Spaces:
Sleeping
Sleeping
| from asyncio import Queue, TaskGroup | |
| import asyncio | |
| from contextlib import asynccontextmanager | |
| import ray | |
| from chat_service import ChatService | |
| # from local_speaker_service import LocalSpeakerService | |
| from text_to_speech_service import TextToSpeechService | |
| from response_state_manager import ResponseStateManager | |
| from ffmpeg_converter import FFMpegConverter | |
| from agent_response import AgentResponse | |
| import json | |
| class RespondToPromptAsync: | |
| def __init__( | |
| self, | |
| response_state_manager:ResponseStateManager, | |
| audio_output_queue): | |
| voice_id="2OviOUQc1JsQRQgNkVBj" | |
| self.llm_sentence_queue = Queue(maxsize=100) | |
| self.speech_chunk_queue = Queue(maxsize=100) | |
| self.voice_id = voice_id | |
| self.audio_output_queue = audio_output_queue | |
| self.response_state_manager = response_state_manager | |
| self.sentence_queues = [] | |
| self.sentence_tasks = [] | |
| # self.ffmpeg_converter = FFMpegConverter.remote(audio_output_queue) | |
| async def prompt_to_llm(self, prompt:str, messages:[str]): | |
| chat_service = ChatService() | |
| async with TaskGroup() as tg: | |
| agent_response = AgentResponse(prompt) | |
| async for text, is_complete_sentance in chat_service.get_responses_as_sentances_async(messages): | |
| if chat_service.ignore_sentence(text): | |
| is_complete_sentance = False | |
| if not is_complete_sentance: | |
| agent_response['llm_preview'] = text | |
| self.response_state_manager.set_llm_preview(text) | |
| continue | |
| agent_response['llm_preview'] = '' | |
| agent_response['llm_sentence'] = text | |
| agent_response['llm_sentences'].append(text) | |
| self.response_state_manager.add_llm_response_and_clear_llm_preview(text) | |
| print(f"{agent_response['llm_sentence']} id: {agent_response['llm_sentence_id']} from prompt: {agent_response['prompt']}") | |
| sentence_response = agent_response.make_copy() | |
| new_queue = Queue() | |
| self.sentence_queues.append(new_queue) | |
| task = tg.create_task(self.llm_sentence_to_speech(sentence_response, new_queue)) | |
| self.sentence_tasks.append(task) | |
| agent_response['llm_sentence_id'] += 1 | |
| async def llm_sentence_to_speech(self, sentence_response, output_queue): | |
| tts_service = TextToSpeechService(self.voice_id) | |
| chunk_count = 0 | |
| async for chunk_response in tts_service.get_speech_chunks_async(sentence_response): | |
| chunk_response = chunk_response.make_copy() | |
| # await self.output_queue.put_async(chunk_response) | |
| await output_queue.put(chunk_response) | |
| chunk_response = { | |
| 'prompt': sentence_response['prompt'], | |
| 'llm_sentence_id': sentence_response['llm_sentence_id'], | |
| 'chunk_count': chunk_count, | |
| } | |
| chunk_id_json = json.dumps(chunk_response) | |
| self.response_state_manager.add_tts_raw_chunk_id(chunk_id_json, sentence_response['llm_sentence_id']) | |
| chunk_count += 1 | |
| async def speech_to_converter(self): | |
| self.ffmpeg_converter = FFMpegConverter(self.audio_output_queue) | |
| await self.ffmpeg_converter.start_process() | |
| self.ffmpeg_converter_task = asyncio.create_task(self.ffmpeg_converter.run()) | |
| while True: | |
| for i, task in enumerate(self.sentence_tasks): | |
| # Skip this task/queue pair if task completed | |
| queue = self.sentence_queues[i] | |
| if task.done() and queue.empty(): | |
| continue | |
| while not queue.empty(): | |
| chunk_response = await queue.get() | |
| audio_chunk_ref = chunk_response['tts_raw_chunk_ref'] | |
| audio_chunk = ray.get(audio_chunk_ref) | |
| await self.ffmpeg_converter.push_chunk(audio_chunk) | |
| break | |
| await asyncio.sleep(0.01) | |
| async def run(self, prompt:str, messages:[str]): | |
| self.task_group_tasks = [] | |
| async with TaskGroup() as tg: # Use asyncio's built-in TaskGroup | |
| t1 = tg.create_task(self.prompt_to_llm(prompt, messages)) | |
| t2 = tg.create_task(self.speech_to_converter()) | |
| self.task_group_tasks.extend([t1, t2]) | |
| async def terminate(self): | |
| # Cancel tasks | |
| if self.task_group_tasks: | |
| for task in self.task_group_tasks: | |
| task.cancel() | |
| for task in self.sentence_tasks: | |
| task.cancel() | |
| # Close FFmpeg converter actor | |
| if self.ffmpeg_converter_task: | |
| self.ffmpeg_converter_task.cancel() | |
| await self.ffmpeg_converter.close() | |
| # ray.kill(self.ffmpeg_converter) | |
| # Flush all queues | |
| # TODO re-enable to interupt when user speaks | |
| # while not self.audio_output_queue.empty(): | |
| # await self.audio_output_queue.get_async() | |
| # # await self.audio_output_queue.get_async(block=False) | |
| while not self.llm_sentence_queue.empty(): | |
| self.llm_sentence_queue.get_nowait() | |
| while not self.speech_chunk_queue.empty(): | |
| self.speech_chunk_queue.get_nowait() | |
| for sentence_queue in self.sentence_queues: | |
| while not sentence_queue.empty(): | |
| sentence_queue.get_nowait() | |