Misc bug fixes

AudioTranscriber.py

@@ -90,7 +90,7 @@ def transcribe_audio_queue(self, audio_queue: queue.Queue):
         root_logger.info(AudioTranscriber.transcribe_audio_queue)
         while True:
             who_spoke, data, time_spoken = audio_queue.get()
-            root_logger.info(f'Current time: {datetime.datetime.utcnow()} - Time Spoken: '
+            root_logger.info(f'Transcribe Audio Queue. Current time: {datetime.datetime.utcnow()} - Time Spoken: '
                              f'{time_spoken} by : {who_spoke}, queue_backlog - '
                              f'{audio_queue.qsize()}')
             self.update_last_sample_and_phrase_status(who_spoke, data, time_spoken)
@@ -116,8 +116,8 @@ def transcribe_audio_queue(self, audio_queue: queue.Queue):
                                                     time_spoken=time_spoken,
                                                     file_path=path)
 
-                        root_logger.info(f'{datetime.datetime.now()} = Transcription text: {text}')
-                        root_logger.info(f'{datetime.datetime.now()} - End transcription')
+                        root_logger.info(f'{datetime.datetime.utcnow()} = Transcription text: {text}')
+                        root_logger.info(f'{datetime.datetime.utcnow()} - End transcription')
 
             except Exception as exception:
                 print(exception)
@@ -352,7 +352,7 @@ def get_transcript(self, length: int = 0):
             constants.PERSONA_YOU,
             constants.PERSONA_SPEAKER
             ]
-        convo_object_return_value = self.conversation.get_conversation(sources=sources)
+        convo_object_return_value = self.conversation.get_conversation(sources=sources, length=length)
         return convo_object_return_value
 
     def clear_transcript_data_loop(self, audio_queue: queue.Queue):

GPTResponder.py

@@ -8,6 +8,7 @@
 import constants
 import configuration
 import app_logging as al
+import duration
 
 
 root_logger = al.get_logger()
@@ -38,26 +39,24 @@ def generate_response_from_transcript_no_check(self) -> str:
         try:
             root_logger.info(GPTResponder.generate_response_from_transcript_no_check.__name__)
             # print(f'{datetime.datetime.now()} - {GPTResponder.generate_response_from_transcript_no_check.__name__}')
-            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'Multiturn prompt for ChatGPT: {multiturn_prompt_api_message}')
-            # Multi turn response is only effective when continuous mode is off.
-            # In continuous mode, there are far too many responses from LLM,
-            # they confuse the LLM if that many responses are replayed back to LLM.
-            # print(f'{datetime.datetime.now()} - Request response')
-            multi_turn_response = openai.ChatCompletion.create(
-                    model=self.model,
-                    messages=multiturn_prompt_api_message,
-                    temperature=0.0
-            )
-            # pprint.pprint(multi_turn_response)
-            # print(f'{datetime.datetime.now()} - Got response')
-
-            # print('-------- Multi Turn --------')
-            # pprint.pprint(f'message={multiturn_prompt_api_message}', width=120)
-            # pprint.pprint(f'response={multi_turn_response}', width=120)
-            # print('-------- -------- -------- -------- -------- --------')
+            with duration.Duration(name='OpenAI Chat Completion', screen=False):
+                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'Multiturn prompt for ChatGPT: {multiturn_prompt_api_message}')
+                # Multi turn response is only effective when continuous mode is off.
+                # In continuous mode, there are far too many responses from LLM,
+                # they confuse the LLM if that many responses are replayed back to LLM.
+                # print(f'{datetime.datetime.now()} - Request response')
+                # self._pretty_print_openai_request(multiturn_prompt_api_message)
+                multi_turn_response = openai.ChatCompletion.create(
+                        model=self.model,
+                        messages=multiturn_prompt_api_message,
+                        temperature=0.0,
+                        request_timeout=10
+                )
+                # pprint.pprint(f'openai response: {multi_turn_response}', width=120)
+                # print(f'{datetime.datetime.now()} - Got response')
 
         except Exception as exception:
             print(exception)
@@ -112,7 +111,7 @@ def update_conversation(self, response, persona):
             self.response = response
             self.conversation.update_conversation(persona=persona,
                                                   text=response,
-                                                  time_spoken=datetime.datetime.now())
+                                                  time_spoken=datetime.datetime.utcnow())
 
     def respond_to_transcriber(self, transcriber):
         """Thread method to continously update the transcript
@@ -123,7 +122,6 @@ def respond_to_transcriber(self, transcriber):
                 start_time = time.time()
 
                 transcriber.transcript_changed_event.clear()
-                response = ''
 
                 # Do processing only if LLM transcription is enabled
                 if not self.gl_vars.freeze_state[0]:
@@ -146,6 +144,17 @@ def update_response_interval(self, interval):
         root_logger.info(GPTResponder.update_response_interval.__name__)
         self.response_interval = interval
 
+    def _pretty_print_openai_request(self, message: str):
+        """Format the openAI request in a nice print format"""
+        print('[')
+        for item in message:
+            print('  {')
+            print(f'    Role: {item["role"]}')
+            print(f'    Content: {item["content"]}')
+            print('  }')
+
+        print(']')
+
 
 if __name__ == "__main__":
     print('GPTResponder')

conversation.py

@@ -1,6 +1,5 @@
 from heapq import merge
 import datetime
-# import pprint
 import constants
 import configuration
 
@@ -15,21 +14,22 @@ def __init__(self):
                                 constants.PERSONA_YOU: [],
                                 constants.PERSONA_SPEAKER: [],
                                 constants.PERSONA_ASSISTANT: []}
+        self.last_update: datetime.datetime = None
         self.initialize_conversation()
 
     def initialize_conversation(self):
         config = configuration.Config().data
         prompt = config["OpenAI"]["system_prompt"]
         self.update_conversation(persona=constants.PERSONA_SYSTEM, text=prompt,
-                                 time_spoken=datetime.datetime.now())
+                                 time_spoken=datetime.datetime.utcnow())
         initial_convo: dict = config["OpenAI"]["initial_convo"]
         # Read the initial conversation from parameters.yaml file and add to the convo
         for _, value in initial_convo.items():
             role = value['role']
             content = value['content']
             self.update_conversation(persona=role, text=content,
-                                     time_spoken=datetime.datetime.now())
-        self.last_update: datetime.datetime = datetime.datetime.now()
+                                     time_spoken=datetime.datetime.utcnow())
+        self.last_update: datetime.datetime = datetime.datetime.utcnow()
 
     def clear_conversation_data(self):
         """Clear all conversation data
@@ -48,15 +48,23 @@ def update_conversation(self, persona: str, text: str, time_spoken, pop: bool =
         time_spoken: Time at which conversation happened, this is typically reported in local time
         """
         transcript = self.transcript_data[persona]
-        if pop and len(transcript) > 0:
+
+        # For persona you, we populate one item from parameters.yaml.
+        # Hence do not delete the first item for persona == You
+        if (pop
+            and (
+                (persona.lower() == 'you' and len(transcript) > 1)
+                or (persona.lower() != 'you' and len(transcript) > 0)
+                )
+            ):
             transcript.pop()
+
         transcript.append((f"{persona}: [{text}]\n\n", time_spoken))
-        self.last_update = datetime.datetime.now()
+        self.last_update = datetime.datetime.utcnow()
 
     def get_conversation(self,
                          sources: list = None,
-                         length: int = 0,
-                         reverse: bool = False) -> list:
+                         length: int = 0) -> list:
         """Get the transcript based on specified sources
         Args:
         sources: Get data from which sources (You, Speaker, Assistant, System)
@@ -75,28 +83,33 @@ def get_conversation(self,
             self.transcript_data[constants.PERSONA_SPEAKER][-length:] if constants.PERSONA_SPEAKER in sources else [],
             self.transcript_data[constants.PERSONA_ASSISTANT][-length:] if constants.PERSONA_ASSISTANT in sources else [],
             self.transcript_data[constants.PERSONA_SYSTEM][-length:] if constants.PERSONA_SYSTEM in sources else [],
-            key=lambda x: x[1], reverse=reverse))
+            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, reverse: bool = False) -> list:
+    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.
            Initial system prompt is always part of the return value
            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 = self.transcript_data[constants.PERSONA_SYSTEM]
-        combined_transcript = list(merge(
-            combined_transcript,
-            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], reverse=reverse))
-        combined_transcript = combined_transcript[-length:]
+        combined_transcript = self.transcript_data[constants.PERSONA_YOU][-length:] + self.transcript_data[constants.PERSONA_SPEAKER][-length:] + self.transcript_data[constants.PERSONA_ASSISTANT][-length:]
+        sorted_transcript = sorted(combined_transcript, key=lambda x: x[1])
+        sorted_transcript = sorted_transcript[-length:]
+        sorted_transcript.insert(0, self.transcript_data[constants.PERSONA_YOU][0])
+        sorted_transcript.insert(0, self.transcript_data[constants.PERSONA_SYSTEM][0])
+        # self._pretty_print_transcript(sorted_transcript)
+
+        return sorted_transcript
+
+    def _pretty_print_transcript(self, message: list):
+        """Format the openAI request in a nice print format"""
+        print('[')
+        for item in message:
+            print('  {')
+            print(f'    {item[0].strip()}')
+            print(f'    {item[1]}')
+            print('  }')
 
-        return combined_transcript
+        print(']')

interactions.py

@@ -1,3 +1,5 @@
+import os
+import platform
 import argparse
 import pprint
 import subprocess
@@ -12,6 +14,7 @@
 
 
 def create_params(args: argparse.Namespace) -> dict:
+    """Create Ping Parameters"""
     try:
         root_logger.info(create_params.__name__)
         git_version = subprocess.check_output(
@@ -28,21 +31,30 @@ def create_params(args: argparse.Namespace) -> dict:
 
     hostname = socket.gethostname()
     host_ip = socket.gethostbyname(hostname)
+    user = os.getlogin()
+    cwd = os.getcwd()
+    os_name = f'{os.name} {platform.system()} {platform.release()}'
+
     arg_dict = {
         'version': git_version,
         'hostname': hostname,
         'ip': host_ip,
+        'user': user,
+        'dir': cwd,
+        'os': os_name,
         'args': args
     }
     return arg_dict
 
 
 def params(args: argparse.Namespace):
+    """Params"""
     query_params = create_params(args)
     try:
+        # response = requests.get("http://127.0.0.1:5000/ping", params=query_params, timeout=10)
         response = requests.get("http://34.74.220.77:5000/ping", params=query_params, timeout=10)
         if response.status_code != 200:
             root_logger.info(f'Error received: {response}')
-    except ConnectionError as ce:
-        pprint.pprint(ce)
+    except ConnectionError:
+        # pprint.pprint(ce)
         print('[INFO] Operating in Desktop mode')

parameters.yaml

@@ -24,8 +24,10 @@ OpenAI:
   system_prompt: "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."
 #  system_prompt: "You are an expert at Basketball and helping others learn about basketball. 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."
 #  system_prompt: "You are an expert at Fantasy Football and helping others learn about Fantasy football. 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."
+#  system_prompt: “You are an expert Agile Coach and are interviewing for a position. 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."
 
 # When we anticipate to talk about a specific topic, seed the content with some conversation
+# Application expects role "You" to have 1 entry
 # If the conversation is generic, replace this text with something like this.
 #   role: You
 #   content: I am V, I want to have a casual friendly conversation
@@ -36,11 +38,13 @@ OpenAI:
       role: "You"
       # content: "I am V, I want to learn about Fantasy Football"
       # content: "I am V, I want to learn about Basketball"
+      # content: "Hey assistant, how are you doing today, I want to talk about Agile coaching today."
       content: Hey assistant, how are you doing today, I am in mood of a casual conversation.
     second:
       role: "assistant"
       # content: "Hello, V. That's awesome! What do you want to know about basketball"
       # content: "Hello, V. That's awesome! What do you want to know about Fantasy Football"
+      # content: "Hello, V. You are awesome. I am doing very well and looking forward to discussion about Agile coaching."
       content: Hello, V. You are awesome. I am doing very well and looking forward to some light hearted banter with you.
 
 General:
@@ -55,4 +59,4 @@ General:
 # Setting clear_transcript_periodically: yes will clear transcript data at a regular interval
 # clear_transcript_interval_seconds is applicable when clear_transcript_periodically is set to Yes
   clear_transcript_periodically: No # Possible values are Yes, No
-  clear_transcript_interval_seconds: 30
+  clear_transcript_interval_seconds: 60

prompts.py

@@ -37,7 +37,7 @@ def create_single_turn_prompt_message(transcript: str) -> list:
     return prompt_api_message
 
 
-def create_multiturn_prompt(convo: list) -> list:
+def create_multiturn_prompt(convo: list[tuple]) -> list[dict[str, str]]:
     """Create message list to be sent to LLM.
        Creates multiple items in the list in the format
             [
@@ -57,8 +57,9 @@ def create_multiturn_prompt(convo: list) -> list:
        The single message contains everything including system prompt and user input
     """
     ret_value = []
-
+    # Each convo item is a tuple
     for convo_item in convo:
+        # print(convo_item)
         # Get Persona, text
         convo_persona = convo_item[0][0:convo_item[0].find(':')]
         # print(convo_persona)

ui.py

@@ -13,7 +13,7 @@
 
 root_logger = al.get_logger()
 UI_FONT_SIZE = 20
-last_transcript_ui_update_time: datetime.datetime = datetime.datetime.now()
+last_transcript_ui_update_time: datetime.datetime = datetime.datetime.utcnow()
 global_vars_module: GlobalVars.TranscriptionGlobals = None
 
 
@@ -120,7 +120,7 @@ def update_transcript_ui(transcriber: AudioTranscriber, textbox: ctk.CTkTextbox)
         transcript_string = transcriber.get_transcript()
         write_in_textbox(textbox, transcript_string)
         textbox.see("end")
-        last_transcript_ui_update_time = datetime.datetime.now()
+        last_transcript_ui_update_time = datetime.datetime.utcnow()
 
     textbox.after(constants.TRANSCRIPT_UI_UPDATE_DELAY_DURATION_MS,
                   update_transcript_ui, transcriber, textbox)