aharrison-fullerandClaude Opus 5
4286e88344
Loads parakeet-mlx in-process (no HTTP hop) and serves it over the Wyoming
protocol. On an M4 Mac mini this transcribes typical voice commands in ~110ms
versus ~1150ms for a whisper.cpp large-v3 setup, with identical accuracy on a
ten-command benchmark.
Two behaviours matter beyond speed: silence returns an empty string rather
than whisper's "Thank you." hallucination, and there is no decoder context
carried between requests.
Notable implementation details, all covered by mutation-checked regression
tests:
- MLX streams are thread-local, so the model is loaded and evaluated on a
single dedicated worker thread. Splitting those raises
"There is no Stream(cpu, 1) in current thread".
- parakeet_mlx.load_audio() shells out to ffmpeg, which is unnecessary here
since Wyoming delivers 16kHz mono PCM. The mel is built directly via
get_logmel(), whose input must be float32 -- it views the complex STFT
output as the input dtype, so anything narrower doubles the mel bin count.
- Wyoming's run loop has no except clause, so an exception escaping
handle_event closes the connection without sending a Transcript and Home
Assistant waits indefinitely. Failures are caught and returned as an empty
transcript instead.
Defaults to parakeet-tdt-0.6b-v2 rather than the newer multilingual v3
because v2 emits digits ("21 degrees") where v3 spells numbers out, and
Home Assistant's local intent matching expects digits.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
92 lines
3.0 KiB
Python
92 lines
3.0 KiB
Python
#!/usr/bin/env python3
|
|
"""Wyoming ASR server for NVIDIA Parakeet via parakeet-mlx (Apple Silicon)."""
|
|
import argparse
|
|
import asyncio
|
|
import logging
|
|
|
|
from wyoming.info import AsrModel, AsrProgram, Attribution, Info
|
|
from wyoming.server import AsyncServer
|
|
|
|
from .engine import ParakeetEngine
|
|
from .handler import ParakeetEventHandler
|
|
|
|
_LOGGER = logging.getLogger(__name__)
|
|
__version__ = "1.0.0"
|
|
|
|
# v2 is English-only but does better inverse text normalisation than the
|
|
# multilingual v3 ("21 degrees" / "30%" rather than "twenty-one degrees" /
|
|
# "thirty percent"), which is what Home Assistant's local intent matching
|
|
# expects. Don't switch to v3 without re-checking that.
|
|
DEFAULT_MODEL = "mlx-community/parakeet-tdt-0.6b-v2"
|
|
|
|
|
|
def build_parser() -> argparse.ArgumentParser:
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument("--uri", required=True, help="unix:// or tcp://")
|
|
parser.add_argument("--model", default=DEFAULT_MODEL, help="HuggingFace model id")
|
|
parser.add_argument("--language", default="en", help="Language code reported to HA")
|
|
parser.add_argument("--debug", action="store_true", help="Log DEBUG messages")
|
|
parser.add_argument("--log-format", default=logging.BASIC_FORMAT)
|
|
parser.add_argument("--version", action="version", version=__version__)
|
|
return parser
|
|
|
|
|
|
def build_info(model: str, language: str) -> Info:
|
|
"""Describe this service to Home Assistant's Wyoming config flow."""
|
|
return Info(
|
|
asr=[
|
|
AsrProgram(
|
|
name="parakeet-mlx",
|
|
description="NVIDIA Parakeet TDT via MLX",
|
|
attribution=Attribution(
|
|
name="senstella", url="https://github.com/senstella/parakeet-mlx"
|
|
),
|
|
installed=True,
|
|
version=__version__,
|
|
models=[
|
|
AsrModel(
|
|
name=model,
|
|
description=model,
|
|
attribution=Attribution(
|
|
name="NVIDIA",
|
|
url="https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2",
|
|
),
|
|
installed=True,
|
|
version=None,
|
|
languages=[language],
|
|
)
|
|
],
|
|
)
|
|
]
|
|
)
|
|
|
|
|
|
async def main() -> None:
|
|
args = build_parser().parse_args()
|
|
|
|
logging.basicConfig(
|
|
level=logging.DEBUG if args.debug else logging.INFO, format=args.log_format
|
|
)
|
|
|
|
_LOGGER.info("Loading %s", args.model)
|
|
engine = ParakeetEngine(args.model)
|
|
await engine.start()
|
|
|
|
wyoming_info = build_info(args.model, args.language)
|
|
server = AsyncServer.from_uri(args.uri)
|
|
_LOGGER.info("Ready on %s", args.uri)
|
|
await server.run(
|
|
lambda *a, **kw: ParakeetEventHandler(wyoming_info, args, engine, *a, **kw)
|
|
)
|
|
|
|
|
|
def run() -> None:
|
|
try:
|
|
asyncio.run(main())
|
|
except KeyboardInterrupt:
|
|
pass
|
|
|
|
|
|
if __name__ == "__main__":
|
|
run()
|