Research map / Large language models / Alignment and preferences
TangoFlux: Super Fast and Faithful Text to Audio Generation with Flow Matching and Clap-Ranked Preference Optimization
Chia-Yu Hung, Navonil Majumder, Zhifeng Kong, Ambuj Mehrish, Amir Zadeh, Chuan Li, Rafael Valle, Bryan Catanzaro, Soujanya Poria
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 |
|---|---|---|---|
| use CRPO preference data instead of Audio Alpaca (ablation) | TangoFlux-base · AudioCaps | CLAP score: 0.448 -> 0.453 | Improved |
| use CRPO preference data instead of BATON (ablation) | TangoFlux-base · AudioCaps | CLAP score: 0.437 -> 0.453 | Improved |
| use CRPO preference data instead of BATON (ablation) | TangoFlux-base · AudioCaps | FD_openl3: 80.5 -> 79.1 | Improved |
| apply CRPO alignment to TangoFlux-base | TangoFlux-base · AudioCaps | CLAP score: 0.431 -> 0.480 | Improved |
| apply CRPO alignment to TangoFlux-base | TangoFlux-base · AudioCaps | FD_openl3: 80.2 -> 75.1 | Improved |
| apply CRPO alignment to TangoFlux-base | TangoFlux-base · AudioCaps | KL_passt: 1.22 -> 1.15 | Improved |
| apply CRPO alignment to TangoFlux-base | TangoFlux-base · AudioCaps multi-event | CLAP score: 0.438 -> 0.488 | Improved |
| CLAP best-of-N reranking, N=1 to 15 | TangoFlux · AudioCaps | CLAP score: 0.480 -> 0.502 | Improved |
| CLAP best-of-N reranking, N=1 to 15 | TangoFlux · AudioCaps | FD_openl3: 75.0 -> 75.1 | No clear effect |
| CLAP best-of-N reranking, N=1 to 15 | Tango 2 · AudioCaps | CLAP score: 0.447 -> 0.473 | Improved |
| raise CFG scale from 3.0 to 5.0 (ablation) | TangoFlux · AudioCaps | FD_openl3: 77.7 -> 74.6 | Improved |
| raise CFG scale from 3.5 to 5.0 (ablation) | TangoFlux · AudioCaps | CLAP score: 0.481 -> 0.472 | Hurt |
Later papers that compared with this method
1 later accepted paper reports TangoFlux as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- AudioStory: Generating Long-Form Narrative Audio with Large Language Models (CVPR 2026 · better on 3 of 4 shared results)
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
- UniFlow-Audio: Unified Flow Matching for Audio Generation from Omni-Modalities (NeurIPS 2026 · similarity 0.88)
- ETTA: Elucidating the Design Space of Text-to-Audio Models (ICML 2025 · similarity 0.87)
- AudioCALM: Continuous Autoregressive Language Modeling for Universal Audio Generation (NeurIPS 2026 · similarity 0.87)
- UniAudio: Towards Universal Audio Generation with Large Language Models (ICML 2024 · similarity 0.87)
- Model-Guided Dual-Role Alignment for High-Fidelity Open-Domain Video-to-Audio Generation (NeurIPS 2025 · similarity 0.86)
- Advancing Audio-Based Text Generation with Imbalance Preference Optimization (AAAI 2025 · similarity 0.86)
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.
