We introduce SIFT (Speech Instruction Fine-Tuning), a 50M-example dataset designed for instruction fine-tuning and pre-training of speech-text large language models (LLMs). SIFT-50M is built from publicly available speech corpora, which collectively contain 14K hours of speech, and leverages LLMs along with off-the-shelf expert models. The dataset spans five languages, encompassing a diverse range of speech understanding as well as controllable speech generation instructions. Using SIFT-50M, we train SIFT-LLM, which outperforms existing speech-text LLMs on instruction-following benchmarks while achieving competitive performance on foundational speech tasks. To support further research, we also introduce EvalSIFT, a benchmark dataset specifically designed to evaluate the instruction-following capabilities of speech-text LLMs.
@article{arxiv.2504.09081,
title = {SIFT-50M: A Large-Scale Multilingual Dataset for Speech Instruction Fine-Tuning},
author = {Prabhat Pandey and Rupak Vignesh Swaminathan and K V Vijay Girish and Arunasish Sen and Jian Xie and Grant P. Strimel and Andreas Schwarz},
journal= {arXiv preprint arXiv:2504.09081},
year = {2025}
}