improve prompt and code

prompts/thanks_prompt.txt

@@ -0,0 +1 @@
+Generate a thank you message quoting the speaker's speech given below:\n\n

prompts/transcript_prompt.txt

@@ -0,0 +1 @@
+Generate a full transcript for an event. The transcript should start with a greeting and a welcome message, followed by inviting each speaker to the stage. Each speaker's invitation should end with the special character ###. The  invitations should have only dialogues without any headings or labels like "host:".  In the last part end the event with thanking all speakers and the attendees. The speaker details are as follows:

src/ai.py

@@ -6,23 +6,12 @@
 )
 
 
-def generate_transcript(document, client: OpenAI):
-    prompt = "Generate a full transcript for an event. \
-        The transcript should start with a greeting and welcome to the event message. \
-            Then, introduce each speaker and invite them to the stage. \
-                Write each speaker welcome as separate paragraph. \
-                    The speaker details are as follows:\n\n"
-
-    transcript = generate_text_with_prompt(document, prompt, client)
-    return transcript
-
-
-def generate_and_save_transcript(pdf_file, client: OpenAI):
+def generate_and_save_transcript(pdf_file, prompt, client: OpenAI):
     # Load the PDF file
     document = load_pdf_file(pdf_file)
 
     # Generate the transcript
-    transcript = generate_transcript(document, client)
+    transcript = generate_text_with_prompt(document, prompt, client)
 
     with open(os.path.join('..', 'transcripts', os.path.basename(pdf_file).replace('.pdf', '.txt')), 'w') as f:
         f.write(transcript)
@@ -38,9 +27,4 @@ def load_transcript(filename):
         transcript = file.read()
 
     # Return the contents of the file
-    return transcript
-
-
-def generate_thank_you_message(speech, client: OpenAI):
-    prompt = "Generate a thank you message quoting the speaker's speech given below:\n\n"
-    return generate_text_with_prompt(speech, prompt, client)
+    return transcript

src/audio.py

@@ -4,6 +4,7 @@
 import keyboard
 from vision import has_speaker_left_stage
 import scipy.io.wavfile as wav
+import cv2
 
 
 def speak(text, client: OpenAI):
@@ -80,7 +81,7 @@ def record_speech_vision(fs=44100):
     return filename
 
 
-def speech_to_text(filename, client: OpenAI):
+def transcribe(filename, client: OpenAI):
     # Open the audio file in binary mode
     with open(filename, 'rb') as audio_file:
         # Convert the audio file to text using OpenAI's Whisper ASR API

src/main.py

@@ -11,21 +11,28 @@
     api_key=os.environ.get("OPENAI_API_KEY"),
 )
 
+with open("prompts/transcript_prompt.txt", 'r') as f:
+    transcript_prompt = f.read()
+
+with open("prompts/thanks_prompt.txt", 'r') as f:
+    thanks_prompt = f.read()
+
 def main():
     parser = argparse.ArgumentParser(description='AI Event Host')
     parser.add_argument('command', help='Command to execute')
     parser.add_argument('pdf_file', help='PDF file with speaker details')
     args = parser.parse_args()
 
     if args.command == 'generate':
-        ai.generate_and_save_transcript(args.pdf_file, client)
+        ai.generate_and_save_transcript(args.pdf_file, transcript_prompt, client)
     elif args.command == 'start':
         transcript = ai.load_transcript(args.pdf_file)
-        for dialogue in transcript.split('\n\n'):
+        dialogues = transcript.split('###')
+        for dialogue in dialogues:
             audio.speak(dialogue, client)
             speech = audio.record_speech()
-            text = audio.speech_to_text(speech, client)
-            thank_you_message = ai.generate_thank_you_message(text, client)
+            text = audio.transcribe(speech, client)
+            thank_you_message = ai.generate_text_with_prompt(text, thanks_prompt, client)
             audio.speak(thank_you_message, client)
 
 if __name__ == '__main__':