Audio Super-Resolution on VCTK 24 kHz (test)
0.74LSDFlowHigh
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FlowHighRTF=0.06 ± 0.02, GFLOPS=30.39, #params=49.40M2026.03 | 0.74 | 27.8 | 0.3 | 1 | — | |
| Decayed SSL guidanceInput SR=24 kHz, Target SR=48 kHz2026.05 | 0.76 | — | 0.429 | 1.06 | 22.68 | |
| Self-adaptive timestep samplingInput SR=24 kHz, Target SR=48 kHz2026.05 | 0.769 | — | 0.423 | 1.043 | 22.53 | |
| Structure-aware regularizationInput SR=24 kHz, Target SR=48 kHz2026.05 | 0.772 | — | 0.427 | 1.048 | 22.27 | |
| NU-Wave 2NFE=8, RTF=0.26 ± 0.02, GFLOPS=18.99, #params=1.8M2026.03 | 0.78 | 27.68 | 0.33 | 1.04 | — | |
| FastWaveNFE=8, RTF=0.30 ± 0.14, GFLOPS=12.87, #params=1.3M2026.03 | 0.83 | 27.22 | 0.48 | 1.05 | — | |
| Uniform baselineInput SR=24 kHz, Target SR=48 kHz2026.05 | 0.831 | — | 0.445 | 1.098 | 22.51 | |
| NVSRInput SR=24 kHz, Target SR=48 kHz2026.05 | 0.845 | — | 0.451 | 1.104 | 22.14 | |
| AudioSRInput SR=24 kHz, Target SR=48 kHz2026.05 | 0.876 | — | 0.482 | 1.132 | 23.76 | |
| FastWaveNFE=4, RTF=0.16 ± 0.03, GFLOPS=12.87, #params=1.3M2026.03 | 0.93 | 27.09 | 0.66 | 1.08 | — | |
| AudioSRRTF=4.99 ± 1.59, GFLOPS=2536.2, #params=1285.40M2026.03 | 1.27 | 23.03 | 0.58 | 1.69 | — |