TEA-ASR — Taiwan Mandarin ASR 🍵

TEA-ASR (Taiwan Everyday Audio) — Traditional-script + Taiwanese-lexicon ASR with robust Mandarin–English code-switch, adapted from Qwen3-ASR with a tokenizer-first procedure. Output is Traditional Chinese. Set the language hint to Chinese for Taiwan speech (best results).

Pick a model: TEA-ASR-1.1 / -mini give the best recognition (incl. dense code-switch). The -fmt variants add a working numeral-style dial — pick 123 (Arabic) or 一二三 (Chinese) and the same audio is rendered in that convention (verified flip control; strongest on multi-digit numbers) — best for formatting-sensitive, Mandarin-dominant audio (subtitles, records). Keep English biases code-switch English toward verbatim transcription (soft control). Main-line models ignore the format controls.