my_voice_assistant_v3/scripts/voice_loop.py

#!/usr/bin/env python3
"""Sprech-Loop: sprechen -> Antwort hoeren -> erneut sprechen.

Nimmt vom Mikrofon auf (Push-to-Talk), schickt das Audio ueber EINE
/ws/voice-Verbindung an das Gateway und spielt die Antwort ab. Das Gespraechs-
gedaechtnis bleibt ueber die `session_id` erhalten.

Beispiele:
    python scripts/voice_loop.py
    python scripts/voice_loop.py --url ws://localhost:8003/ws/voice --session oma-anna
    python scripts/voice_loop.py --device hw:1,0           # bestimmtes Mikrofon (arecord -L)
    python scripts/voice_loop.py --llm-provider openrouter --tts-provider openrouter
    python scripts/voice_loop.py --file frage.wav          # ohne Mikrofon (Test)

Voraussetzungen: laufendes Gateway, `arecord` (Aufnahme), ein Player
(`ffplay`/`aplay`/`paplay`), Python-Paket `websockets`.
"""

from __future__ import annotations

import argparse
import asyncio
import io
import json
import shutil
import signal
import subprocess
import sys
import tempfile
import wave

try:
    import websockets
except ModuleNotFoundError:
    sys.exit("Fehlt: Python-Paket 'websockets' (kommt mit uvicorn[standard]).")

TTS_SAMPLE_RATE = 24000  # Antwort-Audio des Gateways (s16le, mono)


def _require(tool: str) -> str:
    path = shutil.which(tool)
    if not path:
        sys.exit(f"Fehlt: '{tool}' nicht gefunden. Bitte installieren.")
    return path


def _first_player() -> list[str] | None:
    if shutil.which("ffplay"):
        return ["ffplay", "-loglevel", "quiet", "-nodisp", "-autoexit"]
    if shutil.which("aplay"):
        return ["aplay", "-q"]
    if shutil.which("paplay"):
        return ["paplay"]
    return None


def resolve_recorder(choice: str) -> str:
    """Waehlt das Aufnahmewerkzeug. 'auto' bevorzugt PipeWire (pw-record)."""
    if choice != "auto":
        if not shutil.which(choice):
            sys.exit(f"Fehlt: Aufnahmewerkzeug '{choice}' nicht gefunden.")
        return choice
    for tool in ("pw-record", "parecord", "arecord"):
        if shutil.which(tool):
            return tool
    sys.exit("Kein Aufnahmewerkzeug gefunden (pw-record / parecord / arecord).")


def _record_cmd(recorder: str, device: str | None, rate: int, outfile: str) -> list[str]:
    if recorder == "arecord":
        cmd = ["arecord", "-q", "-f", "S16_LE", "-r", str(rate), "-c", "1"]
        if device:
            cmd += ["-D", device]
        return cmd + [outfile]
    if recorder == "pw-record":
        cmd = ["pw-record", "--rate", str(rate), "--channels", "1", "--format", "s16"]
        if device:
            cmd += ["--target", device]
        return cmd + [outfile]
    if recorder == "parecord":
        cmd = ["parecord", f"--rate={rate}", "--channels=1", "--format=s16le",
               "--file-format=wav"]
        if device:
            cmd += [f"--device={device}"]
        return cmd + [outfile]
    sys.exit(f"Unbekanntes Aufnahmewerkzeug: {recorder}")


def record_utterance(recorder: str, device: str | None, rate: int) -> bytes:
    """Push-to-Talk: Enter startet, Enter stoppt die Aufnahme; gibt WAV-Bytes zurueck."""
    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    tmp.close()
    cmd = _record_cmd(recorder, device, rate, tmp.name)

    input("\n[Enter] = Aufnahme START …")
    proc = subprocess.Popen(cmd)
    input("[Enter] = Aufnahme STOP …")
    proc.send_signal(signal.SIGINT)
    try:
        proc.wait(timeout=5)
    except subprocess.TimeoutExpired:
        proc.kill()
    with open(tmp.name, "rb") as fh:
        return fh.read()


def play_pcm(pcm: bytes) -> None:
    player = _first_player()
    if not player:
        print("(kein Player gefunden – Antwort-Audio wird nicht abgespielt)")
        return
    buf = io.BytesIO()
    with wave.open(buf, "wb") as w:
        w.setnchannels(1)
        w.setsampwidth(2)
        w.setframerate(TTS_SAMPLE_RATE)
        w.writeframes(pcm)
    if player[0] == "ffplay":
        # ffplay liest WAV von stdin via pipe:0
        subprocess.run([*player, "-i", "pipe:0"], input=buf.getvalue())
    else:
        subprocess.run(player + ["/dev/stdin"], input=buf.getvalue())


def _start_frame(args) -> dict:
    frame = {"type": "start", "format": "wav"}
    for key in ("stt_provider", "llm_provider", "tts_provider", "language"):
        value = getattr(args, key, None)
        if value:
            frame[key] = value
    return frame


async def one_turn(ws, wav: bytes, start_frame: dict) -> None:
    await ws.send(json.dumps(start_frame))
    await ws.send(wav)
    await ws.send(json.dumps({"type": "end"}))

    audio = b""
    while True:
        msg = await ws.recv()
        if isinstance(msg, (bytes, bytearray)):
            audio = bytes(msg)
            continue
        event = json.loads(msg)
        etype = event.get("type")
        if etype == "transcript":
            print(f"  Du:         {event.get('text','')!r}")
        elif etype == "semantic":
            print(f"  Assistent:  {event.get('text','')!r}")
        elif etype == "emergency":
            print(f"  ⚠ NOTFALL erkannt (Kategorie: {event.get('category')})")
        elif etype == "error":
            print(f"  Fehler {event.get('status','')}: {event.get('detail')}")
            return
        elif etype == "done":
            break
    if audio:
        play_pcm(audio)


async def run(args) -> None:
    url = f"{args.url}?session_id={args.session}"
    if args.token:
        url += f"&token={args.token}"
    start_frame = _start_frame(args)

    async with websockets.connect(url, max_size=None) as ws:
        print(f"Verbunden: {args.url} (Session '{args.session}')")
        if args.file:
            with open(args.file, "rb") as fh:
                wav = fh.read()
            print(f"Sende Datei: {args.file}")
            await one_turn(ws, wav, start_frame)
            return
        recorder = resolve_recorder(args.recorder)
        print(f"Aufnahme mit: {recorder}"
              + (f" (Gerät: {args.device})" if args.device else " (Standardgerät)"))
        print("Sprich nach 'START', stoppe mit Enter. Strg+C beendet den Loop.")
        while True:
            try:
                wav = record_utterance(recorder, args.device, args.rate)
            except KeyboardInterrupt:
                print("\nEnde.")
                return
            if len(wav) < 1000:
                print("  (zu kurz/leer – nochmal)")
                continue
            await one_turn(ws, wav, start_frame)


def main() -> None:
    p = argparse.ArgumentParser(description="Sprech-Loop fuer das Voice-Assistant-Gateway")
    p.add_argument("--url", default="ws://127.0.0.1:8003/ws/voice")
    p.add_argument("--session", default="voice-loop")
    p.add_argument("--token", default=None, help="Bearer-Token, falls AUTH_ENABLED=true")
    p.add_argument("--device", default=None,
                   help="Aufnahmegeraet (arecord: -L; pw-record: --target; parecord: --device)")
    p.add_argument("--recorder", default="auto",
                   choices=["auto", "pw-record", "parecord", "arecord"],
                   help="Aufnahmewerkzeug; 'auto' bevorzugt PipeWire (pw-record)")
    p.add_argument("--rate", type=int, default=16000)
    p.add_argument("--stt-provider", dest="stt_provider", default=None)
    p.add_argument("--llm-provider", dest="llm_provider", default=None)
    p.add_argument("--tts-provider", dest="tts_provider", default=None)
    p.add_argument("--language", default=None)
    p.add_argument("--file", default=None, help="WAV statt Mikrofon senden (Test ohne Aufnahme)")
    args = p.parse_args()
    try:
        asyncio.run(run(args))
    except KeyboardInterrupt:
        print("\nEnde.")


if __name__ == "__main__":
    main()