Metro-ASR is a novel speech recognition architecture designed specifically for Egyptian Arabic with native English code-switching support. It uses a CTC-based non-autoregressive approach for fast, accurate, on-device inference — no decoder, no beam search required.
