Reproduce “LLM-HB: Language-Aware LLM-Guided Hotword Biasing for Code-Switching ASR” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuxuan He, Genshun Wan, Pengcheng Li, Gongping Huang, Jianqing Gao, Yanmin Qian
LLM-HB improves code-switching speech recognition by combining language-aware speech representations with an LLM prompt containing supplied hotwords. It uses a Whisper encoder, an MoE adaptor, and a LoRA-adapted Qwen3-4B language model. On ASRU2019-CS with 15 distractor hotwords after about 2,200 hours of joint training, it achieves 5.85% MER versus 7.34% for its baseline, within those conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.