Turn-taking · end-of-turn detection
01 | 94.0% | 89.0% | 1.0%One false cutoff in 100 incomplete clips. | 32msRuns in parallel with STT — no transcript wait.49 p90 |
02 | 87.0% | 86.0% | 12.0% | 29msWaits for the STT transcript (~0–125ms) before running.57 p90 |
03 | 86.0% | 88.0% | 16.0% | 127msPads every input to 256 tokens regardless of length — slowest as-shipped.193 p90 |
04 | 82.0% | 87.0% | 23.0% | 5ms24 p90 |
05 | 46.9% | 100% | 100%A silence timer ends every turn — the floor semantic detection has to beat. | ~1ms |
| Model | Accuracy synthetic | False-cutoff synthetic |
|---|---|---|
| Turnsensetext | 98.8% | 0.0% |
| LiveKit Intltext | 90.6% | 6.2% |
| LiveKit ENtext | 79.2% | 26.2% |
| Smart Turnaudio | 62.4% | 57.7% |
| Acoustic cue | Complete END | Incomplete WAIT | What it means |
|---|---|---|---|
| Final F0 slope | -119.7 Hz/s | +42.9 Hz/s | complete falls, incomplete rises/holds |
| Final energy slope | -14.9 dB/s | -2.5 dB/s | complete tapers off |