best model by language

Benchmarked on the FLEURS test split at Q8_0. Error is WER — lower is better. Speed is × realtime on Handy target hardware.

accuracy vs speed

Slovenian · 8 models · left is slower, down is better · log speed axis

AMD Ryzen 4750U GPU (Vulkan)
pareto frontier (no model is both faster and more accurate) other models
1 canary-1b-v2!overall pickmost accuratefast pick 12.7611.35–14.61 13.2× 6.7× 1.10 GB CC-BY-4.0
2 whisper-large-v3 19.1518.25–20.02 2.1× 0.6× 1.55 GB Apache-2.0
3 whisper-large-v3-turbo 20.5619.59–21.52 3.4× 0.8× 0.83 GB Apache-2.0
4 parakeet-tdt-0.6b-v3 24.3023.22–25.50 12.5× 7.5× 0.72 GB CC-BY-4.0
5 whisper-medium 33.9832.85–34.99 4.3× 1.1× 0.77 GB Apache-2.0
6 parakeet-primeline 51.0749.27–52.72 12.5× 7.5× 0.72 GB CC-BY-4.0
7 whisper-small 52.7051.43–53.95 12.1× 3.4× 0.25 GB Apache-2.0
8 Fun-ASR-MLT-Nano-2512! 75.2273.63–76.87 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1