Closing the Modality Reasoning Gap for Speech Large Language Models
This addresses a key limitation in speech AI by improving reasoning accuracy for speech-based models, though it is incremental as it builds on existing methods to close a specific gap.
The paper tackled the modality reasoning gap in speech large language models, where reasoning on speech inputs is weaker than on text, by introducing TARS, a reinforcement-learning framework that aligns text- and speech-conditioned trajectories, and it achieved state-of-the-art performance among 7B-scale Speech LLMs on benchmarks like MMSU and OBQA.
Although speech large language models have achieved notable progress, a substantial modality reasoning gap remains: their reasoning performance on speech inputs is markedly weaker than on text. This gap could be associated with representational drift across Transformer layers and behavior deviations in long-chain reasoning. To address this issue, we introduce TARS, a reinforcement-learning framework that aligns text-conditioned and speech-conditioned trajectories through an asymmetric reward design. The framework employs two dense and complementary signals: representation alignment, which measures layer-wise hidden-state similarity between speech- and text-conditioned trajectories, and behavior alignment, which evaluates semantic consistency between generated outputs and reference text completions. Experiments on challenging reasoning benchmarks, including MMSU and OBQA, show that our approach significantly narrows the modality reasoning gap and achieves state-of-the-art performance among 7B-scale Speech LLMs.