Thinking in Depth, Speaking Directly: Recurrent Latent Reasoning for Paralinguistically Grounded Spoken Dialogue

2026-09-29

AI Summary

This study addresses the disconnection between perception and reasoning in empathetic spoken dialogue, as well as the high latency induced by explicit chain-of-thought (CoT) prompting. To this end, it proposes LoopSLM, a framework that employs a recurrent Transformer architecture for latent-space reasoning. It introduces a novel recurrent decoding block reuse mechanism to enhance acoustic grounding and adopts a two-stage decoupled training strategy that separates reasoning from response learning, thereby eliminating reliance on explicit CoT entirely. Experimental results demonstrate that LoopSLM improves reasoning accuracy by over 20 percentage points, reduces generated tokens by 64.5%, and halves inference latency. Furthermore, it outperforms Qwen3-Omni-Thinking across most evaluation metrics, achieving efficient and direct empathetic speech interaction.

Abstract

Empathetic spoken dialogue requires models to use both what is said and how it is said to decide how to respond. Explicit CoT can improve paralinguistic perception and make acoustic cues more explicit in replies, yet does not ensure their effective use in response planning. We call this mismatch the perception-reasoning gap. In addition, CoT may not fully capture acoustic cues in words, and generating it adds inference latency. To address these limitations, we introduce LoopSLM, which builds on looped Transformers for latent reasoning, reusing a decoder block to refine hidden states with acoustic grounding at every pass. Its two-stage training further narrows the perception-reasoning gap by separating learning to reason from learning to respond, enabling direct inference without CoT. On EchoMind, LoopSLM improves paralinguistic understanding, reasoning, and reply quality over Qwen2.5-Omni-7B. Against the CoT-SFT baseline, LoopSLM gains over 20 points in reasoning accuracy while generating 64.5% fewer tokens at half the latency. It also outperforms Qwen3-Omni-Thinking on most empathetic reply metrics with 34x lower latency. Despite training only on dialogue data, LoopSLM improves accuracy on general audio benchmarks.

Problem

• empathetic spoken dialogue
• paralinguistic perception
• perception-reasoning gap
• Chain-of-Thought
• inference latency

Innovation

• Latent Reasoning
• Looped Transformers
• Paralinguistic Grounding
• Two-stage Training
• Spoken Dialogue