AI Paper Atlas

Research map / Large language models / Alignment and preferences

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

Shu-wen Yang, Ming Tu, Andy T. Liu, Xinghua Qu, Hung-yi Lee, Lu Lu, Yuxuan Wang, Yonghui Wu

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward modeling for diverse preferences · 24 citations (OpenAlex)

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 changedWhereMeasuredEffect
add GRPO RL with the distilled reward model after SFTKimi-Audio SFT · ParaS2SBench (all)ParaS2SBench score: 3.955 -> 4.382Improved
add SFT warmup and GRPO to the base Kimi-Audio modelKimi-Audio · ParaS2SBench (all)ParaS2SBench score: 2.350 -> 4.382Improved
add GRPO on top of SFT, judged by humansKimi-Audio SFT · Human evaluation (all)mean opinion score: 3.996 -> 4.303Improved
use multi-stage judge with ground-truth labels instead of end-to-end audio LLM judge (ablation)gpt-audio judge · Judge-human correlationPearson correlation (avg): 0.618 -> 0.776Improved
describe output style with a natural-language tone description (ablation)multi-stage judge O1 · Judge-human correlationPearson correlation (avg): 0.757 -> 0.776Improved
add output emotion tags instead of tone descriptions (ablation)multi-stage judge O2 · Judge-human correlationPearson correlation (avg): 0.776 -> 0.731Hurt
add both emotion tag and tone description for output (ablation)multi-stage judge O2 · Judge-human correlationPearson correlation (avg): 0.776 -> 0.743Hurt
replace ground-truth input labels with captioner predictions (with PolyTone)multi-stage judge O2 · Judge-human correlationPearson correlation (avg): 0.776 -> 0.723Hurt
add PolyTone training to the input style captioners (ablation)multi-stage judge · Judge-human correlationPearson correlation (avg): 0.701 -> 0.723Improved
add PolyTone training to the captioners (ablation)multi-stage judge · Judge-human correlation (emotion)Pearson correlation: 0.645 -> 0.684Improved
add PolyTone training to the captioners (ablation)multi-stage judge · Judge-human correlation (sarcasm)Pearson correlation: 0.669 -> 0.703Improved
RL with the judge on SFT models trained with 10 to 100 hours of data (ablation)Kimi-Audio SFT · SFT data regimesParaS2SBench scoreImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most 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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy