Efficient Chain-of-Modality Reasoning via Progressive Compression for Spoken Language Models

2026-07-22

AI Summary

This work addresses the limited reasoning capabilities of current speech language models in spoken mathematical question answering, primarily due to the ambiguity of spoken mathematical expressions and the difficulty of transferring reasoning mechanisms from text-based large language models. To overcome these challenges, the authors propose an Efficient Chain-of-Multimodal Reasoning framework (ECoM Reasoning), which introduces compressed reasoning into speech language models for the first time. Leveraging a progressive compression training strategy, ECoM Reasoning jointly preserves speech grounding and logical inference while compressing intermediate textual representations. Using only 40% of the original text tokens, the method achieves a 21% accuracy improvement over standard Chain-of-Multimodal (CoM) reasoning without explicit intermediate steps and a 3% gain over full CoM reasoning trajectories on spoken math QA benchmarks, demonstrating significantly more efficient yet highly accurate speech-based reasoning.

Abstract

Spoken language models (SLMs) enable natural human-computer interaction, but their reasoning ability still lags behind that of text-based large language models, especially on spoken mathematical question answering tasks. One important reason is that SLMs reason over purely verbalized mathematical expressions, which are harder to interpret than symbolic text. However, directly transferring text-based reasoning to SLMs is nontrivial due to architectural constraints and the additional computational requirements. To address this challenge, we propose Efficient Chain-of-Modality Reasoning (ECoM Reasoning), the first framework to introduce compressed reasoning into SLMs. By compressing the textual component so that it jointly serves as speech guidance and reasoning representation, ECoM Reasoning improves reasoning accuracy while using a smaller token budget than the standard Chain-of-Modality (CoM) architecture, which generates intermediate text before speech. To train this capability, we further propose Progressive Compression, a curriculum-based strategy that gradually trains the model from full-form reasoning to compressed reasoning. Experiments on spoken mathematical question answering benchmarks show that ECoM Reasoning improves accuracy by 21% over standard CoM without explicit reasoning, and by 3% over CoM with full reasoning traces while using only 40% of the text tokens, demonstrating that it enhances SLM reasoning while remaining inference-efficient.

Problem

• Spoken Language Models
• Reasoning
• Mathematical Question Answering
• Chain-of-Modality
• Efficiency

Innovation

• Efficient Chain-of-Modality Reasoning
• Progressive Compression
• Spoken Language Models
• Compressed Reasoning
• Token Efficiency