Typhoon Whisper Turbo Mlx

--- license: mit language: - th - en tags: - whisper - mlx - thai - speech-recognition - asr base_model: typhoon-ai/typhoon-whisper-turbo library_name: mlx pipeline_tag: automatic-speech-recognition --- # typhoon-whisper-turbo-mlx An [MLX](https://github.com/ml-explore/mlx) (Apple Silicon) conversion of [**typhoon-ai/typhoon-whisper-turbo**](https://huggingface.co/typhoon-ai/typhoon-whisper-turbo) — SCB 10X's Thai fine-tune of OpenAI Whisper large-v3-turbo (~800M params). float16 weights, ready for `mlx-whisper`. ## Usage ```python import mlx_whisper r = mlx_whisper.transcribe("audio.wav", path_or_hf_repo="chayapats/typhoon-whisper-turbo-mlx") print(r["text"]) ``` Runs fully on-device on Apple Silicon — no network, no API key. ## Used by [**OLIV**](https://github.com/chayapats/oliv) — free, open-source (MIT) Thai dictation for macOS. Press a hotkey, speak Thai or English, and the text lands in whatever app you're in. This model is its speech-to-text engine, running locally on the Mac. This conversion exists because of OLIV: in its benchmark, `typhoon-whisper-turbo` runs ahead of a Whisper-large-v3 Thai fine-tune at roughly half the size, with faster decode. Numbers, audio samples, and the full methodology are published here — nothing is hand-picked: 📊 [**OLIV benchmark & methodology**](https://chayapats.github.io/oliv/) ## License & attribution MIT, inherited from typhoon-ai/typhoon-whisper-turbo (SCB 10X) and OpenAI Whisper. This repository redistributes an MLX-format conversion of those weights; all credit for the model belongs to the original authors. See `LICENSE`.

Source context: 63 downloads · 0 likes · Pipeline automatic-speech-recognition · Library mlx · Repo chayapats/typhoon-whisper-turbo-mlx