{"id":791942,"date":"2021-11-09T10:41:15","date_gmt":"2021-11-09T18:41:15","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-academic-program&p=791942"},"modified":"2023-10-04T13:46:36","modified_gmt":"2023-10-04T20:46:36","slug":"deep-noise-suppression-challenge-icassp-2022","status":"publish","type":"msr-academic-program","link":"https:\/\/www.microsoft.com\/en-us\/research\/academic-program\/deep-noise-suppression-challenge-icassp-2022\/","title":{"rendered":"Deep Noise Suppression Challenge \u2013 ICASSP 2022"},"content":{"rendered":"\n\n

<\/p>\n\n\n\n\n\n\n

Program dates:<\/strong> December 2021\u2013February 2022<\/p>\n\n\n\n

Noise suppression has become more important than ever before due to the increasing use of voice interfaces for various applications. Given\u202fthe\u202fmillions of internet-connected devices being employed for audio\/video calls, noise suppression is expected to be effective for all noise types chosen from daily-life scenarios.\u202fThe\u202fIEEE ICASSP 2022\u202fGrand\u202fChallenge is the 4th DNS\u202fchallenge intended to promote industry-academia collaboration on research in real-time noise suppression aimed to maximize the subjective (perceptual) quality of enhanced speech.\u202fThis challenge will extend DNS efforts to\u202ffull band\u202fspeech with a special focus on personalized denoising. In\u202fthe era of hybrid work, personalized denoising is very important to suppress neighboring speaker and\/or background noises.\u202fRecently, DNS research\u202fhas been\u202fmoving fast, and researchers now have state-of-the-art advancements\u202fin deep neural networks (DNNs);\u202fcurrently, deep noise suppression methods leverage the convolutional, recurrent, or hybrid neural network for estimating the enhanced speech from noisy recordings.<\/p>\n\n\n\n

Previous editions of\u202fthe\u202fDNS Challenge provided researchers with a massive training dataset and real test set along with\u202fa\u202fP.808\/P.835\u202ftest framework\u202ffor subjective evaluation\u202fof enhanced speech. In\u202fthe\u202fcurrent\u202fchallenge, we improved the training dataset by cleaning it further and added some more data to\u202fcapture\u202frelevant\u202fDNS scenarios.\u202fWe\u202fcollected\u202fa new test set for\u202ffull band\u202fspeech ensuring high energy content in\u202fhigher\u202ffrequency bands\u202fto eliminate bandlimited clips from some devices. We\u202fincluded new noise types in the test set\u202fcovering contemporary\u202fscenarios and device variety, especially mobile scenarios. Our training data synthesizer script is flexible to allow the exclusion of any subset or addition of new data\u202fby the challenge\u202fparticipants.\u202fThis provides an opportunity for leveraging challenge data along with other corpora for improving DNS performance. Our test set consists of real-world test clips\u202frecorded by crowd-sourced workers and\/or Microsoft employees. We have two dev-test sets for real-time denoising and personalized real-time denoising. Similarly, we have two blind test sets, one for each challenge track.<\/p>\n\n\n\n

Challenge paper can be found ICASSP_2022_4th_Deep_Noise_Suppression_Challenge<\/a><\/p>\n\n\n\n

The tracks in this challenge are:<\/p>\n\n\n\n

Track 1:\u202fReal-Time\u202fnon-personalized DNS\u202ffor\u202ffull band\u202fspeech\u202f<\/b><\/p>\n\n\n\n