import base64 import tempfile import os import requests import gradio as gr import random from openai import OpenAI # Available voices for audio generation VOICES = ["alloy", "ash", "ballad", "coral", "echo", "fable", "onyx", "nova", "sage", "shimmer", "verse"] # Example audio URLs EXAMPLE_AUDIO_URLS = [ "https://cdn.openai.com/API/docs/audio/alloy.wav", "https://cdn.openai.com/API/docs/audio/ash.wav", "https://cdn.openai.com/API/docs/audio/coral.wav", "https://cdn.openai.com/API/docs/audio/echo.wav", "https://cdn.openai.com/API/docs/audio/fable.wav", "https://cdn.openai.com/API/docs/audio/onyx.wav", "https://cdn.openai.com/API/docs/audio/nova.wav", "https://cdn.openai.com/API/docs/audio/sage.wav", "https://cdn.openai.com/API/docs/audio/shimmer.wav" ] def process_text_input(api_key, text_prompt, selected_voice): """Generate audio response from text input""" try: # Initialize OpenAI client with the provided API key client = OpenAI(api_key=api_key) completion = client.chat.completions.create( model="gpt-4o-audio-preview", modalities=["text", "audio"], audio={"voice": selected_voice, "format": "wav"}, messages=[ { "role": "user", "content": text_prompt } ] ) # Save the audio to a temporary file wav_bytes = base64.b64decode(completion.choices[0].message.audio.data) temp_path = tempfile.mktemp(suffix=".wav") with open(temp_path, "wb") as f: f.write(wav_bytes) # Get the text response directly from the API text_response = completion.choices[0].message.content return text_response, temp_path except Exception as e: return f"Error: {str(e)}", None def process_audio_input(api_key, audio_path, text_prompt, selected_voice): """Process audio input and generate a response""" try: if not audio_path: return "Please upload or record audio first.", None # Initialize OpenAI client with the provided API key client = OpenAI(api_key=api_key) # Read audio file and encode to base64 with open(audio_path, "rb") as audio_file: audio_data = audio_file.read() encoded_audio = base64.b64encode(audio_data).decode('utf-8') # Create message content with both text and audio message_content = [] if text_prompt: message_content.append({ "type": "text", "text": text_prompt }) message_content.append({ "type": "input_audio", "input_audio": { "data": encoded_audio, "format": "wav" } }) # Call OpenAI API completion = client.chat.completions.create( model="gpt-4o-audio-preview", modalities=["text", "audio"], audio={"voice": selected_voice, "format": "wav"}, messages=[ { "role": "user", "content": message_content } ] ) # Save the audio response wav_bytes = base64.b64decode(completion.choices[0].message.audio.data) temp_path = tempfile.mktemp(suffix=".wav") with open(temp_path, "wb") as f: f.write(wav_bytes) # Get the text response text_response = completion.choices[0].message.content return text_response, temp_path except Exception as e: return f"Error: {str(e)}", None def transcribe_audio(api_key, audio_path): """Transcribe an audio file using OpenAI's API""" try: if not audio_path: return "No audio file provided for transcription." client = OpenAI(api_key=api_key) with open(audio_path, "rb") as audio_file: transcription = client.audio.transcriptions.create( model="gpt-4o-transcribe", file=audio_file ) return transcription.text except Exception as e: return f"Transcription error: {str(e)}" def download_example_audio(): """Download a random example audio file for testing""" try: # Randomly select one of the example audio URLs url = random.choice(EXAMPLE_AUDIO_URLS) # Get the voice name from the URL for feedback voice_name = url.split('/')[-1].split('.')[0] response = requests.get(url) response.raise_for_status() # Save to a temporary file temp_path = tempfile.mktemp(suffix=".wav") with open(temp_path, "wb") as f: f.write(response.content) return temp_path, f"Loaded example voice: {voice_name}" except Exception as e: return None, f"Error loading example: {str(e)}" def use_example_audio(): """Load random example audio for the interface""" audio_path, message = download_example_audio() return audio_path, message # Create Gradio Interface with gr.Blocks(title="OpenAI Audio Chat App") as app: gr.Markdown("# OpenAI Audio Chat App") gr.Markdown("Interact with GPT-4o audio model through text and audio inputs") # API Key input (used across all tabs) api_key = gr.Textbox( label="OpenAI API Key", placeholder="Enter your OpenAI API key here", type="password" ) with gr.Tab("Text to Audio"): with gr.Row(): with gr.Column(): text_input = gr.Textbox( label="Text Prompt", placeholder="Enter your question or prompt here...", lines=3 ) text_voice = gr.Dropdown( choices=VOICES, value="alloy", label="Voice" ) text_submit = gr.Button("Generate Response") with gr.Column(): text_output = gr.Textbox(label="AI Response (Checks Error)", lines=5) audio_output = gr.Audio(label="AI Response (Audio)") transcribed_output = gr.Textbox(label="Transcription of Audio Response", lines=3) # Function to process text input and then transcribe the resulting audio def text_input_with_transcription(api_key, text_prompt, voice): text_response, audio_path = process_text_input(api_key, text_prompt, voice) # Get transcription of the generated audio if audio_path: transcription = transcribe_audio(api_key, audio_path) else: transcription = "No audio generated to transcribe." return text_response, audio_path, transcription text_submit.click( fn=text_input_with_transcription, inputs=[api_key, text_input, text_voice], outputs=[text_output, audio_output, transcribed_output] ) with gr.Tab("Audio Input to Audio Response"): with gr.Row(): with gr.Column(): audio_input = gr.Audio( label="Audio Input", type="filepath", sources=["microphone", "upload"] ) example_btn = gr.Button("Use Random Example Audio") example_message = gr.Textbox(label="Example Status", interactive=False) accompanying_text = gr.Textbox( label="Accompanying Text (Optional)", placeholder="Add any text context or question about the audio...", lines=2 ) audio_voice = gr.Dropdown( choices=VOICES, value="alloy", label="Response Voice" ) audio_submit = gr.Button("Process Audio & Generate Response") with gr.Column(): audio_text_output = gr.Textbox(label="AI Response (Checks Error)", lines=5) audio_audio_output = gr.Audio(label="AI Response (Audio)") audio_transcribed_output = gr.Textbox(label="Transcription of Audio Response", lines=3) input_transcription = gr.Textbox(label="Transcription of Input Audio", lines=3) # Function to process audio input, generate response, and provide transcriptions def audio_input_with_transcription(api_key, audio_path, text_prompt, voice): # First transcribe the input audio input_transcription = "N/A" if audio_path: input_transcription = transcribe_audio(api_key, audio_path) # Process the audio input and get response text_response, response_audio_path = process_audio_input(api_key, audio_path, text_prompt, voice) # Transcribe the response audio response_transcription = "No audio generated to transcribe." if response_audio_path: response_transcription = transcribe_audio(api_key, response_audio_path) return text_response, response_audio_path, response_transcription, input_transcription audio_submit.click( fn=audio_input_with_transcription, inputs=[api_key, audio_input, accompanying_text, audio_voice], outputs=[audio_text_output, audio_audio_output, audio_transcribed_output, input_transcription] ) example_btn.click( fn=use_example_audio, inputs=[], outputs=[audio_input, example_message] ) with gr.Tab("Voice Samples"): gr.Markdown("## Listen to samples of each voice") def generate_voice_sample(api_key, voice_type): try: if not api_key: return "Please enter your OpenAI API key first.", None, "No transcription available." client = OpenAI(api_key=api_key) completion = client.chat.completions.create( model="gpt-4o-audio-preview", modalities=["text", "audio"], audio={"voice": voice_type, "format": "wav"}, messages=[ { "role": "user", "content": f"This is a sample of the {voice_type} voice. It has its own unique tone and character." } ] ) # Save the audio to a temporary file wav_bytes = base64.b64decode(completion.choices[0].message.audio.data) temp_path = tempfile.mktemp(suffix=".wav") with open(temp_path, "wb") as f: f.write(wav_bytes) # Get transcription transcription = transcribe_audio(api_key, temp_path) return f"Sample generated with voice: {voice_type}", temp_path, transcription except Exception as e: return f"Error: {str(e)}", None, "No transcription available." with gr.Row(): sample_voice = gr.Dropdown( choices=VOICES, value="alloy", label="Select Voice Sample" ) sample_btn = gr.Button("Generate Sample") with gr.Row(): sample_text = gr.Textbox(label="Status") sample_audio = gr.Audio(label="Voice Sample") sample_transcription = gr.Textbox(label="Transcription", lines=3) sample_btn.click( fn=generate_voice_sample, inputs=[api_key, sample_voice], outputs=[sample_text, sample_audio, sample_transcription] ) gr.Markdown(""" ## Notes: - You must provide your OpenAI API key in the field above - The model used is `gpt-4o-audio-preview` for conversation and `gpt-4o-transcribe` for transcriptions - Audio inputs should be in WAV format - Available voices: alloy, ash, ballad, coral, echo, fable, onyx, nova, sage, shimmer, and verse - Each audio response is automatically transcribed for verification - The "Use Random Example Audio" button will load a random sample from OpenAI's demo voices """) if __name__ == "__main__": app.launch()