import os import sys import subprocess import threading from elevenlabs import VoiceSettings from .config import AUDIO_SINK, AGENTS, SILENCE_THRESHOLD, log ELEVENLABS_API_KEY = os.getenv("ELEVENLABS_API_KEY", "") ELEVENLABS_MODEL = os.getenv("ELEVENLABS_MODEL", "eleven_flash_v2_5") _elevenlabs_client = None _current_process: subprocess.Popen | None = None _process_lock = threading.Lock() def _get_elevenlabs(): global _elevenlabs_client if _elevenlabs_client is None: from elevenlabs.client import ElevenLabs _elevenlabs_client = ElevenLabs(api_key=ELEVENLABS_API_KEY) return _elevenlabs_client def stop_speaking(): """Прерывает текущее воспроизведение (barge-in)""" global _current_process with _process_lock: if _current_process and _current_process.poll() is None: _current_process.terminate() try: _current_process.wait(timeout=1) except subprocess.TimeoutExpired: _current_process.kill() _current_process = None def is_speaking() -> bool: with _process_lock: return _current_process is not None and _current_process.poll() is None _barge_in_flag = threading.Event() def start_barge_in_listener(): """Запускает фоновый поток VAD — если услышал голос во время TTS, ставит флаг barge-in.""" _barge_in_flag.clear() def _listen(): import pyaudio import numpy as np try: audio = pyaudio.PyAudio() stream = audio.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=1024) warmup = 8 # ~0.5s прогрев чтобы не словить эхо начала TTS i = 0 while is_speaking(): data = stream.read(1024, exception_on_overflow=False) i += 1 if i < warmup: continue amplitude = np.abs(np.frombuffer(data, dtype=np.int16)).mean() if amplitude > SILENCE_THRESHOLD * 1.5: # порог чуть выше чем для записи _barge_in_flag.set() stop_speaking() break stream.stop_stream() audio.terminate() except Exception: pass t = threading.Thread(target=_listen, daemon=True) t.start() return t def was_barge_in() -> bool: return _barge_in_flag.is_set() def _mpv_cmd() -> list[str]: """Команда mpv для воспроизведения из stdin""" mpv_bin = os.getenv("MPV_PATH", "mpv") cmd = [mpv_bin, "--no-video", "--really-quiet", "--no-terminal"] if AUDIO_SINK: cmd.append(f"--audio-device=pulse/{AUDIO_SINK}") cmd.append("-") return cmd def speak(text: str, agent_id: str = "cosmo"): try: _speak_elevenlabs(text, agent_id) except Exception as e: log.exception("TTS ошибка") print(f"⚠️ Ошибка воспроизведения: {e}") play_error_sound() def _speak_elevenlabs(text: str, agent_id: str): global _current_process client = _get_elevenlabs() voice_id = AGENTS.get(agent_id, AGENTS["cosmo"]).get("tts_voice", "") if not voice_id: log.error(f"tts_voice не задан для {agent_id}") print(f"⚠️ tts_voice не задан для {agent_id}") return voice_settings = VoiceSettings( stability=0.4, # ниже = живее интонация (для multilingual_v2) similarity_boost=0.8, style=0.1, # выше = эмоциональнее use_speaker_boost=True, speed=1.1 ) audio_stream = client.text_to_speech.convert( text=text, voice_id=voice_id, model_id=ELEVENLABS_MODEL, output_format="mp3_22050_32", voice_settings=voice_settings, optimize_streaming_latency=3 ) with _process_lock: _current_process = subprocess.Popen( _mpv_cmd(), stdin=subprocess.PIPE, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, ) proc = _current_process try: for chunk in audio_stream: if proc.poll() is not None: break try: proc.stdin.write(chunk) except BrokenPipeError: break proc.stdin.close() proc.wait() except Exception: proc.kill() finally: with _process_lock: if _current_process is proc: _current_process = None def _play_sound_file(filename: str, wait: bool = False): """Воспроизводит файл из папки sounds/ через mpv. wait=True — блокирует до конца воспроизведения.""" sounds_dir = os.path.join(os.path.dirname(__file__), "..", "sounds") path = os.path.normpath(os.path.join(sounds_dir, filename)) mpv_bin = os.getenv("MPV_PATH", "mpv") cmd = [mpv_bin, "--no-video", "--really-quiet", "--no-terminal", path] if wait: subprocess.run(cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) else: subprocess.Popen(cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) def play_activation_sound(): """Звук активации — неблокирующий""" try: _play_sound_file("Success_Cosmo.mp3", wait=False) except Exception as e: log.warning(f"Ошибка звука активации: {e}") def play_error_sound(): """Звук ошибки — 'не получилось'""" try: _play_sound_file("Error_Cosmo.mp3") except Exception as e: log.warning(f"Ошибка звука ошибки: {e}")