app.py

import gradio as gr
from transformers import pipeline

# Load pipelines for Canary ASR, LLama3 QA, and VITS TTS
asr_pipeline = pipeline("automatic-speech-recognition", model="nvidia/canary-1b", device=0)
qa_pipeline = pipeline("question-answering", model="LLAMA/llama3-base-qa", tokenizer="LLAMA/llama3-base-qa")
tts_pipeline = pipeline("text-to-speech", model="patrickvonplaten/vits-large", device=0)

# Function to capture audio using Canary ASR
def capture_audio():
    print("Listening for cue words...")
    while True:
        audio_input = asr_pipeline(None)[0]['input_values']
        transcript = asr_pipeline(audio_input)[0]['transcription']
        if "hey canary" in transcript.lower():
            print("Cue word detected!")
            break
    print("Listening...")
    return audio_input

# AI assistant function
def ai_assistant(audio_input):
    # Perform automatic speech recognition (ASR)
    transcript = asr_pipeline(audio_input)[0]['transcription']

    # Perform question answering (QA)
    qa_result = qa_pipeline(question=transcript, context="Insert your context here")

    # Convert the QA result to speech using text-to-speech (TTS)
    tts_output = tts_pipeline(qa_result['answer'])

    return tts_output[0]['audio']

if __name__ == "__main__":
    # Create a Gradio interface
    gr.Interface(ai_assistant,
                 inputs=gr.inputs.Audio(capture=capture_audio, label="Speak Here"),
                 outputs=gr.outputs.Audio(type="audio", label="Assistant's Response"),
                 title="AI Assistant",
                 description="An AI Assistant that answers questions based on your speech input.").launch()