Automatic Speech Recognition on SpeechIO
2.2Character Error Rate (CER)FireRedASR2S LLM
Evaluation results
| Method | Links | |
|---|---|---|
| FireRedASR2S LLMModel Size=8B+, Inference Mode=Offline2026.04 | 2.2 | |
| Qwen3-Omni InstructModel Size=30B-A3B, Inference Mode=Offline2026.04 | 2.33 | |
| Qwen3-ASR 1.7BModel Size=2.0B, Inference Mode=Offline2026.04 | 2.55 | |
| NIM4-ASRModel Size=2.3B, Inference Mode=Offline2026.04 | 2.61 | |
| Fun-ASR NanoModel Size=0.8B, Inference Mode=Offline2026.04 | 2.78 | |
| NIM4-ASRModel Size=2.3B, Inference Mode=Stream2026.04 | 2.84 | |
| Qwen3-ASR-1.7BParameters=1.7B2026.01 | 2.88 | |
| Doubao-ASR2026.01 | 2.93 | |
| GLM-ASR NanoModel Size=1.5B, Inference Mode=Offline2026.04 | 3.17 | |
| Step-Audio2 MiniModel Size=8B+, Inference Mode=Offline2026.04 | 3.41 | |
| Qwen3-ASR-0.6BParameters=0.6B2026.01 | 3.44 | |
| Gemini-2.5-Pro2026.01 | 5.3 | |
| Whisper-large-v32026.01 | 7.56 | |
| GPT-4o-Transcribe2026.01 | 12.86 |