Bug fixes for exceptions observed during usage. Add further plumbing …

…for upcoming features. (#41)

AudioRecorder.py

@@ -55,7 +55,7 @@ def __init__(self):
  print("[ERROR] No loopback device found.")
 
  source = sr.Microphone(speaker=True,
- device_index= default_speakers["index"],
+ device_index=default_speakers["index"],
  sample_rate=int(default_speakers["defaultSampleRate"]),
  chunk_size=pyaudio.get_sample_size(pyaudio.paInt16),
  channels=default_speakers["maxInputChannels"])

GPTResponder.py

@@ -1,6 +1,7 @@
 import openai
+import datetime
 import GlobalVars
-from prompts import create_prompt, INITIAL_RESPONSE
+import prompts
 import time
 import conversation
 import constants
@@ -12,7 +13,7 @@
 
 class GPTResponder:
  def __init__(self, convo: conversation.Conversation):
- self.response = INITIAL_RESPONSE
+ self.response = prompts.INITIAL_RESPONSE
  self.response_interval = 2
  self.gl_vars = GlobalVars.TranscriptionGlobals()
  openai.api_key = self.gl_vars.api_key
@@ -22,18 +23,25 @@ def __init__(self, convo: conversation.Conversation):
 
  def generate_response_from_transcript_no_check(self, transcript):
  try:
- prompt_content = create_prompt(transcript)
- response = openai.ChatCompletion.create(
+ # prompt_content = create_prompt(transcript)
+ # prompt_api_message = [{"role": "system", "content": prompt_content}]
+ prompt_api_message = prompts.create_single_turn_prompt_message(transcript)
+ multiturn_prompt_content = self.conversation.get_merged_conversation(length=MAX_PHRASES)
+ multiturn_prompt_api_message = prompts.create_multiturn_prompt(multiturn_prompt_content)
+ # print(f'Usual prompt api message: {prompt_api_message}')
+ # print(f'Multiturn prompt: {multiturn_prompt_api_message}')
+ usual_response = openai.ChatCompletion.create(
  model=self.model,
- messages=[{"role": "system", "content": prompt_content}],
+ messages=prompt_api_message,
  temperature=0.0
  )
+
  except Exception as exception:
  print(exception)
  return ''
- full_response = response.choices[0].message.content
+ usual_full_response = usual_response.choices[0].message.content
  try:
- return full_response.split('[')[1].split(']')[0]
+ return usual_full_response.split('[')[1].split(']')[0]
  except:
  return ''
 
@@ -44,7 +52,7 @@ def generate_response_from_transcript(self, transcript):
  if self.gl_vars.freeze_state[0]:
  return ''
 
- return generate_response_from_transcript_no_check(self, transcript)
+ return self.generate_response_from_transcript_no_check(transcript)
 
  def respond_to_transcriber(self, transcriber):
  while True:
@@ -63,7 +71,7 @@ def respond_to_transcriber(self, transcriber):
  self.response = response
  self.conversation.update_conversation(persona=constants.PERSONA_ASSISTANT,
  text=response,
- time_spoken=end_time)
+ time_spoken=datetime.datetime.now())
 
  remaining_time = self.response_interval - execution_time
  if remaining_time > 0:

TranscriberModels.py

@@ -10,7 +10,7 @@ def get_model(use_api: bool, model: str = None):
  return APIWhisperTranscriber()
 
  model_cleaned = model if model else 'tiny'
- print(f'Using model: {model_cleaned}')
+ print(f'[INFO] Using local model: {model_cleaned}')
  return WhisperTranscriber(model=model_cleaned)
 
 

conversation.py

@@ -1,6 +1,14 @@
 from heapq import merge
 import constants
 import configuration
+import datetime
+
+DEFAULT_PREAMBLE = """You are a casual pal, genuinely interested in the conversation at hand.""" \
+ """Please respond, in detail, to the conversation. Confidently give a """\
+ """straightforward response to the speaker, even if you don't understand """\
+ """them. Give your response in square brackets. DO NOT ask to repeat, """\
+ """and DO NOT ask for clarification. Just answer the speaker directly."""\
+ """A poor transcription of conversation is given below."""
 
 
 class Conversation:
@@ -13,6 +21,8 @@ def __init__(self):
  constants.PERSONA_YOU: [],
  constants.PERSONA_SPEAKER: [],
  constants.PERSONA_ASSISTANT: []}
+ transcript = self.transcript_data[constants.PERSONA_SYSTEM]
+ transcript.append((f"{constants.PERSONA_SYSTEM}: [{DEFAULT_PREAMBLE}]\n\n", datetime.datetime.now()))
  config = configuration.Config().get_data()
 
  def clear_conversation_data(self):
@@ -58,3 +68,23 @@ def get_conversation(self,
  key=lambda x: x[1]))
  combined_transcript = combined_transcript[-length:]
  return "".join([t[0] for t in combined_transcript])
+
+ def get_merged_conversation(self, length: int = 0) -> list:
+ """Creates a prompt to be sent to LLM (OpenAI by default)
+ length: Get the last length elements from the audio transcript.
+ Default value = 0, gives the complete transcript
+ """
+ # print(f'You: Length: {len(self.transcript_data[constants.PERSONA_YOU])}')
+ # print(f'Speaker: Length: {len(self.transcript_data[constants.PERSONA_SPEAKER])}')
+ # print(f'Assistant: Length: {len(self.transcript_data[constants.PERSONA_ASSISTANT])}')
+ # print(f'System: Length: {len(self.transcript_data[constants.PERSONA_SYSTEM])}')
+
+ combined_transcript = list(merge(
+ self.transcript_data[constants.PERSONA_YOU][-length:],
+ self.transcript_data[constants.PERSONA_SPEAKER][-length:],
+ self.transcript_data[constants.PERSONA_ASSISTANT][-length:],
+ key=lambda x: x[1]))
+ combined_transcript = combined_transcript[-length:]
+
+ combined_transcript.insert(0, (f"{constants.PERSONA_SYSTEM}: [{self.transcript_data[constants.PERSONA_SYSTEM][0]}]\n\n", datetime.datetime.now()))
+ return combined_transcript

main.py

@@ -62,7 +62,7 @@ def main():
  if response.status_code != 200:
  print(f'Error received: {response}')
  except ConnectionError:
- print('Operating as a standalone client')
+ print('[INFO] Operating in Desktop mode')
 
  config = configuration.Config().get_data()
 

prompts.py

@@ -8,3 +8,18 @@ def create_prompt(transcript):
  \
 {transcript}.\
 {EPILOGUE}'
+
+
+def create_single_turn_prompt_message(transcript: str):
+ message = f'{PREAMBLE} \
+ \
+{transcript}.\
+{EPILOGUE}'
+
+ prompt_api_message = [{"role": "system", "content": message}]
+ return prompt_api_message
+
+
+def create_multiturn_prompt(convo: list):
+ # print(convo)
+ return ''