| # | Start Time | End Time | Japanese Subtitle Text | Action |
|---|
Local speech-to-text accuracy matrix (Japanese dataset evaluation):
| Model | VRAM | Speed | WER (Error Rate) |
|---|---|---|---|
| whisper-tiny | ~1 GB | 32x real-time | 18.4% |
| whisper-base | ~1 GB | 16x real-time | 12.1% |
| whisper-small.pt ★ | ~2 GB | 6x real-time | 6.3% |
| whisper-medium | ~5 GB | 2x real-time | 4.2% |
| whisper-large-v3 | ~10 GB | 1x real-time | 2.8% |