Get Response from LLM on demand (#44)

- Add feature to get response from LLM on demand in addition to getting it continuously. 
This option is very useful when continuous response from LLM is disabled.
- Bug Fix: Pausing, restarting transcription was not working from File Menu.

GPTResponder.py

@@ -1,20 +1,20 @@
-import openai
 import datetime
+import time
 import GlobalVars
 import prompts
-import time
+import openai
 import conversation
 import constants
 import configuration
 import app_logging as al
 
 
-# Number of phrases to use for generating a response
-MAX_PHRASES = 20
 root_logger = al.get_logger()
 
 
 class GPTResponder:
+ """Handles all interactions with openAI LLM / ChatGPT
+ """
  def __init__(self, convo: conversation.Conversation):
  root_logger.info(GPTResponder.__name__)
  self.response = prompts.INITIAL_RESPONSE
@@ -25,12 +25,16 @@ def __init__(self, convo: conversation.Conversation):
  self.config = configuration.Config().get_data()
  self.model = self.config['OpenAI']['ai_model']
 
- def generate_response_from_transcript_no_check(self, transcript):
+ def generate_response_from_transcript_no_check(self, transcript) -> str:
+ """Ping LLM to get a suggested response right away.
+ Gets a response even if the continuous suggestion option is disabled.
+ """
  try:
  # 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_content = self.conversation.get_merged_conversation(
+ length=constants.MAX_TRANSCRIPTION_PHRASES_FOR_LLM)
  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}')
@@ -42,12 +46,17 @@ def generate_response_from_transcript_no_check(self, transcript):
 
  except Exception as exception:
  print(exception)
- return ''
+ root_logger.error('Error when attempting to get a response from LLM.')
+ root_logger.exception(exception)
+ return prompts.INITIAL_RESPONSE
+
  usual_full_response = usual_response.choices[0].message.content
  try:
  return usual_full_response.split('[')[1].split(']')[0]
- except:
- return ''
+ except Exception as exception:
+ root_logger.error('Error parsing response from LLM.')
+ root_logger.exception(exception)
+ return prompts.INITIAL_RESPONSE
 
  def generate_response_from_transcript(self, transcript):
  """Ping OpenAI LLM model to get response from the Assistant
@@ -59,13 +68,16 @@ def generate_response_from_transcript(self, transcript):
  return self.generate_response_from_transcript_no_check(transcript)
 
  def respond_to_transcriber(self, transcriber):
+ """Thread method to continously update the transcript
+ """
  while True:
 
  if transcriber.transcript_changed_event.is_set():
  start_time = time.time()
 
  transcriber.transcript_changed_event.clear()
- transcript_string = transcriber.get_transcript(length=MAX_PHRASES)
+ transcript_string = transcriber.get_transcript(
+ length=constants.MAX_TRANSCRIPTION_PHRASES_FOR_LLM)
  response = self.generate_response_from_transcript(transcript_string)
 
  end_time = time.time() # Measure end time
@@ -84,5 +96,7 @@ def respond_to_transcriber(self, transcriber):
  time.sleep(0.3)
 
  def update_response_interval(self, interval):
+ """Change the interval for pinging LLM
+ """
  root_logger.info(GPTResponder.update_response_interval.__name__)
  self.response_interval = interval

GlobalVars.py

@@ -1,13 +1,15 @@
 import queue
+import tkinter as tk
+import customtkinter as ctk
 from AudioTranscriber import AudioTranscriber
 import AudioRecorder
-import customtkinter as ctk
 import Singleton
 import app_logging as al
 
 
 root_logger = al.get_logger()
 
+
 class TranscriptionGlobals(Singleton.Singleton):
  """Global constants for audio processing. It is implemented as a Singleton class.
  """
@@ -23,6 +25,8 @@ class TranscriptionGlobals(Singleton.Singleton):
  freeze_state: list = None
  freeze_button: ctk.CTkButton = None
  api_key: str = None
+ filemenu: tk.Menu = None
+ response_textbox: ctk.CTkTextbox = None
 
  def __init__(self, key: str = 'API_KEY'):
  root_logger.info(TranscriptionGlobals.__name__)

constants.py

@@ -7,3 +7,4 @@
 PERSONA_SPEAKER = 'Speaker'
 
 LOG_NAME = 'Transcribe'
+MAX_TRANSCRIPTION_PHRASES_FOR_LLM = 20

main.py

@@ -2,12 +2,12 @@
 import argparse
 from argparse import RawTextHelpFormatter
 import time
-import requests
 import subprocess
+import requests
 from requests.exceptions import ConnectionError
+import customtkinter as ctk
 from AudioTranscriber import AudioTranscriber
 from GPTResponder import GPTResponder
-import customtkinter as ctk
 import TranscriberModels
 import interactions
 import ui
@@ -96,7 +96,11 @@ def main():
  update_interval_slider_label = ui_components[3]
  global_vars.freeze_button = ui_components[4]
  lang_combobox = ui_components[5]
+ filemenu = ui_components[6]
+ response_now_button = ui_components[7]
 
+ global_vars.filemenu = filemenu
+ global_vars.response_textbox = response_textbox
  global_vars.user_audio_recorder.record_into_queue(global_vars.audio_queue)
 
  time.sleep(2)
@@ -135,6 +139,7 @@ def main():
 
  ui_cb = ui.ui_callbacks()
  global_vars.freeze_button.configure(command=ui_cb.freeze_unfreeze)
+ response_now_button.configure(command=ui_cb.update_response_ui_now)
  label_text = f'Update Response interval: {update_interval_slider.get()} seconds'
  update_interval_slider_label.configure(text=label_text)
 

ui.py

@@ -8,6 +8,7 @@
 import GlobalVars
 import GPTResponder
 import app_logging as al
+import constants
 
 
 root_logger = al.get_logger()
@@ -40,15 +41,28 @@ def freeze_unfreeze(self):
  root_logger.info(ui_callbacks.freeze_unfreeze.__name__)
  self.global_vars.freeze_state[0] = not self.global_vars.freeze_state[0] # Invert the state
  self.global_vars.freeze_button.configure(
- text="Suggest Response" if self.global_vars.freeze_state[0] else "Do Not Suggest Response"
+ text="Suggest Responses Continuously" if self.global_vars.freeze_state[0] else "Do Not Suggest Responses Continuously"
  )
 
+ def update_response_ui_now(self):
+ """Get response from LLM right away"""
+ transcript_string = self.global_vars.transcriber.get_transcript(
+ length=constants.MAX_TRANSCRIPTION_PHRASES_FOR_LLM)
+
+ response_string = self.global_vars.responder.generate_response_from_transcript_no_check(transcript_string)
+ self.global_vars.response_textbox.configure(state="normal")
+ write_in_textbox(self.global_vars.response_textbox, response_string)
+ self.global_vars.response_textbox.configure(state="disabled")
+
  def set_transcript_state(self):
+ """Enables, disables transcription.
+ Text of menu item File -> Pause Transcription toggles accordingly"""
  root_logger.info(ui_callbacks.set_transcript_state.__name__)
  self.global_vars.transcriber.transcribe = not self.global_vars.transcriber.transcribe
- self.global_vars.transcript_button.configure(
- text="Pause Transcript" if self.global_vars.transcriber.transcribe else "Start Transcript"
- )
+ if self.global_vars.transcriber.transcribe:
+ self.global_vars.filemenu.entryconfigure(1, label="Pause Transcription")
+ else:
+ self.global_vars.filemenu.entryconfigure(1, label="Start Transcription")
 
 
 def write_in_textbox(textbox: ctk.CTkTextbox, text: str):
@@ -99,7 +113,7 @@ def update_response_ui(responder: GPTResponder,
  freeze_state)
 
 
-def clear_transcriber_context(transcriber: AudioTranscriber, 
+def clear_transcriber_context(transcriber: AudioTranscriber,
  audio_queue: queue.Queue):
  """Reset the transcriber
  Args:
@@ -151,6 +165,8 @@ def create_ui_components(root):
  # Add a "Copy To Clipboard" menu item to the file menu
  editmenu.add_command(label="Copy Transcript to Clipboard", command=ui_cb.copy_to_clipboard)
 
+ # See example of add_radiobutton() at https://www.plus2net.com/python/tkinter-menu.php
+ # Radiobutton would be a good way to display different languages
  # lang_menu = tk.Menu(menubar, tearoff=False)
  # for lang in LANGUAGES_DICT.values():
  # model.change_lang
@@ -172,9 +188,12 @@ def create_ui_components(root):
  response_textbox.grid(row=0, column=1, padx=10, pady=20, sticky="nsew")
  response_textbox.insert("0.0", prompts.INITIAL_RESPONSE)
 
- freeze_button = ctk.CTkButton(root, text="Suggest Response", command=None)
+ freeze_button = ctk.CTkButton(root, text="Suggest Responses Continuously", command=None)
  freeze_button.grid(row=1, column=1, padx=10, pady=3, sticky="nsew")
 
+ response_now_button = ctk.CTkButton(root, text="Suggest Responses Now", command=None)
+ response_now_button.grid(row=2, column=1, padx=10, pady=3, sticky="nsew")
+
  update_interval_slider_label = ctk.CTkLabel(root, text="", font=("Arial", 12),
  text_color="#FFFCF2")
  update_interval_slider_label.grid(row=1, column=0, padx=10, pady=3, sticky="nsew")
@@ -185,9 +204,10 @@ def create_ui_components(root):
  update_interval_slider.grid(row=2, column=0, padx=10, pady=10, sticky="nsew")
 
  lang_combobox = ctk.CTkOptionMenu(root, values=list(LANGUAGES_DICT.values()))
- lang_combobox.grid(row=2, column=1, padx=200, pady=10, sticky="nsew")
+ lang_combobox.grid(row=3, column=0, padx=200, pady=10, sticky="nsew")
 
  # Order of returned components is important.
  # Add new components to the end
  return [transcript_textbox, response_textbox, update_interval_slider,
- update_interval_slider_label, freeze_button, lang_combobox]
+ update_interval_slider_label, freeze_button, lang_combobox,
+ filemenu, response_now_button]