# Autoregressive Audio (AAR) on AudioSet 20k (subset of 100 random 10 s clips)

Source: [Autoregressive Audio (AAR) on AudioSet 20k (subset of 100 random 10 s clips)](<https://www.sota2.com/research/sota/autoregressive-audio-aar-on-audioset-20k-subset-of-100-random-10-s-clips>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-21T12:57:10.942Z

Evaluation of Autoregressive Audio (AAR) performance using reconstruction metrics (LSD, SNR, mel-MSE) on a specific subset of the AudioSet 20k dataset (100 random 10s clips).

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Autoregressive audio (AAR) |
| Dataset (including any stated split) | AudioSet 20k (subset of 100 random 10 s clips) |
| Metric | Compression Ratio |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Compression Ratio](<https://www.sota2.com/research/sota/autoregressive-audio-aar-on-audioset-20k-subset-of-100-random-10-s-clips.md?metric=Compression%20Ratio>): Higher is better; unit: SCALAR.

- [LSD](<https://www.sota2.com/research/sota/autoregressive-audio-aar-on-audioset-20k-subset-of-100-random-10-s-clips.md?metric=LSD>): Lower is better; unit: SCALAR.

- [mel-MSE](<https://www.sota2.com/research/sota/autoregressive-audio-aar-on-audioset-20k-subset-of-100-random-10-s-clips.md?metric=mel-MSE>): Lower is better; unit: SCALAR.

- [SNR](<https://www.sota2.com/research/sota/autoregressive-audio-aar-on-audioset-20k-subset-of-100-random-10-s-clips.md?metric=SNR>): Higher is better; unit: SCALAR.

- [Latency Cosine Score](<https://www.sota2.com/research/sota/autoregressive-audio-aar-on-audioset-20k-subset-of-100-random-10-s-clips.md?metric=Latency%20Cosine%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Compression Ratio | Result date | Report paper |
| --- | --- | --- | --- |
| PolarQuant | 2.75 | Not recorded | 6a0e91bcf984739ed6738d19 |
| TurboQuant-MSE | 2.72 | Not recorded | 6a0e91bcf984739ed6738d19 |
| TurboQuant-QJL | 2.69 | Not recorded | 6a0e91bcf984739ed6738d19 |
| OCTOPUS | 2.64 | Not recorded | 6a0e91bcf984739ed6738d19 |
| PolarQuant | 2.51 | Not recorded | 6a0e91bcf984739ed6738d19 |
| OCTOPUS-QJL | 2.5 | Not recorded | 6a0e91bcf984739ed6738d19 |
| TurboQuant-MSE | 2.48 | Not recorded | 6a0e91bcf984739ed6738d19 |
| TurboQuant-QJL | 2.46 | Not recorded | 6a0e91bcf984739ed6738d19 |
| OCTOPUS | 2.43 | Not recorded | 6a0e91bcf984739ed6738d19 |
| PolarQuant | 2.31 | Not recorded | 6a0e91bcf984739ed6738d19 |
| OCTOPUS-QJL | 2.31 | Not recorded | 6a0e91bcf984739ed6738d19 |
| TurboQuant-MSE | 2.29 | Not recorded | 6a0e91bcf984739ed6738d19 |
| TurboQuant-QJL | 2.26 | Not recorded | 6a0e91bcf984739ed6738d19 |
| OCTOPUS | 2.21 | Not recorded | 6a0e91bcf984739ed6738d19 |
| OCTOPUS-QJL | 2.11 | Not recorded | 6a0e91bcf984739ed6738d19 |

### PolarQuant

Compression Ratio: 2.75. Result ID: 6a0f0110fbc373592c07ae25.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76d11e209131cea6f3df.md>)

Recorded evaluation configuration: {"bits":"2","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### TurboQuant-MSE

Compression Ratio: 2.72. Result ID: 6a0f0110fbc373592c07ae23.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78401e209131cea776a9.md>)

Recorded evaluation configuration: {"bits":"2","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### TurboQuant-QJL

Compression Ratio: 2.69. Result ID: 6a0f0110fbc373592c07ae24.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78401e209131cea776a9.md>)

Recorded evaluation configuration: {"bits":"2","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### OCTOPUS

Compression Ratio: 2.64. Result ID: 6a0f0110fbc373592c07ae26.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

Recorded evaluation configuration: {"bits":"2","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### PolarQuant

Compression Ratio: 2.51. Result ID: 6a0f0110fbc373592c07ae20.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76d11e209131cea6f3df.md>)

Recorded evaluation configuration: {"bits":"3","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### OCTOPUS-QJL

Compression Ratio: 2.5. Result ID: 6a0f0110fbc373592c07ae27.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

Recorded evaluation configuration: {"bits":"2","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### TurboQuant-MSE

Compression Ratio: 2.48. Result ID: 6a0f0110fbc373592c07ae1e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78401e209131cea776a9.md>)

Recorded evaluation configuration: {"bits":"3","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### TurboQuant-QJL

Compression Ratio: 2.46. Result ID: 6a0f0110fbc373592c07ae1f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78401e209131cea776a9.md>)

Recorded evaluation configuration: {"bits":"3","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### OCTOPUS

Compression Ratio: 2.43. Result ID: 6a0f0110fbc373592c07ae21.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

Recorded evaluation configuration: {"bits":"3","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### PolarQuant

Compression Ratio: 2.31. Result ID: 6a0f0110fbc373592c07ae1b.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76d11e209131cea6f3df.md>)

Recorded evaluation configuration: {"bits":"4","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### OCTOPUS-QJL

Compression Ratio: 2.31. Result ID: 6a0f0110fbc373592c07ae22.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

Recorded evaluation configuration: {"bits":"3","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### TurboQuant-MSE

Compression Ratio: 2.29. Result ID: 6a0f0110fbc373592c07ae19.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78401e209131cea776a9.md>)

Recorded evaluation configuration: {"bits":"4","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### TurboQuant-QJL

Compression Ratio: 2.26. Result ID: 6a0f0110fbc373592c07ae1a.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78401e209131cea776a9.md>)

Recorded evaluation configuration: {"bits":"4","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### OCTOPUS

Compression Ratio: 2.21. Result ID: 6a0f0110fbc373592c07ae1c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

Recorded evaluation configuration: {"bits":"4","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

### OCTOPUS-QJL

Compression Ratio: 2.11. Result ID: 6a0f0110fbc373592c07ae1d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0e91bcf984739ed6738d19.md>)

Recorded evaluation configuration: {"bits":"4","residual window scale":"1","V group":"16","per-layer protection":"none"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
