reduce complexity by removing AudioStreamProcessor from StreamingChatService

respond_to_prompt_actor.py

@@ -12,8 +12,7 @@ class PromptToLLMActor:
         load_dotenv()
         self.input_queue = input_queue
         self.output_queue = output_queue
-        self.audio_processor = AudioStreamProcessor()
-        self.chat_service = StreamingChatService(self.audio_processor, voice_id=voice_id)
+        self.chat_service = StreamingChatService(voice_id=voice_id)
         self.cancel_event = None
 
     async def run(self):
@@ -40,8 +39,7 @@ class LLMSentanceToSpeechActor:
         load_dotenv()
         self.input_queue = input_queue
         self.output_queue = output_queue
-        self.audio_processor = AudioStreamProcessor()
-        self.chat_service = StreamingChatService(self.audio_processor, voice_id=voice_id)
+        self.chat_service = StreamingChatService(voice_id=voice_id)
         self.cancel_event = None
 
     async def run(self):
@@ -66,13 +64,14 @@ class SpeechToSpeakerActor:
         load_dotenv()
         self.input_queue = input_queue
         self.audio_processor = AudioStreamProcessor()
-        self.chat_service = StreamingChatService(self.audio_processor, voice_id=voice_id)
+        self.chat_service = StreamingChatService(voice_id=voice_id)
 
     async def run(self):
         while True:
             audio_chunk = await self.input_queue.get_async()
             # print (f"Got audio chunk {len(audio_chunk)}")
             self.chat_service.enqueue_speech_bytes_to_play([audio_chunk])
+            self.audio_processor.add_audio_stream([audio_chunk])
             
     async def cancel(self):
         while not self.input_queue.empty():

streaming_chat_service.py

@@ -5,13 +5,11 @@ import os
 import torch
 import openai
 
-from audio_stream_processor import AudioStreamProcessor
 from speech_service import SpeechService
 
 
 class StreamingChatService:
-    def __init__(self, audio_processor:AudioStreamProcessor()=None, api="openai", model_id = "gpt-3.5-turbo", voice_id="Bella"):
-        self._audio_processor = audio_processor
+    def __init__(self, api="openai", model_id = "gpt-3.5-turbo", voice_id="Bella"):
         self._speech_service = SpeechService(voice_id=voice_id)
         self._api = api
         self._device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -106,43 +104,6 @@ I fell off the pink step, and I had an accident.
         if not has_letters and not has_numbers:
             return True
         return False
-    
-    def _safe_enqueue_text_to_speak(self, text_to_speak):
-        if self.ignore_sentence(text_to_speak):
-            return
-        stream = self._speech_service.stream(text_to_speak)
-        self._audio_processor.add_audio_stream(stream)        
-
-    def respond_to(self, prompt):
-        self._messages.append({"role": "user", "content": prompt})
-        agent_response = ""
-        current_sentence = ""
-
-        response = openai.ChatCompletion.create(
-                model=self._model_id,
-                messages=self._messages,
-                temperature=1.0, # use 1.0 for debugging/deteministic results
-                stream=True
-        )
-
-        for chunk in response:
-            chunk_message = chunk['choices'][0]['delta']
-            if 'content' in chunk_message:
-                chunk_text = chunk_message['content']
-                # print(chunk_text)
-                current_sentence += chunk_text
-                agent_response += chunk_text
-                text_to_speak = self._should_we_send_to_voice(current_sentence)
-                if text_to_speak:
-                    self._safe_enqueue_text_to_speak(text_to_speak)
-                    print(text_to_speak)
-                    current_sentence = current_sentence[len(text_to_speak):]
-
-        if len(current_sentence) > 0:
-            self._safe_enqueue_text_to_speak(current_sentence)
-            print(current_sentence)
-        self._messages.append({"role": "assistant", "content": agent_response})
-        return agent_response
 
     async def get_responses_as_sentances_async(self, prompt, cancel_event):
         self._messages.append({"role": "user", "content": prompt})
@@ -190,6 +151,3 @@ I fell off the pink step, and I had an accident.
             if cancel_event.is_set():
                 return
             yield chunk
-
-    def enqueue_speech_bytes_to_play(self, speech_bytes):
-        self._audio_processor.add_audio_stream(speech_bytes)