Automatic Speech Recognition on Archi (evaluation)
0.402WERwhisper-large-v3
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| whisper-large-v3Params.=1.5B, Tunable=1.5B2026.04 | 0.402 | 0.099 | 0.107 | |
| whisper-large-v3-cyrlParams.=1.5B, Tunable=1.5B2026.04 | 0.422 | 0.111 | 0.119 | |
| w2v2l-custom-cpy1Params.=0.3B, Tunable=0.3B2026.04 | 0.462 | 0.116 | 0.123 | |
| w2v2l-custom-avg-lmParams.=0.3B, Tunable=0.3B2026.04 | 0.465 | 0.116 | 0.122 | |
| w2v2l-custom-avgParams.=0.3B, Tunable=0.3B2026.04 | 0.479 | 0.116 | 0.122 | |
| Qwen2-Audio-7B-Instruct-cyrlParams.=8.4B, Tunable=0.7B2026.04 | 0.539 | 0.156 | 0.166 | |
| wav2vec2-large-ipaParams.=0.3B, Tunable=0.3B2026.04 | 0.559 | 0.128 | 0.135 | |
| Qwen2-Audio-7B-InstructParams.=8.4B, Tunable=0.7B2026.04 | 0.579 | 0.163 | 0.18 | |
| w2v2l-customParams.=0.3B, Tunable=0.3B2026.04 | 0.593 | 0.138 | 0.147 | |
| Qwen2.5-Omni-7BParams.=10.8B, Tunable=0.7B2026.04 | 0.705 | 0.184 | 0.199 | |
| Qwen2.5-Omni-7B-cyrlParams.=10.8B, Tunable=0.7B2026.04 | 0.904 | 0.295 | 0.291 | |
| gpt-4o-transcribeParams.=-, Tunable=-2026.04 | 0.982 | 0.435 | 0.436 | |
| wav2vec2-large-ipa-zsParams.=0.3B, Tunable=-2026.04 | 1 | 0.593 | 0.606 | |
| w2v2l-custom-avg-zsParams.=0.3B, Tunable=-2026.04 | 1 | 0.544 | 0.558 |