Marcus/API/audio_api.py
2026-04-12 18:50:22 +04:00

420 lines
14 KiB
Python

#!/usr/bin/env python3
"""
API/audio_api.py — Marcus Audio API Layer
==========================================
Provides speak() and record_audio() for the Brain layer.
Brain imports ONLY from this API — never from unitree SDK directly.
Speaker: _CallRequestWithParamAndBin (single call, full buffer)
Mic: parec -d 3 (Hollyland wireless, PulseAudio source index from config)
TTS EN: Unitree built-in TtsMaker
TTS AR: Piper ar_JO-kareem-medium → resample → G1 speaker
Usage:
from API.audio_api import AudioAPI
audio = AudioAPI()
audio.speak("Hello", "en")
audio.speak("مرحبا", "ar")
recording = audio.record(seconds=5)
audio.play_pcm(recording)
"""
import json
import logging
import os
import subprocess
import threading
import time
import wave
import numpy as np
# ─── PATH CONFIG ─────────────────────────────────────────
from dotenv import load_dotenv
load_dotenv()
BASE_PATH = os.environ.get("PROJECT_BASE", "/home/unitree")
PROJECT_NAME = "Marcus"
PROJECT_ROOT = os.path.join(BASE_PATH, PROJECT_NAME)
LOG_DIR = os.path.join(PROJECT_ROOT, "logs")
os.makedirs(LOG_DIR, exist_ok=True)
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s [%(name)s] %(levelname)s: %(message)s",
handlers=[
logging.FileHandler(os.path.join(LOG_DIR, "voice.log")),
logging.StreamHandler(),
],
)
log = logging.getLogger("audio_api")
def load_config(name: str) -> dict:
path = os.path.join(PROJECT_ROOT, "Config", f"config_{name}.json")
with open(path, "r") as f:
return json.load(f)
# ─── AUDIO API CLASS ─────────────────────────────────────
class AudioAPI:
"""Marcus audio interface — speak + record + play."""
def __init__(self):
self._config = load_config("Voice")
self._client = None
self._sdk_available = False
self._init_sdk()
# Config shortcuts
self._tts = self._config["tts"]
self._mic = self._config["mic"]
self._spk = self._config["speaker"]
self._target_rate = self._tts["target_sample_rate"]
# Data dir
data_dir = os.path.join(PROJECT_ROOT, self._config["audio"]["data_dir"])
os.makedirs(data_dir, exist_ok=True)
self._data_dir = data_dir
# Speaking lock — prevents mic from hearing TTS output
self._speaking = False
self._speak_lock = threading.Lock()
log.info(self._config["messages"]["ready"])
def _init_sdk(self):
"""Initialize Unitree AudioClient."""
try:
from unitree_sdk2py.core.channel import ChannelFactoryInitialize
from unitree_sdk2py.g1.audio.g1_audio_client import AudioClient
dds_iface = self._config["speaker"]["dds_interface"]
ChannelFactoryInitialize(0, dds_iface)
self._client = AudioClient()
self._client.SetTimeout(10.0)
self._client.Init()
self._client.SetVolume(self._config["speaker"]["volume"])
self._sdk_available = True
log.info("AudioClient initialized on %s", dds_iface)
except Exception as e:
log.error("AudioClient init failed: %s", e)
self._sdk_available = False
# ─── SPEAK ────────────────────────────────────────────
def speak(self, text: str, lang: str = "auto"):
"""
Speak text in the given language.
Mutes mic during playback to prevent self-listening.
lang="en" → built-in TtsMaker
lang="ar" → Piper → resample → G1 speaker
lang="auto" → detect from text
"""
if lang == "auto":
lang = self._detect_lang(text)
log.info("[%s] speak: %s", lang.upper(), text[:80])
with self._speak_lock:
self._speaking = True
self._mute_mic()
try:
if lang == "en":
self._speak_english(text)
elif lang == "ar":
self._speak_arabic(text)
else:
log.warning("Unknown lang '%s', falling back to English", lang)
self._speak_english(text)
except Exception as e:
log.error("%s: %s", self._config["messages"]["error_tts"], e)
finally:
# Small delay so speaker fully stops before mic reopens
time.sleep(0.3)
self._unmute_mic()
self._speaking = False
def _mute_mic(self):
"""Mute the wireless mic to prevent self-listening."""
source = self._mic["source_index"]
subprocess.run(
["pactl", "set-source-mute", source, "1"],
capture_output=True,
)
log.debug("Mic muted")
def _unmute_mic(self):
"""Unmute the wireless mic."""
source = self._mic["source_index"]
subprocess.run(
["pactl", "set-source-mute", source, "0"],
capture_output=True,
)
log.debug("Mic unmuted")
@property
def is_speaking(self) -> bool:
"""True while TTS is playing — voice module checks this."""
return self._speaking
def _speak_english(self, text: str):
"""English TTS via edge-tts."""
self._speak_edge_tts(text, "en")
def _speak_arabic(self, text: str):
"""Arabic TTS via edge-tts."""
self._speak_edge_tts(text, "ar")
def speak_piper_en(self, text: str):
"""Alternative: English via Piper instead of built-in."""
voice = self._tts["piper_voice_en"]
audio, rate = self._piper_synthesize(text, voice)
audio_16k = self._resample(audio, rate)
self._play_pcm(audio_16k)
# ─── PIPER TTS ────────────────────────────────────────
def _piper_synthesize(self, text: str, voice: str) -> tuple:
"""Run Piper CLI, return (audio_int16, sample_rate)."""
cmd = ["piper", "--model", voice, "--output_raw"]
timeout = self._tts["piper_timeout_sec"]
proc = subprocess.run(
cmd,
input=text.encode("utf-8"),
capture_output=True,
timeout=timeout,
)
if proc.returncode != 0:
stderr = proc.stderr.decode()[:300]
raise RuntimeError(f"Piper failed: {stderr}")
audio = np.frombuffer(proc.stdout, dtype=np.int16)
piper_rate = self._tts["piper_sample_rate"]
log.info("Piper: %d samples @ %dHz (%.1fs)", len(audio), piper_rate, len(audio) / piper_rate)
return audio, piper_rate
# ─── RESAMPLE ─────────────────────────────────────────
def _speak_edge_tts(self, text: str, lang: str):
"""Generate speech via edge-tts and play on G1."""
import os as _os
voice = "ar-AE-HamdanNeural" if lang == "ar" else "en-US-GuyNeural"
ts = int(time.time() * 1000)
mp3_path = f"/tmp/edge_{lang}_{ts}.mp3"
wav_path = f"/tmp/edge_{lang}_{ts}.wav"
safe_text = text.replace('"', '\\"')
code = f'import edge_tts, asyncio; asyncio.run(edge_tts.Communicate(\"{safe_text}\", voice=\"{voice}\").save(\"{mp3_path}\"))'
result = subprocess.run(["python3", "-c", code], capture_output=True, text=True, timeout=30)
if result.returncode != 0:
log.error("edge-tts failed: %s", result.stderr[:200])
if lang == "en" and self._sdk_available:
self._client.TtsMaker(text, self._tts.get("builtin_speaker_id", 1))
time.sleep(max(2.0, len(text) * 0.06))
return
try:
from pydub import AudioSegment
a = AudioSegment.from_mp3(mp3_path)
a = a.set_frame_rate(16000).set_channels(1).set_sample_width(2)
a.export(wav_path, format="wav")
import wave
with wave.open(wav_path, "rb") as wf:
audio = np.frombuffer(wf.readframes(wf.getnframes()), dtype=np.int16)
_os.unlink(mp3_path)
_os.unlink(wav_path)
self._play_pcm(audio)
except Exception as e:
log.error("edge-tts conversion error: %s", e)
try: _os.unlink(mp3_path)
except: pass
try: _os.unlink(wav_path)
except: pass
def _resample(self, audio: np.ndarray, src_rate: int) -> np.ndarray:
"""Resample to target rate (16kHz)."""
if src_rate == self._target_rate:
return audio
tl = int(len(audio) * self._target_rate / src_rate)
return np.interp(
np.linspace(0, len(audio), tl, endpoint=False),
np.arange(len(audio)),
audio.astype(np.float64),
).astype(np.int16)
# ─── G1 SPEAKER PLAYBACK ─────────────────────────────
def _play_pcm(self, audio_16k: np.ndarray) -> float:
"""Play 16kHz mono int16 on G1 speaker. Returns duration."""
if not self._sdk_available:
log.warning("SDK not available, cannot play audio")
return 0.0
from unitree_sdk2py.g1.audio.g1_audio_api import (
ROBOT_API_ID_AUDIO_START_PLAY,
ROBOT_API_ID_AUDIO_STOP_PLAY,
)
app_name = self._spk["app_name"]
# Stop previous stream
self._client._Call(
ROBOT_API_ID_AUDIO_STOP_PLAY,
json.dumps({"app_name": app_name}),
)
time.sleep(0.3)
# Build params — unique stream_id every call
pcm = audio_16k.tobytes()
sid = f"s_{int(time.time() * 1000)}"
param = json.dumps({
"app_name": app_name,
"stream_id": sid,
"sample_rate": self._target_rate,
"channels": 1,
"bits_per_sample": 16,
})
# Single call — full buffer
self._client._CallRequestWithParamAndBin(
ROBOT_API_ID_AUDIO_START_PLAY, param, list(pcm)
)
duration = len(audio_16k) / self._target_rate
time.sleep(duration + 0.5)
self._client._Call(
ROBOT_API_ID_AUDIO_STOP_PLAY,
json.dumps({"app_name": app_name}),
)
log.info("Played: %.1fs", duration)
return duration
def play_pcm(self, audio_16k: np.ndarray) -> float:
"""Public wrapper for playing PCM audio."""
return self._play_pcm(audio_16k)
# ─── MIC RECORDING ───────────────────────────────────
def record(self, seconds: float = 5.0) -> np.ndarray:
"""Record from Hollyland wireless mic via parec. Returns int16 array."""
source = self._mic["source_index"]
rate = str(self._mic["rate"])
channels = str(self._mic["channels"])
fmt = self._mic["format"]
# Unmute mic
subprocess.run(
["pactl", "set-source-mute", source, "0"],
capture_output=True,
)
subprocess.run(
["pactl", "set-source-volume", source, "100%"],
capture_output=True,
)
log.info("Recording %.1fs from mic source %s", seconds, source)
proc = subprocess.Popen(
["parec", "-d", source,
f"--format={fmt}", f"--rate={rate}", f"--channels={channels}", "--raw"],
stdout=subprocess.PIPE,
)
time.sleep(seconds)
proc.terminate()
raw = proc.stdout.read()
audio = np.frombuffer(raw, dtype=np.int16)
log.info("Recorded: %d samples, std=%.0f", len(audio), audio.std())
if audio.std() < 50:
log.warning(self._config["messages"]["error_mic"] + " — mic may be silent")
return audio
def save_recording(self, audio: np.ndarray, name: str) -> str:
"""Save recording to Data/Voice/Recordings/."""
path = os.path.join(self._data_dir, f"{name}.wav")
wf = wave.open(path, "wb")
wf.setnchannels(1)
wf.setsampwidth(2)
wf.setframerate(self._target_rate)
wf.writeframes(audio.tobytes())
wf.close()
log.info("Saved: %s", path)
return path
# ─── LANGUAGE DETECTION ───────────────────────────────
@staticmethod
def _detect_lang(text: str) -> str:
"""Detect language from text — Arabic Unicode range check."""
for c in text:
if '\u0600' <= c <= '\u06FF':
return "ar"
return "en"
# ─── STATUS ───────────────────────────────────────────
@property
def is_available(self) -> bool:
return self._sdk_available
# ─── STANDALONE TEST ─────────────────────────────────────
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description="Marcus Audio API Test")
parser.add_argument("--test", action="store_true", help="Run speak tests")
parser.add_argument("--speak", type=str, help="Speak this text")
parser.add_argument("--lang", default="auto", help="Language: en, ar, auto")
parser.add_argument("--record", type=float, default=0, help="Record N seconds")
args = parser.parse_args()
api = AudioAPI()
if args.test:
print("\n--- English built-in ---")
api.speak("Hello, I am Marcus.", "en")
time.sleep(1)
print("\n--- Arabic Piper ---")
api.speak("مرحبا، أنا ماركوس", "ar")
time.sleep(1)
print("\n--- Auto-detect ---")
api.speak("How are you?")
time.sleep(1)
api.speak("كيف حالك؟")
time.sleep(1)
print("\n--- Record 3s + playback ---")
rec = api.record(3.0)
if rec.std() > 50:
api.play_pcm(rec)
print("\nDone.")
elif args.speak:
api.speak(args.speak, args.lang)
elif args.record > 0:
rec = api.record(args.record)
api.save_recording(rec, f"test_{int(time.time())}")
if rec.std() > 50:
api.play_pcm(rec)
else:
parser.print_help()