# Audio Understanding on MMAU (Speech, Sound, Music, Overall)

Source: [Audio Understanding on MMAU (Speech, Sound, Music, Overall)](<https://www.sota2.com/research/sota/audio-understanding-on-mmau-speech-sound-music-overall>)

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

Data updated: 2026-06-01T10:28:10.550Z

Evaluation of Audio Understanding capabilities on the MMAU dataset, reporting domain-specific scores for Speech, Sound, and Music, along with an overall score.

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 | Audio Understanding |
| Dataset (including any stated split) | MMAU |
| Metric | Overall Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 14 of 14 |
| 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

- [Overall Score](<https://www.sota2.com/research/sota/audio-understanding-on-mmau-speech-sound-music-overall.md?metric=Overall%20Score>): Higher is better; unit: PERCENT.

- [Speech Score](<https://www.sota2.com/research/sota/audio-understanding-on-mmau-speech-sound-music-overall.md?metric=Speech%20Score>): Higher is better; unit: PERCENT.

- [Sound Score](<https://www.sota2.com/research/sota/audio-understanding-on-mmau-speech-sound-music-overall.md?metric=Sound%20Score>): Higher is better; unit: PERCENT.

- [Music Score](<https://www.sota2.com/research/sota/audio-understanding-on-mmau-speech-sound-music-overall.md?metric=Music%20Score>): Higher is better; unit: PERCENT.

- [General Score](<https://www.sota2.com/research/sota/audio-understanding-on-mmau-speech-sound-music-overall.md?metric=General%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Overall Score | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen3.5-Omni | 80.4 | Not recorded | 69f1907355a3c32af89fb98b |
| Qwen3-Omni | 77.5 | Not recorded | 69f1907355a3c32af89fb98b |
| Nemotron 3 Nano Omni | 76.9 | Not recorded | 69f1907355a3c32af89fb98b |
| Step-Audio2 | 72.7 | Not recorded | 69df1b7f7d0172140ec7b87d |
| Qwen2.5-Omni | 71.5 | Not recorded | 69df1b7f7d0172140ec7b87d |
| UAS-Audio | 69.4 | Not recorded | 69df1b7f7d0172140ec7b87d |
| Kimi-Audio | 68.2 | Not recorded | 69df1b7f7d0172140ec7b87d |
| UniAudio-Token | 61.1 | Not recorded | 6a1d123bdbd7150506f841e7 |
| UAS-Audio-D | 56 | Not recorded | 69df1b7f7d0172140ec7b87d |
| GLM-4-Voice-Tokenizer | 55.2 | Not recorded | 6a1d123bdbd7150506f841e7 |
| CosyVoice2 | 54.7 | Not recorded | 6a1d123bdbd7150506f841e7 |
| StableToken | 53.2 | Not recorded | 6a1d123bdbd7150506f841e7 |
| WavTokenizer | 51.7 | Not recorded | 6a1d123bdbd7150506f841e7 |
| GLM-4-Voice | 30.3 | Not recorded | 69df1b7f7d0172140ec7b87d |

### Qwen3.5-Omni

Overall Score: 80.4. Result ID: 69f1ae5847f34ce924d7ae84.

[Reporting paper](<https://www.sota2.com/research/paper/69f1907355a3c32af89fb98b.md>)

Recorded evaluation configuration: {"Open-Source":"false","Size":"Flash","Reasoning mode":"Reasoning off"}. This record does not assert independent reproduction.

### Qwen3-Omni

Overall Score: 77.5. Result ID: 69f1ae5847f34ce924d7ae83.

[Reporting paper](<https://www.sota2.com/research/paper/69f1907355a3c32af89fb98b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b881e209131cea88f3d.md>)

Recorded evaluation configuration: {"Open-Source":"true","Size":"30B-A3B","Reasoning mode":"Reasoning off"}. This record does not assert independent reproduction.

### Nemotron 3 Nano Omni

Overall Score: 76.9. Result ID: 69f1ae5847f34ce924d7ae82.

[Reporting paper](<https://www.sota2.com/research/paper/69f1907355a3c32af89fb98b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f1907355a3c32af89fb98b.md>)

Recorded evaluation configuration: {"Open-Source":"true","Size":"30B-A3B","Reasoning mode":"Reasoning off"}. This record does not assert independent reproduction.

### Step-Audio2

Overall Score: 72.7. Result ID: 69df523e47f34ce924c835de.

[Reporting paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b021e209131cea8635d.md>)

Recorded evaluation configuration: {"Audio Input Architecture":"Continuous"}. This record does not assert independent reproduction.

### Qwen2.5-Omni

Overall Score: 71.5. Result ID: 69df523e47f34ce924c835dd.

[Reporting paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77a21e209131cea7482c.md>)

Recorded evaluation configuration: {"Audio Input Architecture":"Continuous"}. This record does not assert independent reproduction.

### UAS-Audio

Overall Score: 69.4. Result ID: 69df523e47f34ce924c835df.

[Reporting paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

Recorded evaluation configuration: {"Audio Input Architecture":"Continuous"}. This record does not assert independent reproduction.

### Kimi-Audio

Overall Score: 68.2. Result ID: 69df523e47f34ce924c835dc.

[Reporting paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78431e209131cea77803.md>)

Recorded evaluation configuration: {"Audio Input Architecture":"Continuous"}. This record does not assert independent reproduction.

### UniAudio-Token

Overall Score: 61.1. Result ID: 6a1d5eb6fbc373592c1743d0.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d123bdbd7150506f841e7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d123bdbd7150506f841e7.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B","Tuning strategy":"tuned identically"}. This record does not assert independent reproduction.

### UAS-Audio-D

Overall Score: 56. Result ID: 69df523e47f34ce924c835db.

[Reporting paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

Recorded evaluation configuration: {"Audio Input Architecture":"Discrete"}. This record does not assert independent reproduction.

### GLM-4-Voice-Tokenizer

Overall Score: 55.2. Result ID: 6a1d5eb6fbc373592c1743ce.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d123bdbd7150506f841e7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76021e209131cea6ae5d.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B","Tuning strategy":"tuned identically"}. This record does not assert independent reproduction.

### CosyVoice2

Overall Score: 54.7. Result ID: 6a1d5eb6fbc373592c1743cd.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d123bdbd7150506f841e7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76591e209131cea6c851.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B","Tuning strategy":"tuned identically"}. This record does not assert independent reproduction.

### StableToken

Overall Score: 53.2. Result ID: 6a1d5eb6fbc373592c1743cf.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d123bdbd7150506f841e7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7baa1e209131cea89c25.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B","Tuning strategy":"tuned identically"}. This record does not assert independent reproduction.

### WavTokenizer

Overall Score: 51.7. Result ID: 6a1d5eb6fbc373592c1743cc.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d123bdbd7150506f841e7.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa771b1e209131cea7109d.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-3B","Tuning strategy":"tuned identically"}. This record does not assert independent reproduction.

### GLM-4-Voice

Overall Score: 30.3. Result ID: 69df523e47f34ce924c835da.

[Reporting paper](<https://www.sota2.com/research/paper/69df1b7f7d0172140ec7b87d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75f91e209131cea6abba.md>)

Recorded evaluation configuration: {"Audio Input Architecture":"Discrete"}. This record does not assert independent reproduction.

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