Skip to content

Model Options

Whisper Models

Available via the --model flag on extract.

Model Speed Accuracy Notes
tiny ⚡⚡⚡⚡⚡ ⭐⭐ Quick tests
base ⚡⚡⚡⚡ ⭐⭐⭐ Fast with decent quality
small ⚡⚡⚡ ⭐⭐⭐⭐ Good balance
medium ⚡⚡ ⭐⭐⭐⭐ Higher accuracy
large-v3 ⭐⭐⭐⭐⭐ Best accuracy (default)
turbo ⚡⚡⚡ ⭐⭐⭐⭐⭐ Fast + accurate

Device

Value Description
auto Detect best available device (default)
cpu Force CPU
cuda Force NVIDIA GPU

Compute Type

Value Requires Speed Notes
float32 CPU or GPU Slower Use on CPU
float16 GPU only Faster Default — will error on CPU
int8 CPU or GPU Fastest Slight accuracy tradeoff

Warning

float16 is the default but requires a GPU. Use --compute-type float32 on CPU systems.