BLINC: Blind Calibration For Training-Free Speech Enhancement Adaptation

Tobias Raichle; Ekaterina Gavrilko; Bin Yang

eess.ASSep 18, 2026

BLINC: Blind Calibration For Training-Free Speech Enhancement Adaptation

Authors: Tobias Raichle, Ekaterina Gavrilko, Bin Yang

Organizations: University of Stuttgart, Institute of Signal Processing and System Theory, Stuttgart, Germany

Abstract

Speech enhancement (SE) models degrade under domain shifts and have to adapt to unseen target domains during deployment. Most existing test-time adaptation (TTA) methods for SE do so by adapting a subset of the model weights using a self-supervised loss, which requires backpropagation at test-time and permanently alters the model. We instead recalibrate the prediction itself and propose BLINC, a training-free TTA method that remaps the predicted time-frequency mask onto a bimodal target distribution by histogram matching. At test-time, the target distribution is parameterized from blind features of the noisy recording, so neither a reference distribution from a classical algorithm nor online metric optimization is involved. BLINC improves the overall quality of both evaluated SE models on almost every target condition and matches or exceeds the loss-based TTA baselines at minimal overhead.

Figures & tables

Panel of Figure 1 — Figure 1: Loss-based TTA minimizes a loss $\mathcal{L}$ at test-time, adjusting either the model weights $\theta$ or the prediction $\widehat{\mathbf{X}}$ . BLINC instead recalibrates the prediction of the frozen model in a single step without a loss.

Panel of Figure 4 — Figure 4: Signal-wise optimal midpoint $b$ over the blind speech-activity fraction $P_{\mathrm{Sp}}$ .

Source	RemixIT	LaDen	MPol	RaR	BLINC
0.003	0.014	0.372	0.032	0.903	0.003

Table 1: RTF

\downarrow

with the AM model on VBD. Factors

\leq 1

indicate real-time capability.

		$\overline{\text{PESQ}}$ $\uparrow$	$\overline{\text{CSIG}}$ $\uparrow$	$\overline{\text{CBAK}}$ $\uparrow$	$\overline{\text{COVL}}$ $\uparrow$	$\overline{\text{SSNR}}$ $\uparrow$	$\overline{\text{SI-SDR}}$ $\uparrow$
AM	Source	$2.05$	$3.07$	$2.78$	$2.52$	$7.42$	$12.28$
	RemixIT	$2.06$	$3.10$	$2.80$	$2.54$	7.48	12.47
	LaDen	$2.13$	$3.13$	$2.80$	$2.59$	$7.01$	$12.33$
	MPol	$2.10$	$3.17$	2.82	$2.60$	$7.40$	$12.35$
	BLINC	2.17	3.21	$2.78$	2.66	$6.19$	$11.50$
CMGAN	Source	$2.60$	$3.75$	$3.02$	$3.15$	$6.00$	$11.32$

Table 2: Evaluation results averaged over the target datasets. SSNR and SI-SDR in dB.

Panel of Figure 5 — Figure 5: $\Delta$ COVL per target domain relative to the source model.

Target	PESQ $\uparrow$	SI-SDR $\uparrow$
Source	2.42	11.49
Constant target	2.45	10.73
BLINC	2.55	11.24
BLINC, SI-SDR	2.29	12.21
BLINC, in-domain	2.65	11.02
Sigmoid oracle	2.75	11.02

Table 3: Ablation study with the AM model on VBD. The lower block uses information unavailable at test-time. SI-SDR in dB.

Speech Enhancement

Speech enhancement improves intelligibility and listening quality by suppressing noise, reverberation, and competing speakers in recorded or live audio.

See all 127 papers

Test-time adaptation for speech enhancement with an autoregressive speech prior
Sep 3, 2026Sofiene Kammoun, Simon Leglaive, Xavier Alameda-Pineda +1Test-Time Adaptation Speech Enhancement
CentraleSup´elec, IETR (UMR CNRS 6164), France · Inria at Univ. Grenoble Alpes, CNRS, LJK, France · Signal Processing Group, University of Hamburg, Germany
Towards Lightweight Adaptation of Speech Enhancement Models in Real-World Environments
Mar 8, 2026Longbiao Cheng, Shih-Chii LiuSpeech Enhancement Memory-Efficient Fine-Tuning
Institute of Neuroinformatics, University of Zurich and ETH Zurich
Domain-Incremental Learning for Generative Speech Enhancement
Sep 28, 2026Manjunath Mulimani, Annamaria Mesaros, Minje Kim +1Domain-Incremental Learning Speech Enhancement
Aalborg University, Department of Electronic Systems, Denmark · Tampere University, Signal Processing Research Centre, Finland · University of Illinois Urbana-Champaign, Siebel School of Computing and Data Science, USA

Sep 3, 2026J/K move · Enter open · S save

Test-time adaptation for speech enhancement with an autoregressive speech prior

Sofiene Kammoun, Simon Leglaive, Xavier Alameda-Pineda, Timo Gerkmann

CentraleSup´elec, IETR (UMR CNRS 6164), France · Inria at Univ. Grenoble Alpes, CNRS, LJK, France · Signal Processing Group, University of Hamburg, Germany

Test-Time Adaptation Speech Enhancement

Open paper AskPDF arXiv

Test-time adaptation (TTA) offers a promising direction for improving speech enhancement models under mismatched acoustic conditions, without requiring access to labeled target data. In this work, we propose a single-utterance TTA method that regularizes a pretrained speech enhancement model using an autoregressive prior trained on clean speech latent representations extracted from a neural audio codec. Adaptation is performed by minimizing the Kullback-Leibler divergence between the enhanced speech distribution and the clean speech prior. Experiments across multiple noisy speech datasets show consistent improvements in speech quality, particularly under training-testing noise mismatch conditions. Code and audio examples are available online.

Abstract

Speech Enhancement

Explore similar work

Test-time adaptation for speech enhancement with an autoregressive speech prior

Towards Lightweight Adaptation of Speech Enhancement Models in Real-World Environments

Domain-Incremental Learning for Generative Speech Enhancement