| """Interruptible voice orchestration, with optional local ASR/TTS adapters. |
| |
| Microphone VAD/streaming ASR and production full-duplex echo cancellation are planned. |
| """ |
| import asyncio |
| import time |
| from .evaluation import word_error_rate |
|
|
|
|
| class VoiceSession: |
| def __init__(self, reply, speak, stop_audio): |
| self.reply, self.speak, self.stop_audio = reply, speak, stop_audio |
| self.generation = 0 |
| self.task = None |
| self.metrics = [] |
|
|
| async def interrupt(self): |
| self.generation += 1 |
| await self.stop_audio() |
| if self.task and not self.task.done(): |
| self.task.cancel() |
| try: |
| await self.task |
| except asyncio.CancelledError: |
| pass |
|
|
| async def transcript(self, text, *, speech_end=None): |
| await self.interrupt() |
| generation = self.generation |
| received = time.perf_counter() |
| async def turn(): |
| first = None |
| async for chunk in self.reply(text): |
| if generation != self.generation: |
| return |
| if first is None: |
| first = time.perf_counter() |
| audio_start = await self.speak(chunk) |
| self.metrics.append({"transcript_ms": None if speech_end is None else (received-speech_end)*1000, |
| "first_token_ms": (first-received)*1000, |
| "first_audio_ms": None if audio_start is None else (audio_start-received)*1000}) |
| self.task = asyncio.create_task(turn()) |
| return self.task |
|
|
|
|
| def transcribe_file(audio_path, model="small", language=None): |
| from faster_whisper import WhisperModel |
| asr = WhisperModel(model, device="cpu", compute_type="int8") |
| start = time.perf_counter() |
| segments, info = asr.transcribe(str(audio_path), language=language, vad_filter=True) |
| text = " ".join(s.text.strip() for s in segments) |
| return {"text": text, "language": info.language, "seconds": time.perf_counter()-start} |
|
|
|
|
| def speak_local(text): |
| import pyttsx3 |
| engine = pyttsx3.init() |
| engine.say(text) |
| engine.runAndWait() |
|
|
