# Large Audio-Language Model Safety Evaluation on Figstep audio

Source: [Large Audio-Language Model Safety Evaluation on Figstep audio](<https://www.sota2.com/research/sota/large-audio-language-model-safety-evaluation-on-figstep-audio>)

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

Data updated: 2026-05-08T08:27:22.918Z

Evaluation of audio-language models for safety compliance using the Figstep audio dataset, measuring performance via metrics such as ASR, BRR, and Safe RR.

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 | Large Audio-Language Model Safety Evaluation |
| Dataset (including any stated split) | Figstep audio |
| Metric | ASR |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 18 of 18 |
| 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

- [ASR](<https://www.sota2.com/research/sota/large-audio-language-model-safety-evaluation-on-figstep-audio.md?metric=ASR>): Higher is better; unit: PERCENT.

- [BRR](<https://www.sota2.com/research/sota/large-audio-language-model-safety-evaluation-on-figstep-audio.md?metric=BRR>): Higher is better; unit: PERCENT.

- [Safe RR](<https://www.sota2.com/research/sota/large-audio-language-model-safety-evaluation-on-figstep-audio.md?metric=Safe%20RR>): Higher is better; unit: PERCENT.

## Recorded results

| Method | ASR | Result date | Report paper |
| --- | --- | --- | --- |
| MDSteer-h2s | 62.8 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-h2s | 58.8 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-c2r | 52.3 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-h2s | 50.8 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 42 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 35.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 34 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 31.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 30 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 28.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-c2r | 26.8 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 25.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| MDSteer-c2r | 15.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| No Defense | 7.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| FSD | 6.4 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| SARSteer | 5.2 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 3.6 | Not recorded | 68fa7c5d1e209131cea8dc06 |
| AdaShield | 2.4 | Not recorded | 68fa7c5d1e209131cea8dc06 |

### MDSteer-h2s

ASR: 62.8. Result ID: 69fd9df447f34ce924e0900f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa757b1e209131cea68414.md>)

[Implementation](<https://github.com/andyrdt/refusal_direction>)

Recorded evaluation configuration: {"Model":"Qwen3-Omni-30B (MoE)"}. This record does not assert independent reproduction.

### MDSteer-h2s

ASR: 58.8. Result ID: 69fd9df447f34ce924e09009.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa757b1e209131cea68414.md>)

[Implementation](<https://github.com/andyrdt/refusal_direction>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Omni-7B"}. This record does not assert independent reproduction.

### MDSteer-c2r

ASR: 52.3. Result ID: 69fd9df447f34ce924e0900a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88272.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Omni-7B"}. This record does not assert independent reproduction.

### MDSteer-h2s

ASR: 50.8. Result ID: 69fd9df447f34ce924e09015.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa757b1e209131cea68414.md>)

[Implementation](<https://github.com/andyrdt/refusal_direction>)

Recorded evaluation configuration: {"Model":"Voxtral-Mini-3B"}. This record does not assert independent reproduction.

### No Defense

ASR: 42. Result ID: 69fd9df447f34ce924e09008.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Omni-7B"}. This record does not assert independent reproduction.

### FSD

ASR: 35.2. Result ID: 69fd9df447f34ce924e09018.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b01e209131cea6e7cf.md>)

Recorded evaluation configuration: {"Model":"Voxtral-Mini-3B"}. This record does not assert independent reproduction.

### No Defense

ASR: 34. Result ID: 69fd9df447f34ce924e09014.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Voxtral-Mini-3B"}. This record does not assert independent reproduction.

### SARSteer

ASR: 31.2. Result ID: 69fd9df447f34ce924e0900d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Omni-7B"}. This record does not assert independent reproduction.

### SARSteer

ASR: 30. Result ID: 69fd9df447f34ce924e09019.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Voxtral-Mini-3B"}. This record does not assert independent reproduction.

### FSD

ASR: 28.2. Result ID: 69fd9df447f34ce924e0900c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b01e209131cea6e7cf.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Omni-7B"}. This record does not assert independent reproduction.

### MDSteer-c2r

ASR: 26.8. Result ID: 69fd9df447f34ce924e09016.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88272.md>)

Recorded evaluation configuration: {"Model":"Voxtral-Mini-3B"}. This record does not assert independent reproduction.

### AdaShield

ASR: 25.2. Result ID: 69fd9df447f34ce924e0900b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727e1e209131cea56bc0.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Omni-7B"}. This record does not assert independent reproduction.

### MDSteer-c2r

ASR: 15.2. Result ID: 69fd9df447f34ce924e09010.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88272.md>)

Recorded evaluation configuration: {"Model":"Qwen3-Omni-30B (MoE)"}. This record does not assert independent reproduction.

### No Defense

ASR: 7.2. Result ID: 69fd9df447f34ce924e0900e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Qwen3-Omni-30B (MoE)"}. This record does not assert independent reproduction.

### FSD

ASR: 6.4. Result ID: 69fd9df447f34ce924e09012.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b01e209131cea6e7cf.md>)

Recorded evaluation configuration: {"Model":"Qwen3-Omni-30B (MoE)"}. This record does not assert independent reproduction.

### SARSteer

ASR: 5.2. Result ID: 69fd9df447f34ce924e09013.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

Recorded evaluation configuration: {"Model":"Qwen3-Omni-30B (MoE)"}. This record does not assert independent reproduction.

### AdaShield

ASR: 3.6. Result ID: 69fd9df447f34ce924e09011.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727e1e209131cea56bc0.md>)

Recorded evaluation configuration: {"Model":"Qwen3-Omni-30B (MoE)"}. This record does not assert independent reproduction.

### AdaShield

ASR: 2.4. Result ID: 69fd9df447f34ce924e09017.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c5d1e209131cea8dc06.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727e1e209131cea56bc0.md>)

Recorded evaluation configuration: {"Model":"Voxtral-Mini-3B"}. This record does not assert independent reproduction.

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