Research map / Large language models / Language, safety and interpretability
Music Flamingo: Scaling Music Understanding in Audio Language Models
Sreyan Ghosh, Arushi Goel, Lasha Koroshinadze, Sang-gil Lee, Zhifeng Kong, Joao Felipe Santos, Ramani Duraiswami, Dinesh Manocha, Wei Ping, Mohammad Shoeybi, Bryan Catanzaro
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add GRPO reinforcement learning with thinking traces after SFT (ablation) | Music Flamingo · MMAU-Pro-Music | accuracy: 63.9 -> 65.6 | Improved |
| add GRPO reinforcement learning with thinking traces after SFT (ablation) | Music Flamingo · MuChoMusic | accuracy: 69.5 -> 74.58 | Improved |
| fine-tune Audio Flamingo 3 on MF-Skills and music data | Audio Flamingo 3 · MMAU Music full-test | accuracy: 73.95 -> 76.83 | Improved |
| fine-tune Audio Flamingo 3 on MF-Skills and music data | Audio Flamingo 3 · Music AVQA | accuracy: 76.7 -> 73.6 | Hurt |
| fine-tune Audio Flamingo 3 on MF-Skills and music data | Audio Flamingo 3 · SongCaps human rating | score (1-10): 6.5 -> 8.3 | Improved |
| fine-tune Audio Flamingo 3 on MF-Skills and music data | Audio Flamingo 3 · NSynth Source | accuracy: 65.5 -> 75.89 | Improved |
| use a music-biased encoder (MERT) instead of captioning-trained Whisper encoder (ablation) | linear probe · GS key classification | accuracy: 40.56 -> 56.12 | Improved |
| use a music-biased encoder (MERT) instead of captioning-trained Whisper encoder (ablation) | linear probe · GTZAN genre classification | accuracy: 91.37 -> 78.96 | Hurt |
| add structured thinking reward and accuracy reward on top of format reward in GRPO | Music Flamingo GRPO · music caption and QA training | reasoning quality | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Audio Flamingo 3 — Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models (NeurIPS 2025) · better on 8 of 9 shared results
- Audio Flamingo 2 — Audio Flamingo 2: An Audio-Language Model with Long-Audio Understanding and Expert Reasoning Abilities (ICML 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LLark: A Multimodal Instruction-Following Language Model for Music (ICML 2024 · similarity 0.91)
- Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models (NeurIPS 2025 · similarity 0.90)
- Audio Flamingo 2: An Audio-Language Model with Long-Audio Understanding and Expert Reasoning Abilities (ICML 2025 · similarity 0.89)
- Audio Flamingo: A Novel Audio Language Model with Few-Shot Learning and Dialogue Abilities (ICML 2024 · similarity 0.89)
- LeVo: High-Quality Song Generation with Multi-Preference Alignment (NeurIPS 2025 · similarity 0.88)
- CMI-RewardBench: Evaluating Music Reward Models with Compositional Multimodal Instruction (ICML 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
