Search papers, labs, and topics across Lattice.
StuPASE improves generative speech enhancement by finetuning the PASE model with dry targets to enhance dereverberation and replacing the GAN-based generative module with a flow-matching module to improve performance under strong additive noise. This approach maintains the low-hallucination properties of PASE while achieving studio-level perceptual quality, even in challenging conditions. Experiments show StuPASE outperforms state-of-the-art speech enhancement methods in both perceptual quality and hallucination control.
Studio-quality speech enhancement without hallucination is now possible, thanks to a clever combination of dry-target finetuning and flow-matching.
Achieving high perceptual quality without hallucination remains a challenge in generative speech enhancement (SE). A representative approach, PASE, is robust to hallucination but has limited perceptual quality under adverse conditions. We propose StuPASE, built upon PASE to achieve studio-level quality while retaining its low-hallucination property. First, we show that finetuning PASE with dry targets rather than targets containing simulated early reflections substantially improves dereverberation. Second, to address performance limitations under strong additive noise, we replace the GAN-based generative module in PASE with a flow-matching module, enabling studio-quality generation even under highly challenging conditions. Experiments demonstrate that StuPASE consistently produces perceptually high-quality speech while maintaining low hallucination, outperforming state-of-the-art SE methods. Audio demos are available at: https://xiaobin-rong.github.io/stupase_demo/.
