Research map / Large language models / Language, safety and interpretability
Closing the Gap Between Text and Speech Understanding in LLMs
Santiago Cuervo, Skyler Seto, Maureen de Seyssel, Richard He Bai, Zijin Gu, Tatiana Likhomanenko, Navdeep Jaitly, Zakaria Aldeneh
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 changed | Where | Measured | Effect |
|---|---|---|---|
| train with distillation (alpha>0) instead of NLL on narrow speech data (ablation) | speech-adapted Qwen2.5-3B · LibriHeavy+Emilia | cross-modal misalignment | Improved |
| increase distillation weight alpha (ablation) | speech-adapted Qwen2.5-3B · LibriHeavy+Emilia | irreducible misalignment E: 0.32 -> 0.13 | Improved |
| use broad-domain FineWeb-Edu synthetic speech with NLL (alpha=0) (ablation) | speech-adapted Qwen2.5-3B · FineWeb-Edu test | misalignment | Mixed |
| add Stage II active selection after Stage I (ablation) | SALAD-3B · Spoken benchmarks avg | accuracy: 68.5 -> 72.0 | Improved |
| add Stage II active selection after Stage I (ablation) | SALAD-7B · Spoken benchmarks avg | accuracy: 73.9 -> 75.4 | Improved |
| active selection instead of uniform selection (ablation) | SALAD-3B · OBQA | accuracy: 71.9 -> 76.7 | Improved |
| active selection instead of uniform selection (ablation) | SALAD-3B · PIQA | accuracy: 78.1 -> 78.1 | No clear effect |
| use SALAD instead of Qwen2.5-Omni | speech-adapted LLM · Spoken benchmarks avg | accuracy: 76.7 vs 75.4 (7B) | Mixed |
Closest papers by meaning
- Geometric Alignment without Functional Equivalence: A Layer-wise Analysis of the Speech-Text Modality Gap (NeurIPS 2026 · similarity 0.90)
- Is Text All You Need? Text as a Universal Information Bottleneck for Speech LLMs (NeurIPS 2026 · similarity 0.89)
- Paralinguistics-Aware Speech-Empowered Large Language Models for Natural Conversation (NeurIPS 2024 · similarity 0.88)
- Scaling Speech-Text Pre-training with Synthetic Interleaved Data (ICLR 2025 · similarity 0.88)
- LLM Alignment--Utility Asymmetry under Semantic-Preserving Transformations (NeurIPS 2026 · similarity 0.88)
- metabench - A Sparse Benchmark of Reasoning and Knowledge in Large Language Models (ICLR 2025 · similarity 0.88)
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.
