app.py

import gradio as gr
from transformers import pipeline


from datasets import DatasetDict, Dataset, load_dataset, Audio
from transformers import WhisperProcessor, WhisperForConditionalGeneration

def transcribe(audio):
    # load model and processor
    processor = WhisperProcessor.from_pretrained("openai/whisper-medium")
    model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-medium")
    
    ds = Dataset.from_dict({"audio": [audio]}).cast_column("audio", Audio())
    ds = ds.cast_column("audio", Audio(sampling_rate=16_000))
    input_speech = next(iter(ds))["audio"]["array"]
    
    input_features = processor(input_speech, return_tensors="pt").input_features 
    forced_decoder_ids = processor.get_decoder_prompt_ids(language = "no", task = "transcribe")

    predicted_ids = model.generate(input_features, forced_decoder_ids = forced_decoder_ids)
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens = True)

    return transcription


gr.Interface(
    title = "OpenAI Whisper ASR Gradio Norwegian Web UI",
    fn=transcribe,
    inputs=[
        gr.inputs.Audio(type="filepath")
    ],
    outputs=[
        "textbox"
    ]
).launch()