Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

Audio Reconstruction research benchmarks – Page 2 · SOTA2 Research

Audio Reconstruction

Benchmarks

Dataset NameSOTA MethodSort

Dataset NameSOTA methodMetricTrendResultsLast Updated

MUSDB18-HQ (test)

H-Codec-2.0 (Large)

0.504Mel Loss

7

Feb 26, 2026

Song Describer Dataset (SDD) (test)

12SI-SDR

6

May 19, 2026

90.5MUSHRA

6

Feb 26, 2026

Audio Evaluation Set 4 modalities, 150 samples per category

3.92ViSQOL

6

Feb 26, 2026

VGGSound (test)

35.45SI-SDR

5

Jul 1, 2026

InternalSFX (test)

Woosh-AE-Public

0.021MelDist

5

Apr 3, 2026

AudioCaps public (test)

Woosh-AE-Public

0.032Mel Distance

5

Apr 3, 2026

86.6MUSHRA Score

5

Feb 26, 2026

80.8MUSHRA Score

5

Feb 26, 2026

SIREN audio segments

0Bach MSE (x1e-3)

5

Feb 26, 2026

SongDescriber no vocals 10 s chunks (eval)

1.27STFT Error

4

Jun 29, 2026

AudioCaps and Song Describer Dataset (60 clips)

4.45MOS

4

Jun 23, 2026

LibriTTS clean 24 kHz (test)

4.063UTMOS

4

Jun 23, 2026

LibriTTS, MTG-Jamendo, and Opencpop Average (test)

4MOS

4

Feb 26, 2026

Librispeech 3 sec

36.24PSNR

4

Feb 26, 2026

Librispeech 1 sec

43.25PSNR

4

Feb 26, 2026

7.44CE Score

3

Mar 24, 2026

LibriSpeech 3s (test-clean)

Instance Pattern Composers

35.38PSNR

2

Feb 26, 2026

LibriSpeech 1s (test-clean)

Instance Pattern Composers

40.11PSNR

2

Feb 26, 2026

Internal speech dataset 200 pairs (test)

232Average Winrate

1

Feb 26, 2026

Page 2 of 2

Popular datasets

LibriSpeech AudioSet LibriTTS Codec-SUPERB SEED

Follow for updates

Join SOTA2 Contact