from transformers import pipeline
import gradio as gr
import requests

def correct_polish_text(text):
    api_url = "https://api.languagetoolplus.com/v2/check"
    params = {
        "text": text,
        "language": "pl",
    }
    response = requests.post(api_url, data=params)
    if response.status_code == 200:
        matches = response.json().get("matches", [])
        corrected_text = text
        for match in reversed(matches):
            start = match["offset"]
            end = start + match["length"]
            replacement = match["replacements"][0]["value"] if match["replacements"] else text[start:end]
            corrected_text = corrected_text[:start] + replacement + corrected_text[end:]
        return corrected_text
    else:
        return text
        
pipe = pipeline(model="marcsixtysix/whisper-base-pl")

def transcribe(audio):
    text = pipe(audio)["text"]
    corrected_text = correct_polish_text(text)
    return corrected_text

demo = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="Polish speech recognition",
    description="Realtime demo for Polish speech recognition using a fine-tuned Whisper Base model.",
)
demo.launch(share=True)