ConferencingSpeech 2022

会议语音2022挑战针对语音质量评估任务的非侵入性深度神经网络模型。我们开源了一个训练语料库，里面有超过86k的不同语言的语音片段，通过众包的方式有各种各样的合成和现场降格及其相应的主观质量分数。18个团队提交了他们的模型，以供本次挑战评估。盲测试集包括来自广泛降解范围的大约4300个剪辑。The 2022 Speech Conference Challenge focuses on non-intrusive deep neural network models for the speech quality assessment task. We have open-sourced a training corpus containing over 86,000 speech clips in diverse languages, which features a variety of synthetic and in-situ degraded speech samples paired with their corresponding subjective quality scores collected through crowdsourcing. A total of 18 teams submitted their models for evaluation in this challenge. The blind test set includes approximately 4,300 clips spanning a wide range of degradation degrees.

搜集汇总

背景与挑战

背景概述

ConferencingSpeech 2022是一个用于语音质量评估的非侵入性深度神经网络模型训练数据集，包含超过86k个多语言语音片段及其主观质量分数。该数据集支持2022年会议语音挑战的评估，盲测试集涵盖约4300个剪辑，由柏林工业大学、印第安纳大学布卢明顿分校和腾讯天籁实验室于2022年发布。