mlsanityData to 7 Oct 2026

ASPIRin: Action Space Projection for Interactivity-Optimized Reinforcement Learning in Full-Duplex Speech Language Models

arXiv 2604.10065 ↗First posted 11 Apr 2026cs.CL cs.AI cs.SDcited by 3

Not on the map. Why?

Abstract (arXiv)

End-to-end full-duplex Speech Language Models (SLMs) require precise turn-taking for natural interaction. However, optimizing temporal dynamics via standard raw-token reinforcement learning (RL) degrades semantic quality, causing severe generative collapse and repetition. We propose ASPIRin, an interactivity-optimized RL framework that explicitly decouples when to speak from what to say. Using Action Space Projection, ASPIRin maps the text vocabulary into a coarse-grained binary state (active speech vs. inactive silence). By applying Group Relative Policy Optimization (GRPO) with rule-based rewards, it balances user interruption and response latency. Empirical evaluations show ASPIRin optimizes interactivity across turn-taking, backchanneling, and pause handling. Crucially, isolating timing from token selection preserves semantic coherence and reduces the portion of duplicate n-grams by over 50% compared to standard GRPO, effectively eliminating degenerative repetition.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 55 of its 59 references that are in the map. First posted on arXiv 11 Apr 2026.

Builds on

55 of 59 references.

References · 55

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  3. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  4. 2024GPT-4o System Cardcited by 4,467
  5. 2022Robust Speech Recognition via Large-Scale Weak Supervisioncited by 3,972
  6. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,655
  7. 2023Qwen Technical Reportcited by 3,273
  8. 2023Neural Codec Language Models are Zero-Shot Text to Speech Synthesizerscited by 737
  9. 2024Qwen2-Audio Technical Reportcited by 543
  10. 2024Moshi: a speech-text foundation model for real-time dialoguecited by 536
  11. 2019Neural Text Generation with Unlikelihood Trainingcited by 489
  12. 2018Texygen: A Benchmarking Platform for Text Generation Modelscited by 484
38 more
  1. 2024CosyVoice: A Scalable Multilingual Zero-shot Text-to-speech Synthesizer based on Supervised Semantic Tokenscited by 323
  2. 2024CosyVoice 2: Scalable Streaming Speech Synthesis with Large Language Modelscited by 282
  3. 2024XTTS: a Massively Multilingual Zero-Shot Text-to-Speech Modelcited by 174
  4. 2025CosyVoice 3: Towards In-the-wild Speech Generation via Scaling-up and Post-trainingcited by 142
  5. 2024VALL-E 2: Neural Codec Language Models are Human Parity Zero-Shot Text to Speech Synthesizerscited by 119
  6. 2025On The Landscape of Spoken Language Models: A Comprehensive Surveycited by 116
  7. 2026Qwen3-TTS Technical Reportcited by 106
  8. 2026Qwen3-ASR Technical Reportcited by 105
  9. 2023Dynamic-SUPERB: Towards A Dynamic, Collaborative, and Comprehensive Instruction-Tuning Benchmark for Speechcited by 99
  10. 2025Full-Duplex-Bench: A Benchmark to Evaluate Full-duplex Spoken Dialogue Models on Turn-taking Capabilitiescited by 94
  11. 2024Dynamic-SUPERB Phase-2: A Collaboratively Expanding Benchmark for Measuring the Capabilities of Spoken Language Models with 180 Taskscited by 70
  12. 2024Language Model Can Listen While Speakingcited by 62
  13. 2026PersonaPlex: Voice and Role Control for Full Duplex Conversational Speech Modelscited by 57
  14. 2025Full-Duplex-Bench v1.5: Evaluating Overlap Handling for Full-Duplex Speech Modelscited by 53
  15. 2025Canary-1B-v2 & Parakeet-TDT-0.6B-v3: Efficient and High-Performance Models for Multilingual ASR and ASTcited by 51
  16. 2025DeSTA2.5-Audio: Toward General-Purpose Large Audio Language Model with Self-Generated Cross-Modal Alignmentcited by 47
  17. 2024DeSTA2: Developing Instruction-Following Speech Language Model Without Speech Instruction-Tuning Datacited by 43
  18. 2025Towards Holistic Evaluation of Large Audio-Language Models: A Comprehensive Surveycited by 38
  19. 2025Full-Duplex-Bench-v2: A Multi-Turn Evaluation Framework for Duplex Dialogue Systems with an Automated Examinercited by 38
  20. 2024DeSTA: Enhancing Speech Language Models through Descriptive Speech-Text Alignmentcited by 32
  21. 2025SALM-Duplex: Efficient and Direct Duplex Modeling for Speech-to-Speech Language Modelcited by 32
  22. 2023SpeechPrompt v2: Prompt Tuning for Speech Classification Taskscited by 31
  23. 2024Speech-Copilot: Leveraging Large Language Models for Speech Processing via Task Decomposition, Modularization, and Program Generationcited by 27
  24. 2025STITCH: Simultaneous Thinking and Talking with Chunked Reasoning for Spoken Language Modelscited by 25
  25. 2024Align-SLM: Textless Spoken Language Models with Reinforcement Learning from AI Feedbackcited by 22
  26. 2024Building a Taiwanese Mandarin Spoken Language Model: A First Attemptcited by 22
  27. 2025Reducing Object Hallucination in Large Audio-Language Models via Audio-Aware Decodingcited by 22
  28. 2022SpeechPrompt: An Exploration of Prompt Tuning on Generative Spoken Language Model for Speech Processing Taskscited by 19
  29. 2025A Preliminary Exploration with GPT-4o Voice Modecited by 16
  30. 2025TASTE: Text-Aligned Speech Tokenization and Embedding for Spoken Language Modelingcited by 15
  31. 2023Investigating Zero-Shot Generalizability on Mandarin-English Code-Switched ASR and Speech-to-text Translation of Recent Foundation Models with Self-Supervision and Weak Supervisioncited by 13
  32. 2025Game-Time: Evaluating Temporal Dynamics in Spoken Language Modelscited by 13
  33. 2024Do Prompts Really Prompt? Exploring the Prompt Understanding Capability of Whispercited by 12
  34. 2024SpeechPrompt: Prompting Speech Language Models for Speech Processing Taskscited by 12
  35. 2025Analyzing Mitigation Strategies for Catastrophic Forgetting in End-to-End Training of Spoken Language Modelscited by 11
  36. 2025Aligning Spoken Dialogue Models from User Interactionscited by 11
  37. 2021Mandarin-English Code-switching Speech Recognition with Self-supervised Speech Representation Modelscited by 9
  38. 2025A Self-Refining Framework for Enhancing ASR Using TTS-Synthesized Datacited by 7

50 most cited of 55 All 55 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Multi-Faceted Interactivity Alignment in Full-Duplex Speech Modelscited by 9
  2. 2026SteerDuplex: Steerable Duplex Speech Dialogue Modelscited by 0
  3. 2026DiffuPlex: Accelerating Full-Duplex Spoken Dialog Models via Rolling Masked Diffusioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Entropy-Regularized Token-Level Policy Optimization for Language Agent Reinforcement
  2. 2026Dual-Axis Generative Reward Model Toward Semantic and Turn-taking Robustness in Interactive Spoken Dialogue Models
  3. 2025SpeakRL: Synergizing Reasoning, Speaking, and Acting in Language Models with Reinforcement Learning
  4. 2026ASymPO: Asymmetric-Scale Policy Optimization for Asynchronous LLM Post-Training Without Behavior Information
  5. 2024Regressing the Relative Future: Efficient Policy Optimization for Multi-turn RLHF
  6. 2025Explore the Reinforcement Learning for the LLM based ASR and TTS system
  7. 2026Words & Weights: Streamlining Multi-Turn Interactions via Co-Adaptation
  8. 2026WavAlign: Enhancing Intelligence and Expressiveness in Spoken Dialogue Models via Adaptive Hybrid Post-Training