mlsanityData to 7 Oct 2026

StuPASE: Towards Low-Hallucination Studio-Quality Generative Speech Enhancement

arXiv 2603.09234 ↗First posted 10 Mar 2026eess.AS cs.AI cs.SDcited by 0

Not on the map. Why?

Abstract (arXiv)

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/.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 36 of its 50 references that are in the map. First posted on arXiv 10 Mar 2026.

Builds on

36 of 50 references.

References · 36

  1. 2022Scalable Diffusion Models with Transformerscited by 5,931
  2. 2022Flow Matching for Generative Modelingcited by 5,426
  3. 2022Robust Speech Recognition via Large-Scale Weak Supervisioncited by 3,972
  4. 2019Common Voice: A Massively-Multilingual Speech Corpuscited by 1,358
  5. 2020HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesiscited by 1,139
  6. 2019LibriTTS: A Corpus Derived from LibriSpeech for Text-to-Speechcited by 639
  7. 2022UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022cited by 381
  8. 2024F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matchingcited by 353
  9. 2020FSD50K: An Open Dataset of Human-Labeled Sound Eventscited by 339
  10. 2022BigVGAN: A Universal Neural Vocoder with Large-Scale Trainingcited by 245
  11. 2016FMA: A Dataset For Music Analysiscited by 205
  12. 2017SEGAN: Speech Enhancement Generative Adversarial Networkcited by 165
24 more
  1. 2021DNSMOS P.835: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressorscited by 165
  2. 2019WHAM!: Extending Speech Separation to Noisy Environmentscited by 154
  3. 2023Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesiscited by 150
  4. 2024E2 TTS: Embarrassingly Easy Fully Non-Autoregressive Zero-Shot TTScited by 149
  5. 2022Conditional Diffusion Probabilistic Model for Speech Enhancementcited by 132
  6. 2020The INTERSPEECH 2020 Deep Noise Suppression Challenge: Datasets, Subjective Testing Framework, and Challenge Resultscited by 128
  7. 2024WavTokenizer: an Efficient Acoustic Discrete Codec Tokenizer for Audio Language Modelingcited by 124
  8. 2022StoRM: A Diffusion-based Stochastic Regeneration Model for Speech Enhancement and Dereverberationcited by 105
  9. 2021Large-scale Self-Supervised Speech Representation Learning for Automatic Speaker Verificationcited by 71
  10. 2024Self-Supervised Speech Representations are More Phonetic than Semanticcited by 70
  11. 2022TF-GridNet: Integrating Full- and Sub-Band Modeling for Speech Separationcited by 62
  12. 2023ICASSP 2023 Deep Noise Suppression Challengecited by 62
  13. 2024SpeechBERTScore: Reference-Aware Automatic Evaluation of Speech Generation Leveraging NLP Evaluation Metricscited by 38
  14. 2025AnyEnhance: A Unified Generative Model with Prompt-Guidance and Self-Critic for Voice Enhancementcited by 23
  15. 2025GenSE: Generative Speech Enhancement via Language Models using Hierarchical Modelingcited by 20
  16. 2025LLaSE-G1: Incentivizing Generalization Capability for LLaMA-based Speech Enhancementcited by 19
  17. 2024URGENT Challenge: Universality, Robustness, and Generalizability For Speech Enhancementcited by 17
  18. 2025FlowSE: Efficient and High-Quality Speech Enhancement via Flow Matchingcited by 17
  19. 2021DPCRN: Dual-Path Convolution Recurrent Network for Single Channel Speech Enhancementcited by 13
  20. 2025Interspeech 2025 URGENT Speech Enhancement Challengecited by 12
  21. 2022Speech Dereverberation with A Reverberation Time Shortening Targetcited by 10
  22. 2025PASE: Leveraging the Phonological Prior of WavLM for Low-Hallucination Generative Speech Enhancementcited by 5
  23. 2025SenSE: Semantic-Aware High-Fidelity Universal Speech Enhancementcited by 3
  24. 2025A Semantic Information-based Hierarchical Speech Enhancement Method Using Factorized Codec and Diffusion Modelcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025DiTSE: High-Fidelity Generative Speech Enhancement via Latent Diffusion Transformers
  2. 2023uSee: Unified Speech Enhancement and Editing with Conditional Diffusion Models
  3. 2025Aligning Generative Speech Enhancement with Perceptual Feedback
  4. 2026LIFT-SE: Linguistic Inference Followed by Flow Transformation for Generative Speech Enhancement ↗
  5. 2026HalluAudio: A Comprehensive Benchmark for Hallucination Detection in Large Audio-Language Models
  6. 2022Speech Enhancement and Dereverberation with Diffusion-based Generative Models
  7. 2025MeanFlowSE: One-Step Generative Speech Enhancement via MeanFlow
  8. 2025MeanSE: Efficient Generative Speech Enhancement with Mean Flows