import gradio as gr import openai import os from gtts import gTTS # from langdetect import detect # Initialize the OpenAI API with your key and organization ID openai.api_key = os.getenv('OPEN_API_KEY') # all_messages = [] # List to store all conversation messages all_messages = [{"role": "system", "content": 'You are an AI chatbot. Respond to all input in 25 words or less'}] # def initialize_session(): # return [{"role": "system", "content": 'You are an AI chatbot. Respond to all input in 25 words or less'}] # all_messages = initialize_session() open("output_audio.mp3", "wb").close() def transcribe(audio, text): global all_messages language = 'en' transcript = {} # if audio: # audio_filename_with_extension = audio + '.wav' # os.rename(audio, audio_filename_with_extension) # audio_file = open(audio_filename_with_extension, "rb") # # Transcribe the audio file using the whisper-1 model # transcript = openai.audio.transcriptions.create(model="whisper-1", file=audio_file) # print("Transcription:", transcript) # # user_input = transcript("text") if isinstance(transcript, dict) else "" # # user_input = transcript.text if isinstance(transcript, dict) else "" # user_input = transcript.get("text", "") if isinstance(transcript, dict) else "" # user_input = transcript.text # print("Transcription:", user_input) # else : # user_input = text if audio: audio_filename_with_extension = audio + '.wav' os.rename(audio, audio_filename_with_extension) audio_file = open(audio_filename_with_extension, "rb") # Transcribe the audio file using the whisper-1 model transcript = openai.audio.transcriptions.create(model="whisper-1", file=audio_file) print("Transcription:", transcript) # user_input = transcript("text") if isinstance(transcript, dict) else "" # user_input = transcript.text if isinstance(transcript, dict) else "" user_input = transcript.get("text", "") if isinstance(transcript, dict) else "" user_input = transcript.text print("Transcription:", user_input) else: user_input = text # if not all_messages or (len(all_messages) == 1 and all_messages[0]["role"] == "system"): # all_messages = initialize_session() conversation_tokens = sum(len(message['content'].split()) for message in all_messages) # Limit the conversation tokens to the model's maximum limit (4096 - a safety margin) max_conversation_tokens = 4000 while conversation_tokens > max_conversation_tokens: oldest_message_tokens = len(all_messages[0]['content'].split()) if conversation_tokens - oldest_message_tokens > max_conversation_tokens: all_messages.pop(0) conversation_tokens -= oldest_message_tokens else: break # all_messages.append = [{"role": "system", "content": 'You are an AI chatbot. Respond to all input in 25 words or less'}] all_messages.append({"role": "user", "content": user_input}) # Generate a chat response using the gpt-3.5-turbo model response = openai.chat.completions.create(model="gpt-3.5-turbo", messages=all_messages) # Get the generated chat response from the model system_message = response.choices[0].message.content # all_messages.append(system_message.to_dict()) all_messages.append({"role": "assistant", "content": system_message}) chatbot_responses = [] # Initialize the list outside the loop # user_input = "" for message in all_messages: print(f"Message: {message}") # print(f"role: {message['role']}") if 'role' in message and 'content' in message: # print(f"role: {message['role']}") if message['role'] == 'user': user_input = message['content'] print(f"ui: {user_input}") elif message['role'] == 'assistant': chatbot_responses.append([user_input, message['content']]) # chatbot_responses.append(message['content']) # print(f"hdjdkdkksdk: {message['content']}") print(f"Assistant response appended: {message['content']}") print(f"cbr: {chatbot_responses}") # For English input, use English language voice tts = gTTS(text=chatbot_responses[-1][1], lang='en', slow=False) tts.save("output_audio.mp3") # tts = gTTS(text=chatbot_responses[-1][1], lang=language, slow=False) # detected_language = detect(user_input) # if detected_language == 'hi': # # For Hindi input, use the Hindi language voice # tts = gTTS(text=chatbot_responses[-1][1], lang='hi', slow=False, lang_check=False) # tts.save("output_audio.mp3") # else: # # # For English input, use English language voice # tts = gTTS(text=chatbot_responses[-1][1], lang='en', slow=False) # tts.save("output_audio.mp3") return [chatbot_responses[-1]], "output_audio.mp3" # Wrap the chatbot response in a list # return chatbot_responses, "output_audio.mp3" # out1 = gr.Audio("output_audio.mp3") css_styles = """ footer { visibility:hidden; } .audio_container { height: 100px; } .svelte-1e89no8 { padding: 10px; } .input_text_container { height: 100px; } @media only screen and (max-width: 450px) { .input_text_container { display: none; } } """ with gr.Blocks(theme=gr.themes.Monochrome(font=[gr.themes.GoogleFont("Inter"), "sans-serif"], spacing_size=gr.themes.sizes.spacing_md, radius_size=gr.themes.sizes.radius_md), css=css_styles, title="NanoAI Voice Assistant") as demo: # chatbot = gr.Chatbot().style(height=450) # Pass the all_messages list to the Chatbot widget chatbot = gr.Chatbot(clear_on_refresh=True).style(height=450) out1 = gr.Audio("output_audio.mp3", autoplay=True, label="Voice Response ") with gr.Row(): with gr.Column(scale=0.35): audio_input = gr.Audio(source="microphone", type="filepath", label="Record your question (Voice)", elem_classes="audio_container") with gr.Column(scale=0.65): text_input = gr.Textbox(label="Or Type your question here (Chat)", placeholder="What would you like to ask the bot...", elem_classes="input_text_container") with gr.Row(): with gr.Column(scale=0.50): text_button = gr.Button("Send a Message") with gr.Column(scale=0.50): reset_button = gr.Button("Clear ❌") text_button.click(transcribe, [audio_input, text_input], [chatbot, out1]) text_button.click(lambda x: gr.update(value=''), [], [text_input]) text_button.click(lambda: None, None, audio_input), text_button.click(lambda: None, None, out1), reset_button.click(lambda: None, None, chatbot), reset_button.click(lambda: None, None, out1) # audio_input = gr.Audio(source="microphone", type="filepath") demo.launch()