Reproduce “Swanada: Split-Encoder Capability-Routed Conformer CTC and Sub-30 MB On-Device Speech Recognition Across Twelve Scripts” 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
Hemanth HM
Swanada enables multilingual speech recognition in browser and edge runtimes by sharing an acoustic encoder across languages and switching small, native-script CTC heads. The runtime exposes the encoder’s shared representation and applies the selected head. On 480 held-out utterances, Studio matches monolithic checkpoints at 5.43% CER while reducing 12-language fleet storage 11.6×; this shows output parity, not improved accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.