app.py

import subprocess

subprocess.run(["python", "-m", "pip", "install", "--upgrade", "pip"])
subprocess.run(["pip", "install", "gradio", "--upgrade"])
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

# Load model and processor
processor = WhisperProcessor.from_pretrained("openai/whisper-large")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
forced_decoder_ids = processor.get_decoder_prompt_ids(language="italian", task="transcribe")

# Custom preprocessing function
def preprocess_audio(audio_data, sampling_rate=16_000):
    # Ensure that the input data is a valid format for the model
    # Convert the audio data to a numpy array with a correct shape
    raw_speech = np.asarray(audio_data, dtype=np.float32)

    # Pad or truncate the audio data to the required length
    if len(raw_speech) > processor.feature_extractor.max_len:
        raw_speech = raw_speech[:processor.feature_extractor.max_len]
    else:
        raw_speech = np.pad(raw_speech, (0, processor.feature_extractor.max_len - len(raw_speech)))

    # Process the audio data using the Whisper processor
    processed_data = processor(
        raw_speech,
        sampling_rate=sampling_rate,
        return_tensors="pt",
        padding=True,
        truncation=True
    )

    return processed_data.input_features

# Function to perform ASR on audio data
def transcribe_audio(audio_data):
    # Preprocess the audio data
    input_features = preprocess_audio(audio_data)

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

    # 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()