Custom models

[raivisdejus]

Turbo model

Turbo models run much faster and require much less VRAM, but may be a bit less accurate.

The following turbo models are available:

• Huggingface whisper - openai/whisper-large-v3-turbo
• Faster whisper - deepdml/faster-whisper-large-v3-turbo-ct2

Custom language models

Below is a list of custom models to use for different languages and whisper types.

Latvian

• Huggingface - AiLab-IMCS-UL/whisper-large-v3-lv-late-cv19
  • Suitable for transcription and subtitle generation on GPU
• Faster whisper
  • RaivisDejus/whisper-large-v3-lv-ailab-ct2 - larger model, requires 6+ GB of GPU memory
  • RaivisDejus/whisper-large-v3-lv-ailab-ct2-int8 - smaller model, Requires 4+ GB of GPU memory
• Whisper.cpp - https://huggingface.co/AiLab-IMCS-UL/whisper-large-v3-lv-late-cv19/resolve/main/ggml-model.bin?download=true
  • Suitable for transcription and subtitle generation on CPU and GPU

Finnish

• Huggingface - Finnish-NLP/whisper-large-finnish-v3
• Faster whisper - Finnish-NLP/whisper-large-finnish-v3-ct2

Chinease

• Huggingface - WANGTINGTING/whisper-large-v2-zh-TW
• Faster whisper - XA9/faster-whisper-large-v2-zh-TW