EARS: An Anechoic Fullband Speech Dataset Benchmarked for Speech Enhancement and Dereverberation

TL;DR

The paper introduces the EARS dataset, a comprehensive high-quality anechoic speech dataset with diverse speaking styles, and benchmarks various speech enhancement and dereverberation methods using instrumental metrics and listening tests.

Contribution

It provides a new large-scale, diverse speech dataset and establishes benchmark evaluations for speech enhancement and dereverberation methods.

Findings

Generative methods are preferred in listening tests.

Benchmarking reveals strengths and weaknesses of current methods.

The dataset enables automatic online evaluation.

Abstract

We release the EARS (Expressive Anechoic Recordings of Speech) dataset, a high-quality speech dataset comprising 107 speakers from diverse backgrounds, totaling in 100 hours of clean, anechoic speech data. The dataset covers a large range of different speaking styles, including emotional speech, different reading styles, non-verbal sounds, and conversational freeform speech. We benchmark various methods for speech enhancement and dereverberation on the dataset and evaluate their performance through a set of instrumental metrics. In addition, we conduct a listening test with 20 participants for the speech enhancement task, where a generative method is preferred. We introduce a blind test set that allows for automatic online evaluation of uploaded data. Dataset download links and automatic evaluation server can be found online.