Reproduce “Understanding Personalization Gains in Whisper 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
Sergei Katkov, Antonio Liotta
The authors compare zero-shot Whisper with pooled and group-specific LoRA adaptation to test whether speech-recognition errors require personalization. On Common Voice, pooled LoRA trains one adapter across selected groups, while personalized LoRA trains separate adapters. For Whisper medium on the hardest 10% of groups, gains are 6.66 versus 5.06 percentage points; this diagnostic result does not show pooled adaptation always wins.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.