import gradio as gr
import os
from transformers import pipeline
from langdetect import detect

def process_audio(audio_file):
    try:
        # audio_file is a tuple (file_obj, file_path)
        audio_path = audio_file if isinstance(audio_file, str) else audio_file.name

        # Transcribe
        asr = pipeline("automatic-speech-recognition", model="openai/whisper-large")
        result = asr(audio_path)
        transcript = result["text"]
    except Exception as e:
        return "Error in transcription: " + str(e), "", "", ""
    try:
        detected_lang = detect(transcript)
    except Exception:
        detected_lang = "unknown"
    lang_map = {'en': 'English', 'hi': 'Hindi', 'ta': 'Tamil'}
    lang_text = lang_map.get(detected_lang, detected_lang)
    transcript_en = transcript
    if detected_lang != "en":
        try:
            asr_translate = pipeline(
                "automatic-speech-recognition",
                model="openai/whisper-large",
                task="translate"
            )
            result_translate = asr_translate(audio_path)
            transcript_en = result_translate["text"]
        except Exception as e:
            transcript_en = f"Error translating: {e}"
    try:
        summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
        summary = summarizer(transcript_en, max_length=100, min_length=30, do_sample=False)
        summary_text = summary[0]["summary_text"]
    except Exception as e:
        summary_text = f"Error summarizing: {e}"
    # Optionally, remove uploaded file if it's saved on disk
    return lang_text, transcript, transcript_en, summary_text

with gr.Blocks() as demo:
    gr.Markdown("## Audio Transcript, Translation & Summary (Powered by Whisper + Hugging Face)")
    audio_input = gr.Audio(source="upload", type="filepath", label="Upload MP3/WAV Audio")
    btn = gr