77d7cd8b56
The first CUDA inference pays a large lazy cost (kernel autotune/cudnn) — ~10s for a cold TTS synth — which would blow the voice loop's ~1s budget on the very first reply. Each worker now runs one dummy inference (TTS: a short phrase; STT: 1s of silence) after model load and before emitting "ready", so "ready" means "hot". Warmup failures are logged and never block startup. Verified: first real call after startup is now TTS ~238ms / STT ~189ms (was ~11s cold for TTS). 12 tests pass. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
110 lines
4.0 KiB
Python
110 lines
4.0 KiB
Python
"""Persistent MeloTTS worker (Korean).
|
|
|
|
MeloTTS lives in its own Python (melo311); loading the model takes seconds, so
|
|
we load it ONCE here and then serve synthesis requests over stdin/stdout. This
|
|
process is launched with the melo311 interpreter by wsai.backends.melo.MeloTTS.
|
|
|
|
MeloTTS (and its deps) print progress straight to stdout, which would corrupt
|
|
the JSON protocol. So on startup we split the streams: a private duplicate of
|
|
the original stdout carries the protocol, and fd 1 is redirected to fd 2 so all
|
|
library chatter lands on stderr instead.
|
|
|
|
Protocol (one JSON object per line, on the protocol channel):
|
|
<- {"text": "...", "out": "/abs/path.wav", "speed": 1.3}
|
|
-> {"ok": true, "out": "/abs/path.wav", "ms": 123}
|
|
-> {"ok": false, "error": "..."}
|
|
On startup, once the model is ready, it emits exactly one line:
|
|
-> {"ready": true, "ms": <load-ms>, "device": "cpu"}
|
|
"""
|
|
|
|
import json
|
|
import os
|
|
import sys
|
|
import time
|
|
|
|
# Split protocol from library noise BEFORE importing anything heavy.
|
|
_proto = os.fdopen(os.dup(1), "w", buffering=1) # private copy of real stdout
|
|
os.dup2(2, 1) # fd1 -> stderr, so stray library prints don't hit the protocol
|
|
|
|
|
|
def _emit(obj: dict) -> None:
|
|
_proto.write(json.dumps(obj) + "\n")
|
|
_proto.flush()
|
|
|
|
|
|
def _log(*a):
|
|
print(*a, file=sys.stderr, flush=True)
|
|
|
|
|
|
def main() -> None:
|
|
lang = "KR"
|
|
requested = os.environ.get("WSAI_MELO_DEVICE", "auto") # cpu | cuda | auto
|
|
from melo.api import TTS # heavy import; only in the melo venv
|
|
|
|
def _has_cuda() -> bool:
|
|
try:
|
|
import torch
|
|
|
|
return torch.cuda.is_available()
|
|
except Exception:
|
|
return False
|
|
|
|
device = requested
|
|
if requested == "auto":
|
|
device = "cuda" if _has_cuda() else "cpu"
|
|
|
|
t0 = time.monotonic()
|
|
try:
|
|
tts = TTS(language=lang, device=device)
|
|
except Exception as exc:
|
|
# CUDA picked but unusable (CPU-only torch, missing libs, OOM): fall back
|
|
# to CPU rather than leaving the whole voice loop dead.
|
|
if device == "cuda":
|
|
_log(f"[melo_worker] CUDA load failed ({exc}); falling back to CPU")
|
|
device = "cpu"
|
|
tts = TTS(language=lang, device=device)
|
|
else:
|
|
raise
|
|
speaker_id = tts.hps.data.spk2id[lang]
|
|
load_ms = int((time.monotonic() - t0) * 1000)
|
|
|
|
# Warm up before signalling ready: the first CUDA synth pays a large lazy
|
|
# cost (kernel autotune/cudnn), ~10s cold vs ~130ms hot, which would blow the
|
|
# voice loop's ~1s budget on the very first reply. Do that dummy synth here so
|
|
# "ready" means "hot". Failures must not block startup.
|
|
warmup_ms = None
|
|
try:
|
|
warm_out = os.path.expanduser("~/.cache/wsai/tts/_warmup.wav")
|
|
os.makedirs(os.path.dirname(warm_out), exist_ok=True)
|
|
w = time.monotonic()
|
|
tts.tts_to_file("워밍업", speaker_id, warm_out, speed=1.3)
|
|
warmup_ms = int((time.monotonic() - w) * 1000)
|
|
except Exception as exc:
|
|
_log(f"[melo_worker] warmup skipped: {exc}")
|
|
|
|
_emit({"ready": True, "ms": load_ms, "device": device, "warmup_ms": warmup_ms})
|
|
_log(f"[melo_worker] model ready in {load_ms} ms on {device} (warmup {warmup_ms} ms)")
|
|
|
|
for line in sys.stdin:
|
|
line = line.strip()
|
|
if not line:
|
|
continue
|
|
try:
|
|
req = json.loads(line)
|
|
text = req["text"]
|
|
out = req["out"]
|
|
speed = float(req.get("speed", 1.0))
|
|
if out.startswith("/tmp") or out.startswith("/dev/shm"):
|
|
raise ValueError(f"refusing RAM-backed tmpfs path: {out}")
|
|
s = time.monotonic()
|
|
tts.tts_to_file(text, speaker_id, out, speed=speed)
|
|
ms = int((time.monotonic() - s) * 1000)
|
|
_emit({"ok": True, "out": out, "ms": ms})
|
|
except Exception as exc: # keep the worker alive across bad requests
|
|
_emit({"ok": False, "error": f"{type(exc).__name__}: {exc}"})
|
|
_log(f"[melo_worker] error: {exc}")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|