Reproduce “Preserve to Expand: Source-Aware Adaptation and Model Merging for Multilingual LLM-Based 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
Yuchen Zhang, Haralambos Mouratidis, Ravi Shekhar
The authors study connector-only multilingual speech recognition expansion using source-language calibration to protect connector columns during target-language adaptation. The speech encoder and language model remain frozen, and the update is merged with the base. On Slavic languages, SSU-Wanda+TIES reduces source WER from 90.88% with standard TIES to 43.60%, while target WER rises from 37.03% to 51.69%, in Common Voice tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.