Multimodal Question Answering on CaReSound
93.5Yes/No AccuracyAudioTSLM (1.4B)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AudioTSLM (1.4B)Model type=ALM2026.03 | 93.5 | 86.5 | 42.6 | 67.32 | 64.32 | 95.19 | |
| AudioTSLMEncoder=Raw2026.03 | 93.5 | 86.48 | 42.6 | 67.32 | 64.32 | 95.19 | |
| AudioTSLMEncoder=Mel2026.03 | 93.5 | 86.32 | 42.69 | 66.87 | 63.59 | 95.07 | |
| AudioTSLMEncoder=Whisper2026.03 | 93.5 | 86.46 | 42.47 | 66.73 | 63.96 | 95.09 | |
| AudioTSLMTemporal Context=30s2026.03 | 93.5 | 86.5 | 42.6 | 67.3 | 64.3 | 95.2 | |
| AudioTSLMEncoder=CLAP2026.03 | 93.44 | 86.3 | 43.13 | 67.26 | 64.53 | 95.13 | |
| AudioTSLMEncoder=Wav2Vec2026.03 | 93.21 | 85.64 | 42.38 | 66.43 | 63.49 | 95.03 | |
| CaReAQA (3B)Model type=ALM2026.03 | 93.12 | 84.6 | 34.85 | 61.17 | 51.07 | 94.23 | |
| AudioTSLMTemporal Context=20s2026.03 | 92 | 82.6 | 41.3 | 65.5 | 62.3 | 94.8 | |
| AudioTSLMTemporal Context=10s2026.03 | 91.1 | 80.4 | 39.5 | 63.9 | 60.7 | 94.6 | |
| Audio-Flamingo3 (8B)Model type=ALM2026.03 | 85 | 54.5 | 5.51 | 28.83 | 28.42 | 88.51 | |
| Qwen2.5-Omni (0.5B)Model type=ALM2026.03 | 84.62 | 52.3 | 6.74 | 13.51 | 22.76 | 86.48 | |
| Qwen2-Audio (7B)Model type=ALM2026.03 | 17.83 | 17.1 | 4.33 | 29.51 | 35.92 | 89.96 |