Speech Recognition on Switchboard
6.6WERAR Beam
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AR Beam2026.06 | 6.6 | — | — | — | |
| AR Greedy2026.06 | 6.9 | — | — | — | |
| Audio Flamingo 3 + CoATBackbone=Audio Flamingo 3, Method=CoAT2026.06 | 7.18 | — | — | — | |
| NAR-MBR|Z|=64, Niter=12026.06 | 7.3 | — | — | — | |
| NAR-MBR|Z|=256, Niter=12026.06 | 7.3 | — | — | — | |
| NAR-MBR|Z|=256, Niter=102026.06 | 7.3 | — | — | — | |
| Qwen2.5-Omni-7B + CoATBackbone=Qwen2.5-Omni-7B, Method=CoAT2026.06 | 7.35 | — | — | — | |
| NAR-MBR|Z|=64, Niter=102026.06 | 7.4 | — | — | — | |
| NARNiter=102026.06 | 7.6 | — | — | — | |
| NAR-MBR|Z|=256, Niter=02026.06 | 7.7 | — | — | — | |
| NARNiter=12026.06 | 7.8 | — | — | — | |
| NAR-MBR|Z|=64, Niter=02026.06 | 7.8 | — | — | — | |
| NARNiter=02026.06 | 7.9 | — | — | — | |
| Audio Flamingo 3Backbone=Audio Flamingo 3, Method=Baseline2026.06 | 8.01 | — | — | — | |
| Qwen2-Audio + CoATBackbone=Qwen2-Audio, Method=CoAT2026.06 | 9.25 | — | — | — | |
| ECLMCorrection model size=0.5B, Decoding strategy=greedy, Base model=Parakeet-TDT-v22024.05 | 10.3 | — | — | 7 | |
| Parakeet-TDT-v2Correction method=None2024.05 | 11.1 | — | — | — | |
| Seed-ASR (ML)Language=EN2024.07 | 11.59 | — | — | — | |
| WhisperBackbone=Large V2, Evaluation Protocol=Zero-shot2022.12 | 13.8 | 51.2 | — | — | |
| Whisper Large-v2Language=EN2024.07 | 13.8 | — | — | — | |
| Qwen2.5-Omni-7BBackbone=Qwen2.5-Omni-7B, Method=Baseline2026.06 | 14.83 | — | — | — | |
| KM 25HzToken Rate=252026.01 | 15 | — | 9.5 | — | |
| KM 12.5HzToken Rate=12.52026.01 | 17.3 | — | 11.2 | — | |
| Kanade 25HzToken Rate=252026.01 | 18.6 | — | 11.7 | — | |
| Qwen2-AudioBackbone=Qwen2-Audio, Method=Baseline2026.06 | 19.06 | — | — | — | |
| Kanade 12.5HzToken Rate=12.52026.01 | 24.6 | — | 15.9 | — | |
| FACodecToken Rate=4802026.01 | 25.5 | — | 16.4 | — | |
| DualCodecToken Rate=1002026.01 | 28.2 | — | 18.1 | — | |
| wav2vec 2.0Backbone=Large, Language Model=None2022.12 | 28.3 | — | — | — | |
| PASTToken Rate=4002026.01 | 28.9 | — | 18.8 | — | |
| TiCodecToken Rate=3002026.01 | 29.1 | — | 18.9 | — | |
| STToken Rate=4002026.01 | 29.4 | — | 19.1 | — | |
| MimiToken Rate=1002026.01 | 30.6 | — | 20.1 | — | |
| StableCodecToken Rate=252026.01 | 45 | — | 30.2 | — | |
| WavTokenizerToken Rate=402026.01 | 67.2 | — | 46.8 | — | |
| X-Codec 2Token Rate=502026.01 | 103.3 | — | 75.6 | — | |
| BiCodecToken Rate=502026.01 | 108.8 | — | 78.4 | — |