app.py

import gradio as gr
from transformers import pipeline
import numpy as np
import time

# Initialize the pipelines
transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-tiny.en")
classifier = pipeline("zero-shot-classification", model="MoritzLaurer/DeBERTa-v3-base-mnli-fever-anli")

candidate_labels = ["dim the light", "turn on light fully", "turn off light fully", "raise the light", "not about lighting"]
last_update_time = time.time() - 5  # Initialize with a value to ensure immediate first update

# Buffer to hold the last updated values
last_transcription = ""
last_classification = ""

def transcribe_and_classify(stream, new_chunk):
    global last_update_time, last_transcription, last_classification
    sr, y = new_chunk
    y = y.astype(np.float32)
    y /= np.max(np.abs(y))

    # Concatenate new audio chunk to the stream
    if stream is not None:
        stream = np.concatenate([stream, y])
    else:
        stream = y

    # Transcribe the last 10 seconds of audio
    transcription = transcriber({"sampling_rate": sr, "task": "transcribe", "language": "english", "raw": stream})["text"]
    last_transcription = transcription  # Update the buffer
    
    # Classify the transcribed text
    if transcription.strip():
        output = classifier(transcription, candidate_labels, multi_label=False)
        top_label = output['labels'][0]
        top_score = output['scores'][0]
        last_classification = f"{top_label.upper()}, score: {top_score:.2f}"
    
    # Return the last updated transcription and classification
    return stream, last_transcription, last_classification

# Define the Gradio interface
demo = gr.Interface(
    fn=transcribe_and_classify,
    inputs=[
        "state", 
        gr.Audio(sources=["microphone"])
    ],
    outputs=[
        "state", 
        "text", 
        "text"
    ],

)

# Launch the demo
demo.launch(debug=True)