FlowTSE: Target Speaker Extraction with Flow Matching

TL;DR

FlowTSE introduces a novel generative target speaker extraction method using flow matching, achieving competitive results with simpler architecture and improved phase reconstruction via a specialized vocoder.

Contribution

This work presents FlowTSE, a simple conditional flow matching model for TSE, and a new vocoder for phase reconstruction, reducing complexity and enhancing performance.

Findings

FlowTSE matches or outperforms strong baselines on standard benchmarks.

The proposed vocoder improves phase reconstruction quality.

FlowTSE simplifies the TSE pipeline with competitive results.

Abstract

Target speaker extraction (TSE) aims to isolate a specific speaker's speech from a mixture using speaker enrollment as a reference. While most existing approaches are discriminative, recent generative methods for TSE achieve strong results. However, generative methods for TSE remain underexplored, with most existing approaches relying on complex pipelines and pretrained components, leading to computational overhead. In this work, we present FlowTSE, a simple yet effective TSE approach based on conditional flow matching. Our model receives an enrollment audio sample and a mixed speech signal, both represented as mel-spectrograms, with the objective of extracting the target speaker's clean speech. Furthermore, for tasks where phase reconstruction is crucial, we propose a novel vocoder conditioned on the complex STFT of the mixed signal, enabling improved phase estimation. Experimental results on standard TSE benchmarks show that FlowTSE matches or outperforms strong baselines.