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

Estonian · 9 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 12.7212.04–13.43 13.2× 6.7× 1.10 GB CC-BY-4.0
2 parakeet-primeline 17.1716.36–18.00 12.5× 7.5× 0.72 GB CC-BY-4.0
3 parakeet-tdt-0.6b-v3 17.9617.06–18.80 12.5× 7.5× 0.72 GB CC-BY-4.0
4 whisper-large-v3 18.3017.41–19.25 2.1× 0.6× 1.55 GB Apache-2.0
5 whisper-large-v3-turbo 18.4417.50–19.37 3.4× 0.8× 0.83 GB Apache-2.0
6 whisper-medium 31.5230.39–32.63 4.3× 1.1× 0.77 GB Apache-2.0
7 nemotron-3.5-asr-streaming-0.6b 31.8430.63–32.85 14.5× 7.5× 0.70 GB OpenMDW-1.1
8 whisper-small 54.7953.58–55.93 12.1× 3.4× 0.25 GB Apache-2.0
9 Fun-ASR-MLT-Nano-2512! 64.2262.44–66.21 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1