# ParaS2S: Benchmarking and Aligning Spoken Language Models for Paralinguistic-aware Speech-to-Speech Interaction

arXiv: https://arxiv.org/abs/2511.08723 · HTML: https://mlsanity.com/paper/2511.08723/

## Key facts

- Cited by 24 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 11 Nov 2025.

## Built on by (most cited first, 24 of 24)

- [τ-Voice: Benchmarking Full-Duplex Voice Agents on Real-World Domains](https://mlsanity.com/paper/2603.13686/) (2026)
- [Multi-Faceted Interactivity Alignment in Full-Duplex Speech Models](https://mlsanity.com/paper/2606.11167/) (2026)
- [VoxSafeBench: Not Just What Is Said, but Who, How, and Where](https://mlsanity.com/paper/2604.14548/) (2026)
- [AQUA-Bench: Beyond Finding Answers to Knowing When There Are None in Audio Question Answering](https://mlsanity.com/paper/2601.12248/) (2026)
- [ParaBridge: Bridging Paralinguistic Perception and Dialogue Behavior in Speech Language Models](https://mlsanity.com/paper/2606.10581/) (2026)
- [Walking Through Uncertainty: An Empirical Study of Uncertainty Estimation for Audio-Aware Large Language Models](https://mlsanity.com/paper/2604.25591/) (2026)
- [WavAlign: Enhancing Intelligence and Expressiveness in Spoken Dialogue Models via Adaptive Hybrid Post-Training](https://mlsanity.com/paper/2604.14932/) (2026)
- [Reflecting Twice before Speaking with Empathy: Self-Reflective Alternating Inference for Empathy-Aware End-to-End Spoken Dialogue](https://mlsanity.com/paper/2601.18281/) (2026)
- [Resurfacing Paralinguistic Awareness in Large Audio Language Models](https://mlsanity.com/paper/2603.11947/) (2026)
- [SpeechParaling-Bench: A Comprehensive Benchmark for Paralinguistic-Aware Speech Generation](https://mlsanity.com/paper/2604.20842/) (2026)
- [Hear2Act: Benchmarking When Prosody Should Change What an Assistant Does](https://mlsanity.com/paper/2608.19515/) (2026)
- [MP-Bench: Evaluating Voice Agents as a Multiparty Conversation Participant](https://mlsanity.com/paper/2609.13076/) (2026)
- [Aligning Paralinguistic Understanding and Generation in Speech LLMs via Multi-Task Reinforcement Learning](https://mlsanity.com/paper/2603.15981/) (2026)
- [TiCo: Time-Controllable Spoken Dialogue Model](https://mlsanity.com/paper/2603.22267/) (2026)
- [OmniACBench: A Benchmark for Evaluating Context-Grounded Acoustic Control in Omni-Modal Models](https://mlsanity.com/paper/2603.23938/) (2026)
- [The Sound of Absence: Audio-Language Embedding Models Struggle with Negation](https://mlsanity.com/paper/2607.12290/) (2026)
- [Do Audio Language Models Use Paralinguistic Evidence? Counterfactual Audits for Response Evaluation](https://mlsanity.com/paper/2608.06718/) (2026)
- [VoiceLongMemEval: Do Assistants Remember How You Sounded?](https://mlsanity.com/paper/2609.00570/) (2026)
- [SDiaReward: Modeling and Benchmarking Spoken Dialogue Rewards with Modality and Colloquialness](https://mlsanity.com/paper/2603.14889/) (2026)
- [ParaPairAudioBench: Paralinguistic Pairwise Audio Benchmark for LALM-as-a-Judge](https://mlsanity.com/paper/2606.24648/) (2026)
- [Towards High-Level Semantic Intelligence](https://mlsanity.com/paper/2607.24082/) (2026)
- [S3-Bench: Evaluating Speech Interaction Models as Scientific Voice Assistants](https://mlsanity.com/paper/2609.09852/) (2026)
- [NSV-Shift: A Contrastive Benchmark for Non-Speech Vocalization Understanding and Response Adaptation in Speech-to-Speech Models](https://mlsanity.com/paper/2609.33899/) (2026)
- [Personalization Without Stereotyping: Do Speech LLMs Know When to Turn a Deaf Ear?](https://mlsanity.com/paper/2610.13550/) (2026)

## Abstract (arXiv)

Speech-to-Speech (S2S) models have shown promising dialogue capabilities, but their ability to handle paralinguistic cues - such as emotion, tone, and speaker attributes - and to respond appropriately in both content and style remains under-explored. Progress is further hindered by the scarcity of high-quality and expressive demonstrations. To address this, we introduce a new reinforcement learning (RL) framework for paralinguistic-aware S2S, ParaS2S, which evaluates and optimizes both response content and speaking style directly at the waveform level. We first construct ParaS2SBench, a benchmark that evaluates the naturalness of input-output pairs in terms of content and speaking style using expressive and challenging queries. For the automatic judge, we propose a PolyTone training strategy and a multi-stage framework, preventing the style hallucination of end-to-end audio LLM judging. Our judge correlates well with human preferences and is scalable, enabling the model to interact and learn from unlabeled speech via RL. Experiments show that existing S2S models fail to respond appropriately to paralinguistic attributes, performing no better than pipeline-based baselines. Our RL approach (ParaS2SAlign) achieves a 10% relative improvement in the appropriateness of response content and speaking style on ParaS2SBench over supervised fine-tuning (SFT), surpassing all prior models while requiring substantially fewer paired demonstrations than pure SFT. Our findings highlight the need for a scalable and accurate automatic evaluator for speech-to-speech interaction.