best model by language
Benchmarked on the FLEURS test split at Q8_0. Error is CER — lower is better. Speed is × realtime on Handy target hardware.
accuracy vs speed
Cantonese · 6 models
AMD Ryzen 4750U GPU (Vulkan)
pareto frontier (no model is both faster and more accurate) other models
highlighted rows are recommended picks or on pareto frontier
| | | | | | |
| 1 | Qwen3-ASR-1.7Baccurate pick | 6.13 | 4.1× | 2.5× | 2.04 GB | Apache-2.0 |
| 2 | Qwen3-ASR-0.6Boverall pick | 7.91 | 9.7× | 5.3× | 0.79 GB | Apache-2.0 |
| 3 | Fun-ASR-MLT-Nano-2512!fast pick | 12.72 | 8.8× | 5.5× | 0.83 GB | FunASR Model Open Source License Agreement v1.1 |
| 4 | whisper-large-v3 | 22.06 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 5 | whisper-large-v3-turbo | 34.62 | 3.1× | 1.4× | 0.83 GB | Apache-2.0 |
| 6 | SenseVoiceSmall | 37.44 | 27.7× | 19.7× | 0.24 GB | model-license (FunASR MODEL_LICENSE) |