--- license: other language: - zh - nan - hak - en - id base_model: openai/whisper-large-v2 pipeline_tag: automatic-speech-recognition tags: - dginfra - whisper - automatic-speech-recognition - taiwanese - hakka - indonesian - ctranslate2 - faster-whisper --- # Taiwan-Tongues-ASR-CE Model v2.0 Taiwan-Tongues-ASR-CE-v2.0 是 [Taiwan-Tongues-ASR-CE-pretrained-v2.0](https://huggingface.co/adi-gov-tw/Taiwan-Tongues-ASR-CE-pretrained-v2.0) 的 CTranslate2 / faster-whisper 推論版本。此版本新增印尼文(Indonesian, `id`)語音辨識能力,並持續支援華語、台語、客語與英語。 這個 repository 提供 `model.bin` 與 tokenizer 設定,適合使用 faster-whisper 進行部署或批次轉錄。若需要 Transformers checkpoint、繼續訓練或微調,請使用 [Taiwan-Tongues-ASR-CE-pretrained-v2.0](https://huggingface.co/adi-gov-tw/Taiwan-Tongues-ASR-CE-pretrained-v2.0)。 專案原始碼與訓練流程請見 [Taiwan-Tongues-ASR-CE](https://github.com/adi-gov-tw/Taiwan-Tongues-ASR-CE)。 ## Version | Version | Base model | Format | Main update | |---|---|---|---| | v2.0 | `openai/whisper-large-v2` | CTranslate2 (`model.bin`) | Added Indonesian (`id`) support | ## Supported Languages | Language | Code | |---|---| | Mandarin Chinese | `zh` | | Taiwanese Hokkien | `nan` | | Hakka | `hak` | | English | `en` | | Indonesian | `id` | ## Files This repository contains the CTranslate2 / faster-whisper inference assets: - `model.bin` - `config.json` - `tokenizer.json` - `tokenizer_config.json` - `vocabulary.json` ## Usage Install faster-whisper: ```bash pip install faster-whisper ``` Run transcription: ```python from faster_whisper import WhisperModel model = WhisperModel( "adi-gov-tw/Taiwan-Tongues-ASR-CE-v2.0", device="cuda", compute_type="float16", ) segments, info = model.transcribe( "audio.wav", language="id", task="transcribe", ) for segment in segments: print(segment.text) ``` Change `language` to `zh`, `nan`, `hak`, `en`, or `id` according to the input audio. For CPU inference, set `device="cpu"` and choose an appropriate `compute_type`, such as `"int8"`. ## Notes - This is an inference-only CTranslate2 / faster-whisper package. - For the Transformers checkpoint or fine-tuning workflow, use [Taiwan-Tongues-ASR-CE-pretrained-v2.0](https://huggingface.co/adi-gov-tw/Taiwan-Tongues-ASR-CE-pretrained-v2.0). - The license is listed as `other`; please refer to the project and release documentation before redistribution or commercial use.
