app.py

import subprocess
subprocess.run(["pip", "install", "gradio", "--upgrade"])
subprocess.run(["pip", "install", "transformers"])
subprocess.run(["pip", "install", "torchaudio", "--upgrade"])

import gradio as gr
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import torchaudio
import torch

# Load model and processor
processor = Wav2Vec2Processor.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-italian")
model = Wav2Vec2ForCTC.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-italian")

def preprocess_audio(audio_data):
    # Apply any custom preprocessing to the audio data here if needed
    return processor(audio_data, return_tensors="pt").input_features

# Function to perform ASR on audio data
def transcribe_audio(input_features):
    print("Received audio data:", input_features)  # Debug print

    # Check if audio_data is None or not a tuple of length 2
    if audio_data is None or not isinstance(input_features, tuple) or len(input_features) != 2:
        return "Invalid audio data format."

    sample_rate, waveform = input_features

    # Check if waveform is None or not a NumPy array
    if waveform is None or not isinstance(waveform, torch.Tensor):
        return "Invalid audio data format."

    try:
        # Convert audio data to mono and normalize
        audio_data = torchaudio.transforms.Resample(sample_rate, 100000)(waveform)
        audio_data = torchaudio.functional.gain(input_features, gain_db=5.0)

        # Apply custom preprocessing to the audio data if needed
        input_values = processor(input_features[0], return_tensors="pt").input_values

        # Perform ASR
        with torch.no_grad():
            logits = model(input_values).logits

        # Decode the output
        predicted_ids = torch.argmax(logits, dim=-1)
        transcription = processor.batch_decode(predicted_ids)

        return transcription[0]

    except Exception as e:
        return f"An error occurred: {str(e)}"

# Create Gradio interface
audio_input = gr.Audio(sources=["microphone"])
gr.Interface(fn=transcribe_audio, inputs=audio_input, outputs="text").launch()