# SNOAR: Self-training with Noise filtering, Observation adding, and Remixing for Speech Denoising Without Clean References

arXiv: https://arxiv.org/abs/2610.20863 · HTML: https://mlsanity.com/paper/2610.20863/

## Key facts

- No citing papers in the map yet (data through 19 Oct 2026).
- Builds on 20 of its 29 references that are in the map.
- First posted on arXiv 14 Oct 2026.

## Builds on (most cited first, 20 of 20)

- [SDR - half-baked or well done?](https://mlsanity.com/paper/1811.02508/) (2018)
- [UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022](https://mlsanity.com/paper/2204.02152/) (2022)
- [Pseudo-Labeling and Confirmation Bias in Deep Semi-Supervised Learning](https://mlsanity.com/paper/1908.02983/) (2019)
- [DNSMOS: A Non-Intrusive Perceptual Objective Speech Quality metric to evaluate Noise Suppressors](https://mlsanity.com/paper/2010.15258/) (2020)
- [WHAM!: Extending Speech Separation to Noisy Environments](https://mlsanity.com/paper/1907.01160/) (2019)
- [TF-GridNet: Integrating Full- and Sub-Band Modeling for Speech Separation](https://mlsanity.com/paper/2211.12433/) (2022)
- [Self-Training: A Survey](https://mlsanity.com/paper/2202.12040/) (2022)
- [How Bad Are Artifacts?: Analyzing the Impact of Speech Enhancement Errors on ASR](https://mlsanity.com/paper/2201.06685/) (2022)
- [Differentiable Consistency Constraints for Improved Deep Speech Enhancement](https://mlsanity.com/paper/1811.08521/) (2018)
- [VERSA: A Versatile Evaluation Toolkit for Speech, Audio, and Music](https://mlsanity.com/paper/2412.17667/) (2024)
- [Statistical Speech Enhancement Based on Probabilistic Integration of Variational Autoencoder and Non-Negative Matrix Factorization](https://mlsanity.com/paper/1710.11439/) (2017)
- [RemixIT: Continual self-training of speech enhancement models via bootstrapped remixing](https://mlsanity.com/paper/2202.08862/) (2022)
- [OWSM v4: Improving Open Whisper-Style Speech Models via Data Scaling and Cleaning](https://mlsanity.com/paper/2506.00338/) (2025)
- [Noisy-target Training: A Training Strategy for DNN-based Speech Enhancement without Clean Speech](https://mlsanity.com/paper/2101.08625/) (2021)
- [The CHiME-7 UDASE task: Unsupervised domain adaptation for conversational speech enhancement](https://mlsanity.com/paper/2307.03533/) (2023)
- [Unsupervised speech enhancement with diffusion-based generative models](https://mlsanity.com/paper/2309.10450/) (2023)
- [Continual self-training with bootstrapped remixing for speech enhancement](https://mlsanity.com/paper/2110.10103/) (2021)
- [Analysis of Noisy-target Training for DNN-based speech enhancement](https://mlsanity.com/paper/2211.01198/) (2022)
- [Objective and subjective evaluation of speech enhancement methods in the UDASE task of the 7th CHiME challenge](https://mlsanity.com/paper/2402.01413/) (2024)
- [Exploiting Noise Inseparability for Weakly-Supervised Discriminative Speech Denoising Using Noisy Targets](https://mlsanity.com/paper/2606.02327/) (2026)

## Abstract (arXiv)

Unsupervised speech denoising methods such as MixIT and noisy-target training can be trained directly on noisy recordings, avoiding the sim-to-real gap of synthetic training data, but remain well below supervised performance. Teacher--student self-training, as in RemixIT, narrows this gap by remixing a teacher's estimates into new mixtures that supervise a student; however, since the student is trained only toward the teacher's estimates, teacher errors, in particular over-suppressed speech, are never recovered and propagate as the teacher is updated from the student. We propose SNOAR, a self-training framework that corrects the teacher's estimates before use: the noise leaked into its speech estimate is removed by differential noise filtering (DNF), and part of the noisy observation is added back in to restore suppressed speech and attenuate artifacts. We show that the two corrections depend on each other, as without DNF the loop collapses to the identity mapping. Compared to RemixIT, SNOAR is more stable along training and outperforms it in signal-based, perceptual, and word error rate metrics, reaching 97% of the SI-SDR improvement of a fully-supervised topline on WHAM!; on real CHiME-3 recordings, it improves every metric of three different teachers, while RemixIT degrades their perceptual quality.