app.py

import gradio as gr
import ffmpeg
import io
import tempfile
import srt
import datetime
from transformers import WhisperForConditionalGeneration, WhisperProcessor, pipeline, MT5ForConditionalGeneration, MT5Tokenizer

# مدل‌ها
asr_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base.en")
asr_processor = WhisperProcessor.from_pretrained("openai/whisper-base.en")
translator = pipeline("text2text-generation", model="persiannlp/mt5-small-parsinlu-translation_en_fa", tokenizer="persiannlp/mt5-small-parsinlu-translation_en_fa")

def process(video):
    # استخراج فایل صوتی
    ytmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    (
        ffmpeg
        .input(video)
        .output(ytmp.name, format="wav", ac=1, ar="16000")
        .run(quiet=True, overwrite_output=True)
    )
    # ASR
    input_feat = asr_processor(ytmp.name, return_tensors="pt", sampling_rate=16000)
    out = asr_model.generate(**input_feat)
    segments = asr_processor.batch_decode(out, skip_special_tokens=True)  # متن کامل

    # برای زمان‌بندی دستی، تقسیم به جملات
    subs = []
    text = segments[0]
    lines = [l for l in text.split('.') if l.strip()]
    for i, line in enumerate(lines):
        start = datetime.timedelta(seconds=i*5)
        end = datetime.timedelta(seconds=(i+1)*5)
        # ترجمه
        tr = translator(line.strip(), max_length=128)[0]['generated_text']
        subs.append(srt.Subtitle(index=i+1, start=start, end=end, content=tr))

    srt_data = srt.compose(subs)
    return video, srt_data

demo = gr.Interface(
    fn=process,
    inputs=gr.Video(source="upload", format="mp4"),
    outputs=[gr.Video(), gr.Textbox(label="Subtitles (SRT)")],
    title="Offline English→Persian Subtitle Maker",
    description="ویدیو را آپلود کنید؛ زیرنویس فارسی تولید می‌شود."
)

if __name__ == "__main__":
    demo.launch()