AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Data-Centric Lessons To Improve Speech-Language Pretraining

Vishaal Udandarao, Zhiyun Lu, Xuankai Chang, Yongqiang Wang, Albin Madappally Jose, Fartash Faghri, Joshua P Gardner, Chung-Cheng Chiu

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Speech and audio language models

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
use fine instead of coarse speech-text interleaving (ablation)speech-text pretraining · SQA avg (SWQ/STQ/SLQ)accuracy (%): 37.6 -> 40.7Improved
mix in Krist synthetic data (ablation)web-crawl pretraining · SQA avg (SWQ/STQ/SLQ)accuracy (%): 40.7 -> 41.5Mixed
mix in Quest QA-format synthetic data (ablation)web-crawl pretraining · SQA avg (SWQ/STQ/SLQ)accuracy (%): 40.7 -> 47.9Improved
mix both Quest and Krist (token-proportional) (ablation)web-crawl pretraining · SQA avg (SWQ/STQ/SLQ)accuracy (%): 40.7 -> 42.1Improved
use deterministic instead of stochastic modality sampling (ablation)interleaved pretraining · SQA avg (SWQ/STQ/SLQ)accuracy (%): 41.4 -> 42.4Improved
deterministic sampling effect on text benchmarks (ablation)interleaved pretraining · CoreEN / MMLUaccuracy (%): 60.6 -> 60.1 CoreEN; 65.7 -> 65.2 MMLUHurt
apply all data interventions with audio-loss maskingunderstanding-only SpeechLM · SQA avg (SWQ/STQ/SLQ)accuracy (%): 42.5 -> 51.8Improved
mask loss on audio tokens (ablation)SpeechLM pretraining · SQA avg (SWQ/STQ/SLQ)accuracy (%): 42.4 -> 51.8Improved
fine interleaving after SFT (ablation)SpeechLM · spoken-alpacaLC win rate (%): 42.6 -> 44.3Improved
add Quest synthetic data after SFT (ablation)SpeechLM (fine) · spoken-alpacaLC win rate (%): 44.3 -> 47.4Improved
add synthetic data on audio-response Eval 5 (ablation)SpeechLM (fine) · Audio Eval 5win rate (%): 41.9 -> 39.3Hurt
fine interleaving reduces text-audio reverse KL (ablation)speech-text pretraining · Spoken-LLaMA-Questionsreverse-KL: 3.20 -> 2.21Improved

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