Audio Signal Enhancement with Learning from Positive and Unlabelled Data

TL;DR

This paper proposes a novel audio signal enhancement method that learns from positive and unlabelled data, avoiding the need for clean signals and using real noisy recordings for improved real-world performance.

Contribution

It introduces a positive and unlabelled data learning framework for audio enhancement, bypassing the need for synthetic clean-noisy pairs and enabling training on real data.

Findings

Effective separation of signal activity and inactivity in spectrograms.

Improved performance on real noisy data compared to traditional supervised methods.

Utilization of noise clips as positive data for training.

Abstract

Supervised learning is a mainstream approach to audio signal enhancement (SE) and requires parallel training data consisting of both noisy signals and the corresponding clean signals. Such data can only be synthesised and are mismatched with real data, which can result in poor performance on real data. Moreover, clean signals may be inaccessible in certain scenarios, which renders this conventional approach infeasible. Here we explore SE using non-parallel training data consisting of noisy signals and noise, which can be easily recorded. We define the positive (P) and the negative (N) classes as signal inactivity and activity, respectively. We observe that the spectrogram patches of noise clips can be used as P data and those of noisy signal clips as unlabelled data. Thus, learning from positive and unlabelled data enables a convolutional neural network to learn to classify each spectrogram patch as P or N to enable SE.