Beyond U-Net: A Latent-Representation-Aligned Skip-Free Backbone for Flow-Matching Speech Enhancement
For speech enhancement researchers, this work offers a more efficient alternative to U-Net-based generative models, improving quality while maintaining few-step inference.
This paper proposes a skip-free encoder-decoder backbone for flow-matching speech enhancement, using Latent Representation Alignment (LRA) to align bottleneck and decoder representations with clean latent features from a frozen Descript Audio Codec. The method achieves improved PESQ and perceptual quality on WSJ0-CHiME3 and VoiceBank-DEMAND with only five function evaluations.
Generative models, particularly diffusion and score-based approaches, have recently achieved strong performance in speech enhancement, but their iterative sampling process limits real-time deployment. Flow Matching offers an efficient alternative by transporting noisy speech toward clean speech through an ordinary differential equation with few function evaluations. In this work, we propose a skip-free encoder-decoder backbone for flow-matching speech enhancement, guided by Latent Representation Alignment (LRA). Instead of relying on U-Net skip connections, which may transfer noise-correlated low-level features to the decoder, the proposed model aligns its bottleneck and decoder representations with clean latent features extracted from a frozen Descript Audio Codec encoder-decoder without quantization. This codec-aligned supervision promotes compact clean-speech representations while preserving efficient few-step inference. Experiments on WSJ0-CHiME3 and VoiceBank-DEMAND show improved PESQ and perceptual quality, especially on VoiceBank-DEMAND, using only five function evaluations.