SwinDS-BWE: A Parameter-Efficient Swin-1D Lattice Generator with Decision-Science Discriminators for Speech Bandwidth Extension

Tarikul Islam Tamiti; Sajid Fardin Dipto; Luke Baja-Ricketts; David Vergano; Anomadarshi Barua

cs.SDOct 6, 2026

SwinDS-BWE: A Parameter-Efficient Swin-1D Lattice Generator with Decision-Science Discriminators for Speech Bandwidth Extension

Authors: Tarikul Islam Tamiti, Sajid Fardin Dipto, Luke Baja-Ricketts, David Vergano, Anomadarshi Barua

Organizations: Department of Cyber Security Engineering, George Mason University, Virginia, USA

Abstract

We propose SwinDS-BWE, a decision-science-inspired bandwidth extension (BWE) model with two coupled contributions: (1) Swin Transformers with lattice-style cross-stream interaction yield locality-aware modeling with linear-in-sequence per-window attention cost and reduce the prior AP-BWE generator size 0.5x from 33M to 17M. (2) Three novel lightweight decision-science-inspired discriminators augment AP-BWE's performance: a CVaR discriminator tail-pools activations to emphasize worst high-frequency (HF) segments, a Chance-Constraint HF discriminator penalizes excessive HF energy via a differentiable barrier, and a Multi-Criteria Utility discriminator learns convex style weights over spectral criteria. SwinDS-BWE surpasses prior AP-BWE with a 30x smaller discriminator (42.3M vs. 1.36M) and higher fidelity on English and French datasets. This work shows that decision-science-inspired critics can supervise BWE to reduce discriminator size.

Figures & tables

Figure 1: Our proposed CVaRD, CCD, and MCUD discriminators. (Right corner) Results of subjective MOS.

Method	Size	Data	NISQA-MOS $\uparrow$			STOI $\uparrow$			PESQ $\uparrow$			SI-SDR $\uparrow$			SI-SNR $\uparrow$			LSD $\downarrow$			WER % $\downarrow$
Method	Size	Data	4–16	8–16	16–48	4–16	8–16	16–48	4–16	8–16	16–48	4–16	8–16	16–48	4–16	8–16	16–48	4–16	8–16	16–48	4–16	8–16	16–48
Unprocessed	-	VCTK clean	2.79	3.67	4.43	0.55	0.61	0.61	1.15	1.51	1.41	-11.03	-8.07	-6.07	-10.53	-7.62	-5.63	3.27	2.27	2.85	90.0	6.1	2.1
Unprocessed	-	MLS clean	2.29	3.12	-	0.51	0.57	-	1.05	1.34	-	-14.15	-18.89	-	-13.67	-17.71	-	3.48	2.48	-	91.1	6.8	-
EBEN, 2023	29.7M	VCTK clean	2.59	2.69	2.53	0.89	0.98	0.98	2.64	3.69	3.71	11.94	19.94	20.82	11.94	19.94	20.83	1.03	0.78	0.92	19.1	12.4	8.5
EBEN, 2023	29.7M	MLS clean	2.14	2.28	-	0.87	0.96	-	2.47	3.55	-	11.81	18.24	-	11.74	18.38	-	1.17	0.88	-	19.4	12.7	-
AERO, 2023	36.4M	VCTK clean	2.79	2.75	2.88	0.83	0.94	0.99	2.62	3.65	3.69	13.60	20.70	21.56	13.60	20.70	21.56	1.09	0.97	0.75	20.3	13.4	9.8

Table 1: Comparison for clean English-VCTK and MLS French dataset (MLS does not have 48 kHz speech).

Addition of Single Decision-Science Discriminators
SL	MRAD	MRPD	CVaRD	CCD	MCUD	LSD	STOI	PESQ	SNR	N-MOS
1	✗	✗	✓	✗	✗	1.06	0.94	2.32	12.89	3.36
2	✗	✗	✗	✓	✗	1.25	0.95	2.61	14.35	2.99
3	✗	✗	✗	✗	✓	1.18	0.95	2.63	14.2	3.01
Decision Science Discriminators with MRAD and MRPD Pair
4	✗	✗	✓	✓	✓	1.06	0.94	2.27	11.8	3.23

Table 2: Ablation of decision-science discriminators for 4

\rightarrow

16 kHz. Here, N-MOS = NISQA-MOS and SNR = SI-SNR.

Freq. range	LSD	STOI	PESQ	SI-SDR	SI-SNR	NISQA-MOS
2-16 kHz	1.11	0.87	1.61	7.63	7.62	4.13
2-48 kHz	1.13	0.83	1.19	7.6	7.59	4.01
4-16 kHz	0.99	0.94	2.34	12.76	12.68	4.11
8-16 kHz	0.77	0.998	3.69	17.59	17.43	4.29
8-48 kHz	0.93	0.996	2.42	15.53	15.42	4.52
12-48 kHz	0.85	0.998	3.21	18.01	17.91	4.51

Table 3: Performance over different frequency ranges.

Model	Fq. Range	GP (M)	DP (M)	MACs (M)	FLOPs (M)	RTF (GPU)
AP-BWE	4-16 kHz	29.76	42.3	14236.65	28473.31	0.0023x
AP-BWE	16-48 kHz	29.76	42.3	14236.65	28473.31	0.0025x
SwinDS-BWE	4-16 kHz	17.09	1.36	5277.35	10554.69	0.0032x
SwinDS-BWE	16-48 kHz	17.09	1.36	5277.35	10554.69	0.0034x

Table 4: Computational complexity of SwinDS-BWE.

Explore similar work

UDSS-BWE: Uncertainty- and Decision-Science Inspired Swin BandWidth Extension
Sep 28, 2026Tarikul Islam Tamiti, Sajid Fardin Dipto, David Vergano +2Bandwidth Extension Perceptual Quality
Department of Cyber Security Engineering, George Mason University, USA.
AnyBand: Unified Multi-Bandwidth Speech Extension via Frequency-Aware In-Context Spectral Infilling
Aug 1, 2026Junchuan Zhao, Minh Duc Vu, Bowen Zhang +1Speech Enhancement Bandwidth Extension
School of Computing, National University of Singapore · Department of Statistics & Data Science, National University of Singapore · College of Computing and Data Science, Nanyang Technological University
A Survey of Advancing Audio Super-Resolution and Bandwidth Extension from Discriminative to Generative Models
May 15, 2026Ningyuan Yang, Yize Li, Diego A. Cuji +4Generative Super-Resolution Real-World Image Super-Resolution
Department of Electrical and Computer Engineering Stony Brook University · Department of Electrical and Computer Engineering Northeastern University · Department of Electrical and Computer Engineering University of Illinois Chicago & Discovery Partners Institute

Sep 28, 2026, cs.SDJ/K move · Enter open · S save

UDSS-BWE: Uncertainty- and Decision-Science Inspired Swin BandWidth Extension

Tarikul Islam Tamiti, Sajid Fardin Dipto, David Vergano, Luke Baja-Ricketts+1

Department of Cyber Security Engineering, George Mason University, USA.

Bandwidth Extension Perceptual Quality High-Frequency Reconstruction Uncertainty

Open paper AskPDF arXiv

Bandwidth extension (BWE) is fundamentally localized: the most perceptual distortions are not average-case distortions, but rare high-frequency (HF) transients that standard, risk-neutral objectives tend to smooth away. To close this gap, we seek solutions in the risk-sensitive and uncertainty-aware decision science rules and present UDSS-BWE, which introduces five decision-science and uncertainty-aware discriminators: CVaRD (does tail pooling to amplify HF artifacts), CCD (a primal-dual augmented Lagrangian to prevent HF overboost), MCUD (a learnable utility over spectral flatness/ centroid/ rolloff), EDD (captures epistemic uncertainty), and DROD (captures entropic KL-DRO aggregation). UDSS-BWE is also designed as a complex valued adversarial BWE framework that uses Swin-based generators, a lightweight dual-stream shifted-window backbone, to capture local and long-range structure efficiently, while learnable lattice coupling provides controlled cross-stream exchange. UDSS-BWE is optimized extensively and achieves better perceptual quality with 3.89x fewer parameters (72M vs.18.5M) over two English and French datasets under clean and noisy conditions. To the best of our knowledge, this work shows how multi disciplinary decision-science-inspired and uncertainty theories can be successfully used to design efficient discriminators for producing more nuanced audios, establishing a new baseline in the BWE task.

Abstract

Explore similar work

UDSS-BWE: Uncertainty- and Decision-Science Inspired Swin BandWidth Extension

AnyBand: Unified Multi-Bandwidth Speech Extension via Frequency-Aware In-Context Spectral Infilling

A Survey of Advancing Audio Super-Resolution and Bandwidth Extension from Discriminative to Generative Models