ICASSP 2023 Deep Noise Suppression Challenge

• DOI: 10.48550/arxiv.2303.11510
• Publication Date: 2023-01-01
• Authors: Harishchandra Dubey; Ashkan Aazami; Vishak Gopal; Babak Naderi; Sebastian Braun; Ross Cutler; Alex Ju; Mehdi Zohourian; Min Tang; Hannes Gamper; Mehrsa Golestaneh; Robert Aichner
• Keywords: Headset; SIGNAL (programming language)

Deep Speech Enhancement Challenge is the 5th edition of deep noise suppression (DNS) challenges organized at ICASSP 2023 Signal Processing Grand Challenges. DNS were during 2019-2023 to stimulate research in speech enhancement (DSE). Previous INTERSPEECH 2020, 2021, and 2022. From prior editions, we learnt that improving signal quality (SIG) challenging particularly presence simultaneously active interfering talkers noise. This challenge aims develop models for joint denosing, dereverberation talkers. When primary talker wears a headphone, certain acoustic properties their such as direct-to-reverberation (DRR), ratio (SNR) etc. make it possible suppress neighboring even without enrollment data talker. motivated us create two tracks this challenge: (i) Track-1 Headset; (ii) Track-2 Speakerphone. Both has fullband (48kHz) training testset, each testclips corresponding (10-30s duration) Each track invited submissions personalized non-personalized all which are evaluated through same subjective evaluation. Most submitted models, team winner both where best improvement 0.145 0.141 challenge's Score compared noisy blind testset.