 2618448
什么是DOI,文献DOI怎么找? 1431076
科研通“疑难数据库(出版商)”最低求助积分说明 664359
邀请新用户注册赠送积分活动 649542