pyVideoTrans

Appearance

What is the HuggingFace_ASR Speech Recognition Service? ​

HuggingFace_ASR is a speech recognition service added in pyVideoTrans v3.91. It supports using open-source models from HuggingFace for speech recognition. This service covers models for various languages — suitable for non-Chinese speech recognition (English, Japanese, Vietnamese, Thai, etc.).

Prerequisites ​

  • pyVideoTrans version ≥ v3.91
  • First use of a model requires downloading from HuggingFace (internet connection required)
  • Chinese users may need to use a mirror site or download manually (see below)
  • GPU acceleration is recommended for significantly faster recognition (NVIDIA GPU)

Auto-download ​

On first use, the software automatically downloads from:

  • International: https://huggingface.co
  • Chinese mirror: https://hf-mirror.com

Due to network conditions, automatic download may fail for Chinese users. If it fails, see the Manual Download section below.

Supported Models & Languages ​

English ​

ModelLanguageNotes
nvidia/parakeet-ctc-1.1bEnglishNVIDIA, high accuracy

Japanese ​

ModelLanguageNotes
reazon-research/japanese-wav2vec2-large-rs35khJapanesewav2vec2 architecture
kotoba-tech/kotoba-whisper-v2.0JapaneseWhisper optimized for Japanese

Vietnamese ​

ModelLanguageNotes
vinai/Phowhisper-largeVietnameseWhisper-large-v3 fine-tuned

Thai ​

ModelLanguageNotes
biodatlab/whisper-th-large-v3ThaiWhisper-large-v3 Thai version

Manual Download ​

If automatic download fails, download the model files manually. All models follow the same storage rules:

  1. Create the corresponding folder inside the models directory (same level as sp.exe / sp.py)
  2. Open the model download page and download all files
  3. Place the downloaded files into the created folder

Note: Do not rename files after downloading. If the download directory already has files with the same name (e.g., xxx(1)), delete the old files first, then rename.

nvidia/parakeet-ctc-1.1b (English) ​

  1. Create folder: models/models--nvidia--parakeet-ctc-1.1b
  2. URL: https://huggingface.co/nvidia/parakeet-ctc-1.1b/tree/main
  3. Download all files from this page and place them in the folder

reazon-research/japanese-wav2vec2-large-rs35kh (Japanese) ​

  1. Create folder: models/models--reazon-research--japanese-wav2vec2-large-rs35kh
  2. URL: https://huggingface.co/reazon-research/japanese-wav2vec2-large-rs35kh/tree/main
  3. Download all files and place them in the folder

kotoba-tech/kotoba-whisper-v2.0 (Japanese) ​

  1. Create folder: models/models--kotoba-tech--kotoba-whisper-v2.0
  2. URL: https://huggingface.co/kotoba-tech/kotoba-whisper-v2.0/tree/main
  3. Download all files and place them in the folder

biodatlab/whisper-th-large-v3 (Thai) ​

  1. Create folder: models/models--biodatlab--whisper-th-large-v3
  2. URL: https://huggingface.co/biodatlab/whisper-th-large-v3/tree/main
  3. Download all files and place them in the folder

vinai/Phowhisper-large (Vietnamese) ​

  1. Create folder: models/models--vinai--Phowhisper-large
  2. URL: https://huggingface.co/vinai/Phowhisper-large/tree/main
  3. Download all files and place them in the folder

openai/whisper-large-v3 ​

  1. Create folder: models/models--openai--whisper-large-v3
  2. URL: https://huggingface.co/openai/whisper-large-v3/tree/main
  3. Download all files and place them in the folder

zai-org/GLM-ASR-Nano-2512 ​

  1. Create folder: models/models--zai-org--GLM-ASR-Nano-2512
  2. URL: https://huggingface.co/zai-org/GLM-ASR-Nano-2512/tree/main
  3. Download all files and place them in the folder

anke01/whisper-small-uyghur ​

  1. Create folder: models/models--anke01--whisper-small-uyghur
  2. URL: https://huggingface.co/anke01/whisper-small-uyghur/tree/main
  3. Download all files and place them in the folder

Common Errors & Troubleshooting ​

Auto-download failed ​

  • Cause: Cannot reach HuggingFace or mirror site is unstable
  • Solution: Use the manual download method above, or configure a proxy and retry

Incorrect model file names ​

  • Cause: Browser auto-renamed files during download (e.g., xxx(1))
  • Solution: Remove files with (1) suffix, place correctly named files in the model folder

Empty recognition results or errors ​

  • Cause: Incomplete model files or incorrect folder structure
  • Solution: Check that all required files are in the model folder — re-download if necessary