High-frequency reconstruction in bandwidth extension (BWE) depends on joined interactions among nonlinear spectral, temporal, and perceptual features; assigning a separate discriminator to each such feature may be redundant and may fail to model their joint interactions. We introduce quantum-informed generative model Q-BWE, where we design a five-qubit Quantum Chaos Discriminator (QCD) instead of using separate discriminator for every features, for modeling nonlinear joined-feature interactions efficiently. Our QCD explores a compact five-qubit quantum-inspired inductive bias based on data re-uploading circuit, trainable controlled NOT (CNOT) rotations, quantum entangling gates, and statevector measurement map. Five features for five qubits are selected from a 26-dimensional speech-dynamics bank and are encoded by a two-layer statevector circuit, producing 31 measurements map of local behavior, correlations, mixedness, and Born entropy. QCD guides a Lattice-ConformerNeXt generator with FiLM conditioning and a cosine-masked residual pathway, while delayed-reward Thompson sampling adaptively controls its contribution. The circuit is simulated on classical hardware. Overall, Q-BWE achieves competitive performance on VCTK English, and MLS French datasets in multilingual and multi speaker setups under clean and noisy conditions across a broad set of evaluation metrics with approximately 1.72x fewer parameters (33.5M vs. 19.4M) compared to recent CIS-BWE baseline published in ACL.
Figures & tables
Figure 1: Overall Q-BWE architecture with FiLM conditioning and raised cosine mask and QCD.
Figure 2: Overall QCD architecture: feature-bank selection, normalization, bounded angle encoding, simulated statevector map, and the scorer and matching heads.
Method
Size
Data
NISQA-MOS ↑
STOI ↑
PESQ ↑
SI-SDR ↑
SI-SNR ↑
LSD ↓
WER % ↓
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
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
MLS
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
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
MLS
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
–
AP-BWE, 2024
72M
VCTK
3.86
3.97
4.49
0.94
0.99
0.99
2.55
3.69
3.72
13.42
18.26
20.86
13.35
18.07
20.74
0.96
0.74
0.75
13.7
4.2
1.7
Table 1: Comparison with baseline on clean English VCTK and French MLS. MLS does not contain 48 kHz speech.
Figure 3: Qualitative comparison at 4–16 kHz.
Row
System
N-MOS ↑
STOI ↑
PESQ ↑
SI-SNR ↑
LSD ↓
\scriptsize1⃝
No FiLM
4.15
0.95
2.83
13.09
1.03
\scriptsize2⃝
No-Mask
4.39
0.95
2.72
12.85
0.97
\scriptsize3⃝
No QCD (only MRAD+MRPD)
4.15
0.95
2.85
13.15
1.01
\scriptsize4⃝
No FiLM & QCD
4.20
0.95
2.81
13.05
0.98
\scriptsize5⃝
No Entanglement.
4.27
0.95
2.76
13.05
0.99
\scriptsize6⃝
XYZ-only
4.2
0.95
2.83
13.14
0.98
Table 2: Ablation study on VCTK at 4 → 16 kHz.
SL
QCD inputs
STOI ↑
PESQ ↑
SI-SNR ↑
N-MOS ↑
\scriptsize1⃝
5 features
0.9478
2.735
13.27
3.99
\scriptsize2⃝
7 features
0.9462
2.731
13.11
4.06
Table 3: 5 and 7 qubit ablation for 4 → 16 kHz BWE.
SL
Metric
EBEN
AERO
AP-BWE
Q-BWE ZS
Q-BWE Matched
\scriptsize1⃝
NISQA-MOS ↑
1.01
1.52
2.74
1.466
1.854
\scriptsize2⃝
SI-SNR ↑
4.23
4.39
4.36
5.459
5.557
\scriptsize3⃝
LSD ↓
1.41
1.43
1.37
1.592
1.367
Table 4: Performance on noisy VCTK for 4 → 16 kHz.
SL
Condition
Pair
LSD ↓
STOI ↑
PESQ ↑
SI-SNR ↑
N-MOS ↑
\scriptsize1⃝
G.711 A-law
ZS/M
0.98/0.99
0.94/0.94
2.62/2.72
13.05/13.04
4.12/4.24
\scriptsize2⃝
G.711 μ -law
ZS/M
0.98/0.99
0.94/0.95
2.68/2.75
13.03/13.024
4.14/4.24
\scriptsize3⃝
AMR-NB 4.75
ZS/M
1.36/1.13
0.82/0.83
1.85/1.80
-16.88/-16.95
2.88/3.46
\scriptsize4⃝
AMR-NB 12.2
ZS/M
1.17/1.08
0.87/0.88
2.41/2.33
-16.88/-16.89
3.88/4.13
\scriptsize5⃝
Loss 3%
R/Z
0.99/0.99
0.93/0.92
2.26/2.14
9.91/10.97
3.85/2.75
\scriptsize6⃝
Loss 5%
R/Z
1.01/1.01
0.92/0.91
2.05/1.89
8.45/9.94
3.59/2.26
Table 5: Channel-degradation results for 4 → 16 kHz.
Model
Freq.
Par.(M)
MAC(M)
FLOP(M)
RTF(GPU)
Inf.(ms)
AP-BWE
16–48 kHz
72.07
14236.65
28473.31
0.0025
16.60
CIS-BWE
16–48 kHz
33.74
6790.86
13581.73
0.0028
16.64
Q-BWE
16–48 kHz
18.21
3225.68
6451.36
0.0032
19.65
Table 6: Overall computational complexity.
Row
Setting
LSD ↓
STOI ↑
PESQ ↑
SI-SNR ↑
N-MOS ↑
\scriptsize1⃝
Fixed 0.05
0.99
0.95
2.83
13.17
4.22
\scriptsize2⃝
Fixed 0.10
0.99
0.95
2.83
13.16
4.28
\scriptsize3⃝
Fixed 0.25
0.99
0.95
2.84
13.26
4.32
\scriptsize4⃝
Fixed 0.50
0.99
0.95
2.8
13.09
4.26
\scriptsize5⃝
Fixed 1.00
0.98
0.95
2.82
13.11
4.3
\scriptsize6⃝
Thompson
0.98
0.94
2.79
13.43
4.39
Table 7: Ablation on fixed and Thompson weighting.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Object
Shape
Invariant
**\scriptsize\arabic{appendixrow}**⃝ Input waveform x
(B,T)
finite float; model rate 16 kHz
**\scriptsize\arabic{appendixrow}**⃝ Complex STFT X
Speech enhancement improves intelligibility and listening quality by suppressing noise, reverberation, and competing speakers in recorded or live audio.
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.
Tarikul Islam Tamiti, Sajid Fardin Dipto, Luke Baja-Ricketts +2
Department of Cyber Security Engineering, George Mason University, Virginia, USA
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.
Tarikul Islam Tamiti, Sajid Fardin Dipto, David Vergano +2
Department of Cyber Security Engineering, George Mason University, USA.
Speech Bandwidth Extension (BWE) reconstructs missing high frequencies from a band-limited signal. Please note that perceptually important information is often concentrated in short high-frequency (HF) events. Therefore, an effective BWE requires both accurate HF generation and explicit control over where, how often, and in what spectral form HF energy should be produced. To address the above HF supervision, we present HFSWO, a compact complex-spectral Swin-based dual-stream adversarial framework with two newly proposed specialized discriminators: 1. High-Frequency Safeguard Discriminator (HFSD) converts excessive framewise HF-energy ratios into a differentiable safeguard-constraint penalty optimized through a projected primal--dual augmented-Lagrangian update; 2. Weighted Objective Scoring Discriminator (WOSD) learns an interpretable convex objective scoring over spectral flatness, centroid, and rolloff. Our rigorous experiments on VCTK English and MLS French datasets across multiple frequency ranges, including noisy conditions, show significantly improved perceptual scores with 74.6% fewer parameters than the existing baseline AP-BWE, establishing a new baseline in the BWE task in multilingual and multispeaker setups.
Tarikul Islam Tamiti, Sajid Fardin Dipto, Anomadarshi Barua