Game translation and speech latency benchmarks
TTFA (time to first audio) measures the time from a complete captured frame to the first matching speech in Windows audio loopback.
| v1.5.0 input | Trials | Median TTFA |
|---|---|---|
| Short synthetic dialogue | 3 | 910.24 ms |
| Two synthetic sentences | 3 | 914.23 ms |
The test replayed frames at four frames per second with warm PaddleOCR on CPU, local Gemma 4 E4B Q4_K_M and VieNeu Nano. It used a separate WASAPI endpoint. These figures measure replay latency. Live game FPS, RDP and the time a game takes to reveal text are outside the measurement.
Measurement setup and individual trials · Historical benchmarks
