services.wyoming.faster-whisper.servers.<name>.model

Name of the voice model to use. Can also be a HuggingFace model ID or a path to a custom model directory.

With useTranformers enabled, a HuggingFace transformers Whisper model ID from HuggingFace like openai/whisper-tiny.en must be used.

Compressed models (int8) are slightly less accurate, but smaller and faster. Distilled models are uncompressed and faster and smaller than non-distilled models.

Available models:

• tiny-int8 (compressed)
• tiny
• tiny.en (English only)
• base-int8 (compressed)
• base
• base.en (English only)
• small-int8 (compressed)
• distil-small.en (distilled, English only)
• small
• small.en (English only)
• medium-int8 (compressed)
• distil-medium.en (distilled, English only)
• medium
• medium.en (English only)
• large
• large-v1
• distil-large-v2 (distilled, English only)
• large-v2
• distil-large-v3 (distilled, English only)
• large-v3
• turbo (faster than large-v3)

Type

string

Default

"tiny-int8"

Example

"Systran/faster-distil-whisper-small.en"

Declared

<nixpkgs/nixos/modules/services/home-automation/wyoming/faster-whisper.nix>