marauder-actual
102 lines
3.4 KiB
Python
102 lines
3.4 KiB
Python
"""Piper TTS adapter for chat.saiden.dev.
|
|
|
|
Synthesises text → WAV bytes by subprocess'ing the `piper` CLI binary
|
|
(already installed on every host that runs marauder-os).
|
|
|
|
Designed to fail silently — if piper is missing or synthesis errors,
|
|
the chat still works, just without voice.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import logging
|
|
import os
|
|
import shutil
|
|
import tempfile
|
|
from pathlib import Path
|
|
|
|
log = logging.getLogger("chat-saiden.tts")
|
|
|
|
# Where the voice .onnx files live across hosts.
|
|
# Order: env override → macOS marauder → linux marauder → linux marauder-agent (mesh node) → linux ~/.local
|
|
_VOICE_SEARCH_PATHS = [
|
|
Path.home() / "Library/Application Support/marauder/voices",
|
|
Path("/home") / os.environ.get("USER", "marauder") / ".local/share/marauder/voices",
|
|
Path.home() / ".local/share/marauder/voices",
|
|
Path.home() / ".local/share/psn/voices",
|
|
Path.home() / ".local/share/piper/voices",
|
|
]
|
|
|
|
|
|
def _resolve_voice_path(name: str) -> Path | None:
|
|
"""Return absolute path to a voice model by short name, or None."""
|
|
# explicit override
|
|
override = os.environ.get("TTS_VOICE_PATH")
|
|
if override:
|
|
p = Path(override)
|
|
return p if p.exists() else None
|
|
|
|
for base in _VOICE_SEARCH_PATHS:
|
|
candidate = base / f"{name}.onnx"
|
|
if candidate.exists():
|
|
return candidate
|
|
return None
|
|
|
|
|
|
PIPER_BIN = shutil.which("piper") or os.environ.get("PIPER_BIN")
|
|
|
|
|
|
class TTS:
|
|
"""Subprocess-based piper synthesizer with graceful fallback."""
|
|
|
|
def __init__(self, voice: str = "en_US-amy-medium") -> None:
|
|
self.voice = voice
|
|
self.voice_path = _resolve_voice_path(voice)
|
|
self.bin = PIPER_BIN
|
|
if not self.bin:
|
|
log.warning("piper binary not found on PATH — TTS disabled")
|
|
elif not self.voice_path:
|
|
log.warning("voice '%s' not found in known locations — TTS disabled", voice)
|
|
else:
|
|
log.info("TTS enabled — voice=%s path=%s bin=%s", voice, self.voice_path, self.bin)
|
|
|
|
@property
|
|
def available(self) -> bool:
|
|
return bool(self.bin and self.voice_path)
|
|
|
|
async def synthesize(self, text: str) -> bytes | None:
|
|
"""Return WAV bytes, or None on failure / unavailable."""
|
|
if not self.available:
|
|
return None
|
|
if not text.strip():
|
|
return None
|
|
|
|
# piper wants an output file path (no stdout streaming for WAV in older versions)
|
|
out = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
|
|
out.close()
|
|
out_path = out.name
|
|
try:
|
|
proc = await asyncio.create_subprocess_exec(
|
|
self.bin,
|
|
"--model", str(self.voice_path),
|
|
"--output_file", out_path,
|
|
stdin=asyncio.subprocess.PIPE,
|
|
stdout=asyncio.subprocess.DEVNULL,
|
|
stderr=asyncio.subprocess.PIPE,
|
|
)
|
|
_, stderr = await proc.communicate(text.encode("utf-8"))
|
|
if proc.returncode != 0:
|
|
log.error("piper exited %s: %s", proc.returncode, stderr.decode("utf-8", "replace")[:300])
|
|
return None
|
|
with open(out_path, "rb") as f:
|
|
return f.read()
|
|
except Exception:
|
|
log.exception("piper synthesis failed")
|
|
return None
|
|
finally:
|
|
try:
|
|
os.unlink(out_path)
|
|
except OSError:
|
|
pass
|