Gaussian Flow Bridges for Audio Domain Transfer with Unpaired Data

TL;DR

This paper introduces Gaussian Flow Bridges, a novel unsupervised generative approach for audio domain transfer that manipulates audio characteristics without requiring paired data, showing promising results in reverberation and distortion tasks.

Contribution

The paper presents a new framework using Gaussian Flow Bridges for unpaired audio domain transfer, enabling continuous control over target domain properties.

Findings

Competitive performance in reverberation manipulation

Effective distortion modification without paired data

Potential for further research in unsupervised audio transfer

Abstract

Audio domain transfer is the process of modifying audio signals to match characteristics of a different domain, while retaining the original content. This paper investigates the potential of Gaussian Flow Bridges, an emerging approach in generative modeling, for this problem. The presented framework addresses the transport problem across different distributions of audio signals through the implementation of a series of two deterministic probability flows. The proposed framework facilitates manipulation of the target distribution properties through a continuous control variable, which defines a certain aspect of the target domain. Notably, this approach does not rely on paired examples for training. To address identified challenges on maintaining the speech content consistent, we recommend a training strategy that incorporates chunk-based minibatch Optimal Transport couplings of data samples and noise. Comparing our unsupervised method with established baselines, we find competitive performance in tasks of reverberation and distortion manipulation. Despite encoutering limitations, the intriguing results obtained in this study underscore potential for further exploration.