app.py

import subprocess
subprocess.run(["pip", "install", "datasets"])
subprocess.run(["pip", "install", "transformers"])
subprocess.run(["pip", "install", "torch", "torchvision", "torchaudio", "-f", "https://download.pytorch.org/whl/torch_stable.html"])

import gradio as gr
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import torchaudio

# Load model and processor
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
model.config.forced_decoder_ids = None

# Function to perform ASR on audio data
def transcribe_audio(audio_data):
    # Convert audio data to mono and normalize
    audio_data = torchaudio.functional.to_mono(audio_data)
    audio_data = torchaudio.functional.gain(audio_data, gain_db=5.0)

    # Resample if needed (Whisper model requires 16 kHz sampling rate)
    if audio_data[1] != 16000:
        audio_data = torchaudio.transforms.Resample(audio_data[1], 16000)(audio_data[0])

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

    # Generate token ids
    predicted_ids = model.generate(processed_input)

    # Decode token ids to text
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)

    return transcription[0]

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