Settings

Providers, models, clip defaults and pipeline tuning.

Transcription

Whisper model and speed/accuracy tuning.

Model

Backend

Compute type

Beam size1 = greedy/fastest; higher = more accurate, slower.

Voice activity detection (skip silence)

CPU threads0 = use all cores.Language hintISO 639-1 code (en, es, …) or blank to auto-detect.Device

Applies to faster-whisper; openai-whisper auto-detects on its own.