app.py

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

# Load the Wav2Vec2 model and processor
model_name = "facebook/wav2vec2-base-960h"
processor = Wav2Vec2Processor.from_pretrained(model_name)
model = Wav2Vec2ForCTC.from_pretrained(model_name)

# Function to process a single chunk of audio
def process_chunk(chunk, sample_rate):
    # Resample the audio to 16000 Hz if necessary
    if sample_rate != 16000:
        resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
        chunk = resampler(chunk)
    
    # Ensure the audio is in the correct format
    chunk = chunk.squeeze().numpy()
    
    # Process the audio to the format expected by the model
    input_values = processor(chunk, sampling_rate=16000, return_tensors="pt").input_values
    
    # Perform inference
    with torch.no_grad():
        logits = model(input_values).logits
    
    # Decode the logits to get the predicted text
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)[0]
    
    return transcription

# Function to perform speech recognition on the entire audio
def speech_recognition(audio_path):
    # Load the audio file
    waveform, sample_rate = torchaudio.load(audio_path)
    
    # Split the waveform into chunks of 30 seconds
    chunk_length = 30 * sample_rate  # 30 seconds in samples
    chunks = [waveform[:, i:i + chunk_length] for i in range(0, waveform.size(1), chunk_length)]
    
    # Use multiprocessing to process chunks in parallel
    with mp.Pool(mp.cpu_count()) as pool:
        results = pool.starmap(process_chunk, [(chunk, sample_rate) for chunk in chunks])
    
    # Combine the transcriptions
    transcription = " ".join(results)
    
    return transcription.strip()

# Create the Gradio interface
inputs = gr.Audio(type="filepath", label="Input Audio")
outputs = gr.Textbox(label="Transcription")

interface = gr.Interface(
    fn=speech_recognition,
    inputs=inputs,
    outputs=outputs,
    title="Speech Recognition using Wav2Vec2",
    description="Upload a audio file or record the audio to get the transcription using the Wav2Vec2 model.",
    article="This assignement is developed by Pranshu Swaroop",
)

# Launch the interface
if __name__ == "__main__":
    interface.launch()