Listen to What You Want: Neural Network-Based Universal Sound Selector

• DOI: 10.21437/interspeech.2020-2210
• Publication Date: 2020-10-27T09:22:11Z
• Authors: Tsubasa Ochiai; Marc Delcroix; Yuma Koizumi; Hiroaki Ito; Keisuke Kinoshita; Shoko Araki
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Sound (cs.SD); 02 engineering and technology; Computer Science - Sound; Machine Learning (cs.LG); 03 medical and health sciences; Audio and Speech Processing (eess.AS); FOS: Electrical engineering, electronic engineering, information engineering; 0202 electrical engineering, electronic engineering, information engineering; 0305 other medical science; Electrical Engineering and Systems Science - Audio and Speech Processing

Being able to control the acoustic events (AEs) which we want listen would allow development of more controllable hearable devices.This paper addresses AE sound selection (or removal) problems, that define as extraction suppression) all sounds belong one or multiple desired classes.Although this problem could be addressed with a combination source separation followed by classification, is sub-optimal way solving problem.Moreover, usually requires knowing maximum number sources, may not practical when dealing AEs.In paper, propose instead universal neural network enables directly select from mixture given user-specified target classes.The proposed framework can explicitly optimized simultaneously classes, independently sources in mixture.We experimentally show method achieves promising performance and generalized mixtures are unseen during training.