Audio Frontend Inference on LibriSpeech 160s duration (test)
0.076Latency (ms)MelT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MelTPlatform=H100 80GB, Backend=Fastest available, Context Length=160s2026.05 | 0.076 | 1.92 | 23.2 | 2.74 | — | 309 | |
| STFT+MelPlatform=H100 80GB, Backend=Fastest available, Context Length=160s2026.05 | 0.145 | — | 63.4 | — | 438.1 | — | |
| MelTPlatform=V100 32GB, Backend=Fastest available, Context Length=160s2026.05 | 0.345 | 1.36 | 90.9 | 1.2 | — | 263.4 | |
| STFT+MelPlatform=V100 32GB, Backend=Fastest available, Context Length=160s2026.05 | 0.468 | — | 109.5 | — | 233.9 | — | |
| MelTPlatform=M4 Pro, Backend=Fastest available, Context Length=160s2026.05 | 0.914 | 1.49 | 13.5 | 3.4 | — | 16 | |
| STFT+MelPlatform=M4 Pro, Backend=Fastest available, Context Length=160s2026.05 | 1.362 | — | 45.8 | — | 17.3 | — | |
| MelTPlatform=A18 Pro, Backend=Fastest available, Context Length=160s2026.05 | 2.664 | 3.75 | 10.6 | 3.52 | — | 3.71 | |
| STFT+MelPlatform=A18 Pro, Backend=Fastest available, Context Length=160s2026.05 | 9.981 | — | 37.3 | — | 3.68 | — |