app.py

from transformers import WhisperProcessor, WhisperForConditionalGeneration
import gradio as gr
import torch
import torchaudio
from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read

MODEL_NAME = "dataprizma/whisper-large-v3-turbo"

processor = WhisperProcessor.from_pretrained(MODEL_NAME)
model = WhisperForConditionalGeneration.from_pretrained(MODEL_NAME)


def transcribe(audio_file):

    global model
    global processor

    # Move to GPU if available
    device = "cuda" if torch.cuda.is_available() else "cpu"
    model = model.to(device)

    # Load and preprocess audio
    waveform, sample_rate = torchaudio.load(audio_file)
    if sample_rate != 16000:
        waveform = torchaudio.functional.resample(waveform, sample_rate, 16000)

    # Convert to mono if needed
    if waveform.shape[0] > 1:
        waveform = waveform.mean(dim=0, keepdim=True)

    # Process audio
    input_features = processor(
        waveform.squeeze().numpy(),
        sampling_rate=16000,
        return_tensors="pt",
        language="uz"
    ).input_features.to(device)

    # Generate transcription
    with torch.no_grad():
        predicted_ids = model.generate(input_features)

    # Decode
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
    return transcription

demo = gr.Blocks()

file_transcribe = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type="filepath", label="Audio file"),
    outputs="text",
    title="Whisper Large V3: Transcribe Audio",
    description="Whisper Large V3 fine-tuned for Uzbek language by Dataprizma",
)

with demo:
    gr.TabbedInterface([file_transcribe], ["Audio file"])

demo.launch()