app.py

import gradio as gr
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import torch
import soundfile as sf

# Correctly load the Wav2Vec2Processor and model
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-xlsr-53")
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-xlsr-53")

def transcribe_audio(audio):
    """
    Takes an audio file, processes it using Hugging Face Wav2Vec2 model, 
    and returns the transcribed text.
    """
    # Read the audio file
    audio_input, _ = sf.read(audio.name)
    
    # Process audio input using the processor
    input_values = processor(audio_input, return_tensors="pt").input_values
    
    # Get model logits (raw prediction)
    logits = model(input_values).logits
    
    # Decode the prediction into text
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)
    
    return transcription[0]

# Create a Gradio interface for users to upload audio files
iface = gr.Interface(fn=transcribe_audio, 
                     inputs=gr.Audio(source="upload", type="file"), 
                     outputs="text", 
                     title="Voice Login System", 
                     description="Upload an audio file for transcription using Wav2Vec2 model.")
                     
iface.launch()