app.py

import gradio as gr
from gradio_client import Client

# Call the existing chat model
chat_client = Client("Futuresony/Mr.Events")

# Load the Whisper model - using the public space as the specific model was not found
# You might need to adjust this if a specific space for whisper-large-v3-turbo becomes available or accessible
try:
    whisper_client = Client("openai/whisper-large-v3-turbo")
except Exception as e:
    print(f"Could not load 'openai/whisper-large-v3-turbo', falling back to 'openai/whisper'. Error: {e}")
    whisper_client = Client("openai/whisper")

def chat_with_model(user_input):
    """Sends text input to the chat model and returns the response."""
    result = chat_client.predict(
        query=user_input,
        api_name="/chat"
    )
    return result

def transcribe_audio(audio_input):
    """Transcribes audio input using the Whisper model."""
    if audio_input is None:
        return ""
    try:
        # The API name might vary depending on the specific Whisper space
        # Common API names are "/predict", "/run"
        transcript = whisper_client.predict(audio_input, api_name="/predict")
        return transcript
    except Exception as e:
        print(f"Error during transcription: {e}")
        return f"Transcription failed: {e}"

def send_message(transcribed_text, manual_text_input):
    """Determines which input to send to the chat model and returns the response."""
    if transcribed_text:
        return chat_with_model(transcribed_text)
    elif manual_text_input:
        return chat_with_model(manual_text_input)
    else:
        return "Please provide audio or text input."

# Create the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## 💬 Test the ABSA Model Chat with Audio Input")
    with gr.Row():
        with gr.Column(scale=3):
            audio_input = gr.Audio(sources=["microphone", "upload"], label="Audio Input")
            transcribed_text = gr.Textbox(label="Transcribed Text", lines=3, interactive=False) # Make transcribed text non-interactive
            manual_text_input = gr.Textbox(label="Or type your message here", lines=3)
            submit_btn = gr.Button("Send")
        with gr.Column(scale=5):
            output_text = gr.Textbox(label="Model Response", lines=6, interactive=False) # Make output non-interactive

    # Connect interface elements to functions
    audio_input.change(fn=transcribe_audio, inputs=audio_input, outputs=transcribed_text)
    submit_btn.click(fn=send_message, inputs=[transcribed_text, manual_text_input], outputs=output_text)

# Launch the Gradio application
demo.launch(share=True) # share=True is required for Colab