mlsanityData to 7 Oct 2026

Improving Text-to-Audio Instruction Following via Fine-Grained Feedback from Audio-Aware Large Language Models

arXiv 2607.13408 ↗First posted 15 Jul 2026eess.AS cs.AI cs.CLcited by 0

On the map: LLM alignment / RLHF / preference optimisation → LLM preference optimization methods

Abstract (arXiv)

Recent text-to-audio models generate high-quality audio, but often fail to follow instructions involving multiple sound events and temporal order. This gap arises because existing evaluation and training signals mainly emphasize global similarity or perceptual quality, with limited supervision on instruction-level correctness. We propose an instruction-level framework that uses audio-aware large language models (ALLMs) as fine-grained judges to verify target event presence and temporal relations in generated audio. After validating ALLM judgments on benchmarks and through human verification, we use their feedback to construct preference pairs for direct preference optimization. We further introduce S3Bench, a narrative benchmark for evaluating multi-event temporal instruction following. Experiments show that our method improves event completeness, temporal ordering, and joint instruction-following accuracy across existing benchmarks and S3Bench, while maintaining audio quality.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 56 of its 62 references that are in the map. First posted on arXiv 15 Jul 2026.

Builds on

56 of 62 references.

References · 56

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,655
  3. 2016Improved Techniques for Training GANscited by 2,938
  4. 2022Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentationcited by 707
  5. 2025Qwen2.5-Omni Technical Reportcited by 704
  6. 2022AudioLM: a Language Modeling Approach to Audio Generationcited by 611
  7. 2023AudioLDM: Text-to-Audio Generation with Latent Diffusion Modelscited by 592
  8. 2025Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAscited by 491
  9. 2023Simple and Controllable Music Generationcited by 480
  10. 2023MusicLM: Generating Music From Textcited by 442
  11. 2023AudioLDM 2: Learning Holistic Audio Generation with Self-supervised Pretrainingcited by 343
  12. 2022AudioGen: Textually Guided Audio Generationcited by 292
38 more
  1. 2024MMAU: A Massive Multi-Task Audio Understanding and Reasoning Benchmarkcited by 281
  2. 2022Diffsound: Discrete Diffusion Model for Text-to-sound Generationcited by 256
  3. 2025Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Modelscited by 210
  4. 2024Stable Audio Opencited by 192
  5. 2023Text-to-Audio Generation using Instruction-Tuned LLM and Latent Diffusion Modelcited by 145
  6. 2025Audio Flamingo 2: An Audio-Language Model with Long-Audio Understanding and Expert Reasoning Abilitiescited by 127
  7. 2025On The Landscape of Spoken Language Models: A Comprehensive Surveycited by 116
  8. 2025MMAR: A Challenging Benchmark for Deep Reasoning in Speech, Audio, Music, and Their Mixcited by 113
  9. 2023SoundStorm: Efficient Parallel Audio Generationcited by 111
  10. 2023Dynamic-SUPERB: Towards A Dynamic, Collaborative, and Comprehensive Instruction-Tuning Benchmark for Speechcited by 99
  11. 2023Joint Audio and Speech Understandingcited by 99
  12. 2024Tango 2: Aligning Diffusion-based Text-to-Audio Generations through Direct Preference Optimizationcited by 97
  13. 2023Make-An-Audio 2: Temporal-Enhanced Text-to-Audio Generationcited by 90
  14. 2023BLSP: Bootstrapping Language-Speech Pre-training via Behavior Alignment of Continuation Writingcited by 75
  15. 2024Dynamic-SUPERB Phase-2: A Collaboratively Expanding Benchmark for Measuring the Capabilities of Spoken Language Models with 180 Taskscited by 70
  16. 2024TangoFlux: Super Fast and Faithful Text to Audio Generation with Flow Matching and Clap-Ranked Preference Optimizationcited by 67
  17. 2023AudioChatLlama: Towards General-Purpose Speech Abilities for LLMscited by 59
  18. 2025MMAU-Pro: A Challenging and Comprehensive Benchmark for Holistic Evaluation of Audio General Intelligencecited by 59
  19. 2023CompA: Addressing the Gap in Compositional Reasoning in Audio-Language Modelscited by 43
  20. 2024BLSP-Emo: Towards Empathetic Large Speech-Language Modelscited by 36
  21. 2024Understanding Sounds, Missing the Questions: The Challenge of Object Hallucination in Large Audio-Language Modelscited by 30
  22. 2024EzAudio: Enhancing Text-to-Audio Generation with Efficient Diffusion Transformercited by 30
  23. 2024Can Large Audio-Language Models Truly Hear? Tackling Hallucinations with Multi-Task Assessment and Stepwise Audio Reasoningcited by 29
  24. 2024Speech-Copilot: Leveraging Large Language Models for Speech Processing via Task Decomposition, Modularization, and Program Generationcited by 27
  25. 2025Audio Large Language Models Can Be Descriptive Speech Quality Evaluatorscited by 25
  26. 2025AudioJudge: Understanding What Works in Large Audio Model Based Speech Evaluationcited by 20
  27. 2024AudioTime: A Temporally-aligned Audio-text Benchmark Datasetcited by 19
  28. 2024ETTA: Elucidating the Design Space of Text-to-Audio Modelscited by 17
  29. 2025Audio-Aware Large Language Models as Judges for Speaking Stylescited by 17
  30. 2025InstructTTSEval: Benchmarking Complex Natural-Language Instruction Following in Text-to-Speech Systemscited by 14
  31. 2025Game-Time: Evaluating Temporal Dynamics in Spoken Language Modelscited by 13
  32. 2024SpeechPrompt: Prompting Speech Language Models for Speech Processing Taskscited by 12
  33. 2025Teaching Audio-Aware Large Language Models What Does Not Hear: Mitigating Hallucinations through Synthesized Negative Samplescited by 11
  34. 2026AQAScore: Evaluating Semantic Alignment in Text-to-Audio Generation via Audio Question Answeringcited by 8
  35. 2025From Alignment to Advancement: Bootstrapping Audio-Language Alignment with Synthetic Datacited by 7
  36. 2023Towards General-Purpose Text-Instruction-Guided Voice Conversioncited by 6
  37. 2025T2A-Feedback: Improving Basic Capabilities of Text-to-Audio Generation via Fine-grained AI Feedbackcited by 6
  38. 2025IMPACT: Iterative Mask-based Parallel Decoding for Text-to-Audio Generation with Diffusion Modelingcited by 6

50 most cited of 56 All 56 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025IFEval-Audio: Benchmarking Instruction-Following Capability in Audio-based Large Language Models
  2. 2026JASTIN: Aligning LLMs for Zero-Shot Audio and Speech Evaluation via Natural Language Instructions
  3. 2026AnyAudio-Judge: A Dynamic Rubric-Based Benchmark and Evaluator for Audio Instruction Following
  4. 2026A Closer Look at Failure Modes in Temporal Understanding of Large Audio-Language Models
  5. 2026AudioCALM: Continuous Autoregressive Language Modeling for Universal Audio Generation ↗
  6. 2026Beyond Transcription: Unified Audio Schema for Perception-Aware AudioLLMs
  7. 2025Towards Reliable Large Audio Language Model
  8. 2025AudioMarathon: A Comprehensive Benchmark for Long-Context Audio Understanding and Efficiency in Audio LLMs