Downloads · 30 days
19
58% of all-time downloads
rhap-llm-audio/faster-Taiwan-Tongues-ASR-CE-pretrained
faster-Taiwan-Tongues-ASR-CE-pretrained is a automatic speech recognition model from rhap-llm-audio. Use it when you need speech turned into text. It is set up for ctranslate2. The card lists the license as other.
Converted faster-whisper / CTranslate2 model for OpenAI Whisper.
Downloads · 30 days
19
58% of all-time downloads
All-time downloads
33
Public
Repo size
3.1 GB
Likes
0
Public
Click a slice to open those files.
.bin3.1 GB · 100%
From the Hugging Face model README
Converted faster-whisper / CTranslate2 model for OpenAI Whisper.
rhap-llm-audio/Taiwan-Tongues-ASR-CE-pretrainedrhap-llm-audio/faster-Taiwan-Tongues-ASR-CE-pretrainedfloat32 (see CTranslate2 quantization)cpuwhisper-to-faster-whisperfrom faster_whisper import WhisperModel
model = WhisperModel(
"rhap-llm-audio/faster-Taiwan-Tongues-ASR-CE-pretrained",
device="cpu",
compute_type="float32",
)
segments, info = model.transcribe("audio.mp3")
for seg in segments:
print(f"[{seg.start:.2f} → {seg.end:.2f}] {seg.text}")
This model card includes the metadata block speaches filters on
(library_name: ctranslate2, pipeline_tag: automatic-speech-recognition,
language:). After a regular POST /v1/models download, speaches will
list this in GET /v1/models.
curl -X POST http://localhost:8000/v1/models \
-H 'Content-Type: application/json' \
-d '{"model": "rhap-llm-audio/faster-Taiwan-Tongues-ASR-CE-pretrained"}'
curl http://localhost:8000/v1/audio/transcriptions \
-F "[email protected]" \
-F "model=rhap-llm-audio/faster-Taiwan-Tongues-ASR-CE-pretrained"
| File | Purpose |
|---|---|
model.bin | CTranslate2 weights (float32) |
config.json | faster-whisper model config |
vocabulary.json | tokenizer vocab |
tokenizer.json | HF tokenizer |
preprocessor_config.json | whisper feature-extractor config |
README.md | this card (speaches filter requires it) |
Inherited from upstream rhap-llm-audio/Taiwan-Tongues-ASR-CE-pretrained. See source repo for the original model
license. The conversion process itself adds no additional licensing terms.
本模型為 Taiwan-Tongues-ASR-CE 專案所使用的 Automatic Speech Recognition (ASR) 模型,配合對應的語料庫進行預訓練與微調。
模型包含以下檔案:
config.json — 模型結構與超參數設定model.bin — 預訓練權重preprocessor_config.json — 音訊前處理設定tokenizer.json — 分詞器設定vocabulary.json — 詞彙表請參考 Taiwan-Tongues-ASR-CE 專案最新且完整訓練方法及步驟
請先安裝 Python 3.8+ 及以下套件(建議使用虛擬環境):
pip install torch transformers datasets evaluate
依照 專案資料集格式 放置語音資料與標註檔案。
train/train-000000.tar...) 搭配 metadata,能更方便地進行流式訓練。train.tsv / test.tsv 載入,但不建議。可直接執行 run.sh,或根據需求修改參數:
bash run.sh
主要參數說明:
--model_name_or_path:預訓練模型名稱(如 openai/whisper-large-v3,或本模型路徑 ./)--corpus_data_dir:語料資料夾(如 sample_corpus)--dataset_config_name:資料集組合(如 train_ds_01+train_ds_02)--language:語言代碼(如 zh、nan、hak、en)run.sh 及 train_asr.py。訓練完成後,模型與相關設定會儲存在 output/ 目錄:
output/
├── config.json
├── model.bin
├── preprocessor_config.json
├── tokenizer.json
├── vocabulary.json
已訓練好的國語、台語、客語、英語模型存放於 models/ 目錄,包含:
model.bin:模型權重config.json、preprocessor_config.json、tokenizer.json、vocabulary.json這些檔案可以直接放入 Hugging Face transformers 進行推論或微調。
歡迎研究人員與開發者使用本模型,並透過 Issue 或 Pull Request 回饋改進。
本模型依照 Taiwan-Tongues-ASR-CE 專案授權條款釋出,使用者需遵守相關規範。