Audio stream-in simulation
Pretend the chosen audio file is arriving incrementally (TTS / mic),
even though it's actually a complete file on disk. full = one
POST. chunked = slice into pieces, sleep
push_delay between. tts_stream = same as chunked but
delay = chunk_sec / tts_speed. Only the last chunk carries
final=true.
Defaults model product TTS: hold=3s chunk=0.5s tts_speed=2×.
Reaction summary:
• chunk_sec ↓ → more POSTs, finer dribble (TTS-like).
↑ → bursty, closer to full.
• delay_sec = 0 → instant; = chunk_sec → realtime arrival
(stresses lookahead); > chunk_sec → engine starves.
• tts_speed = 1 → realtime (most stressful); = 2 → expected
HTTP PCM TTS; ≥ 10 → ≈ full; < 1 → starves.
Full reaction table + recipes:
docs/avatar-media-pipeline-v2.md#audio-stream-in-simulation