Compute and memory efficient universal sound source separation

Abstract

Recent progress in audio source separation lead by deep learning has enabled many neural network models to provide robust solutions to this fundamental estimation problem. In this study, we provide a family of efficient neural network architectures for general purpose audio source separation while focusing on multiple computational aspects that hinder the application of neural networks in real-world scenarios. The backbone structure of this convolutional network is the SUccessive DOwnsampling and Resampling of Multi-Resolution Features (SuDoRM-RF) as well as their aggregation which is performed through simple one-dimensional convolutions. This mechanism enables our models to obtain high fidelity signal separation in a wide variety of settings where variable number of sources are present and with limited computational resources (e.g. floating point operations, memory footprint, number of parameters and latency). Our experiments show that SuDoRM-RF models perform comparably and even surpass several state-of-the-art benchmarks with significantly higher computational resource requirements. The causal variation of SuDoRM-RF is able to obtain competitive performance in real-time speech separation of around 10dB scale-invariant signal-to-distortion ratio improvement (SI-SDRi) while remaining up to 20 times faster than real-time on a laptop device.

Few-Shot Learning of New Sound Classes for Target Sound Extraction

“…In future works, we will consider training the system on a larger dataset, which may include audio clips with multiple AE classes, or unlabeled data [8,9]. Moreover, we will also investigate other EL [13] to provide more discriminative embedding vectors and extend the model to perform online processing [30].…”

Separate But Together: Unsupervised Federated Learning for Speech Enhancement from Non-IID Data

“…We force the estimated sources to add up to the input mixture using a mixture consistency layer [28] at the output of our separation model. For all the other parameters we choose the default settings provided in [29] for a sampling rate of 16k Hz.…”