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

Lithuanian · 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 13.4512.74–14.16 13.2× 6.7× 1.10 GB CC-BY-4.0
2 parakeet-tdt-0.6b-v3 22.2021.36–23.02 12.5× 7.5× 0.72 GB CC-BY-4.0
3 parakeet-primeline 23.0822.26–23.89 12.5× 7.5× 0.72 GB CC-BY-4.0
4 whisper-large-v3 24.5523.68–25.41 2.1× 0.6× 1.55 GB Apache-2.0
5 whisper-large-v3-turbo 25.1124.23–26.06 3.4× 0.8× 0.83 GB Apache-2.0
6 whisper-medium 43.3442.26–44.40 4.3× 1.1× 0.77 GB Apache-2.0
7 whisper-small 70.2168.92–71.71 12.1× 3.4× 0.25 GB Apache-2.0
8 Fun-ASR-MLT-Nano-2512! 78.4276.74–80.45 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1