Result
Nothing generated yet. Pick a preset, type a line, hit Generate.
What to expect on this box
- Fast path (CUDA graphs): TTFA ≈ 0.23 s, generation ≈ 1.66× realtime — a 20 s clip takes ~33 s.
- 24 kHz mono output, ≤ 120 s of audio per request (longer text is cut at the token cap).
- One GPU slot: extra requests queue here, raw
/v1/*callers get409. - Elderly / child voices speak slower — same text, ~1.7× the duration.
History
loading…