app.py

import gradio as gr
import whisper
import os
from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
from docx import Document
from reportlab.pdfgen import canvas
from reportlab.pdfbase.ttfonts import TTFont
from reportlab.pdfbase import pdfmetrics
from reportlab.lib.pagesizes import A4
import arabic_reshaper
from bidi.algorithm import get_display
from pptx import Presentation
import subprocess
import shlex
import yt_dlp

# Load the Whisper model (smaller model for faster transcription)
model = whisper.load_model("tiny")

# Load M2M100 translation model for different languages
def load_translation_model(target_language):
    lang_codes = {
        "fa": "fa",  # Persian (Farsi)
        "es": "es",  # Spanish
        "fr": "fr",  # French
        "de": "de",  # German
        "it": "it",  # Italian
        "pt": "pt",  # Portuguese
        "ar": "ar",  # Arabic
        "zh": "zh",  # Chinese
        "hi": "hi",  # Hindi
        "ja": "ja",  # Japanese
        "ko": "ko",  # Korean
        "ru": "ru",  # Russian
    }
    target_lang_code = lang_codes.get(target_language)
    if not target_lang_code:
        raise ValueError(f"Translation model for {target_language} not supported")

    tokenizer = M2M100Tokenizer.from_pretrained("facebook/m2m100_418M")
    translation_model = M2M100ForConditionalGeneration.from_pretrained("facebook/m2m100_418M")

    tokenizer.src_lang = "en"
    tokenizer.tgt_lang = target_lang_code

    return tokenizer, translation_model

def translate_text(text, tokenizer, model):
    try:
        inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
        translated = model.generate(**inputs, forced_bos_token_id=tokenizer.get_lang_id(tokenizer.tgt_lang))
        return tokenizer.decode(translated[0], skip_special_tokens=True)
    except Exception as e:
        raise RuntimeError(f"Error during translation: {e}")

# Helper function to format timestamps in SRT format
def format_timestamp(seconds):
    milliseconds = int((seconds % 1) * 1000)
    seconds = int(seconds)
    hours = seconds // 3600
    minutes = (seconds % 3600) // 60
    seconds = seconds % 60
    return f"{hours:02}:{minutes:02}:{seconds:02},{milliseconds:03}"

# Corrected write_srt function
def write_srt(transcription, output_file, tokenizer=None, translation_model=None):
    with open(output_file, "w") as f:
        for i, segment in enumerate(transcription['segments']):
            start = segment['start']
            end = segment['end']
            text = segment['text']
            
            if translation_model:
                text = translate_text(text, tokenizer, translation_model)
            
            start_time = format_timestamp(start)
            end_time = format_timestamp(end)
            
            f.write(f"{i + 1}\n")
            f.write(f"{start_time} --> {end_time}\n")
            f.write(f"{text.strip()}\n\n")

# Embedding subtitles into video (hardsub)
def embed_hardsub_in_video(video_file, srt_file, output_video):
    command = f'ffmpeg -i "{video_file}" -vf "subtitles=\'{srt_file}\'" -c:v libx264 -crf 23 -preset medium "{output_video}"'
    try:
        process = subprocess.run(shlex.split(command), capture_output=True, text=True, timeout=300)
        if process.returncode != 0:
            raise RuntimeError(f"ffmpeg error: {process.stderr}")
    except subprocess.TimeoutExpired:
        raise RuntimeError("ffmpeg process timed out.")
    except Exception as e:
        raise RuntimeError(f"Error running ffmpeg: {e}")

# Helper function to write Word documents
def write_word(transcription, output_file, tokenizer=None, translation_model=None, target_language=None):
    doc = Document()
    rtl = target_language == "fa"
    for i, segment in enumerate(transcription['segments']):
        text = segment['text']
        if translation_model:
            text = translate_text(text, tokenizer, translation_model)
        para = doc.add_paragraph(f"{i + 1}. {text.strip()}")
        if rtl:
            para.paragraph_format.right_to_left = True
    doc.save(output_file)

# Helper function to reverse text for RTL
def reverse_text_for_rtl(text):
    return ' '.join([word[::-1] for word in text.split()])

# Helper function to write PDF documents
def write_pdf(transcription, output_file, tokenizer=None, translation_model=None, target_language=None):
    # Create PDF with A4 page size
    c = canvas.Canvas(output_file, pagesize=A4)
    
    # Get the directory where app.py is located
    app_dir = os.path.dirname(os.path.abspath(__file__))

    # Define font paths for different languages
    fonts = {
        'fa': os.path.join(app_dir, 'B-NAZANIN.TTF'),  # Persian Font
        'ar': os.path.join(app_dir, 'Amiri-Regular.ttf'),  # Arabic Font
        'default': 'Arial'  # Default font for other languages
    }

    # Register and set the appropriate font
    font_path = fonts.get(target_language, fonts['default'])
    
    if os.path.exists(font_path):
        try:
            pdfmetrics.registerFont(TTFont('custom_font', font_path))
            c.setFont('custom_font', 12)
        except Exception as e:
            raise RuntimeError(f"Error registering font: {e}.")
    else:
        raise FileNotFoundError(f"Font file not found at {font_path}. Please ensure it is available.")

    # Initialize y position from top of page
    y_position = A4[1] - 50  # Start 50 points from top
    line_height = 20

    # Process each segment
    for i, segment in enumerate(transcription['segments']):
        text = segment['text']

        # Translate if translation model is provided
        if translation_model:
            text = translate_text(text, tokenizer, translation_model)

        # Format the line with segment number
        line = f"{i + 1}. {text.strip()}"

        # For RTL languages like Persian and Arabic, reshape and reorder text
        if target_language in ['fa', 'ar']:
            reshaped_text = arabic_reshaper.reshape(line)
            bidi_text = get_display(reshaped_text)
        else:
            bidi_text = line  # For LTR languages, no reshaping needed

        # Add new page if needed
        if y_position < 50:  # Leave 50 points margin at bottom
            c.showPage()
            c.setFont('custom_font', 12)
            y_position = A4[1] - 50

        # Draw the text right-aligned for RTL languages, otherwise left-aligned
        if target_language in ['fa', 'ar']:
            c.drawRightString(A4[0] - 50, y_position, bidi_text)  # Right align for RTL
        else:
            c.drawString(50, y_position, bidi_text)  # Left align for LTR

        # Update y position for next line
        y_position -= line_height

    # Save the PDF
    c.save()
    return output_file



# Helper function to write PowerPoint slides
def write_ppt(transcription, output_file, tokenizer=None, translation_model=None):
    ppt = Presentation()
    for i, segment in enumerate(transcription['segments']):
        text = segment['text']
        if translation_model:
            text = translate_text(text, tokenizer, translation_model)
        slide = ppt.slides.add_slide(ppt.slide_layouts[5])
        title = slide.shapes.title
        title.text = f"{i + 1}. {text.strip()}"
    ppt.save(output_file)

# Function to download YouTube video
def download_youtube_video(url):
    ydl_opts = {
        'format': 'mp4',
        'outtmpl': 'downloaded_video.mp4',
    }
    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
        ydl.download([url])
    return 'downloaded_video.mp4'

# Transcribing video and generating output
def transcribe_video(video_file, video_url, language, target_language, output_format):
    if video_url:
        video_file_path = download_youtube_video(video_url)
    else:
        video_file_path = video_file.name

    result = model.transcribe(video_file_path, language=language)
    video_name = os.path.splitext(video_file_path)[0]
    if target_language != "en":
        try:
            tokenizer, translation_model = load_translation_model(target_language)
        except Exception as e:
            raise RuntimeError(f"Error loading translation model: {e}")
    else:
        tokenizer, translation_model = None, None

    srt_file = f"{video_name}.srt"
    write_srt(result, srt_file, tokenizer, translation_model)

    if output_format == "SRT":
        return srt_file
    elif output_format == "Video with Hardsub":
        output_video = f"{video_name}_with_subtitles.mp4"
        try:
            embed_hardsub_in_video(video_file_path, srt_file, output_video)
            return output_video
        except Exception as e:
            raise RuntimeError(f"Error embedding subtitles in video: {e}")
    elif output_format == "Word":
        word_file = f"{video_name}.docx"
        write_word(result, word_file, tokenizer, translation_model, target_language)
        return word_file
    elif output_format == "PDF":
        pdf_file = f"{video_name}.pdf"
        write_pdf(result, pdf_file, tokenizer, translation_model)
        return pdf_file
    elif output_format == "PowerPoint":
        ppt_file = f"{video_name}.pptx"
        write_ppt(result, ppt_file, tokenizer, translation_model)
        return ppt_file

# Gradio interface with YouTube URL
iface = gr.Interface(
    fn=transcribe_video,
    inputs=[
        gr.File(label="Upload Video File (or leave empty for YouTube link)"),  # Removed 'optional=True'
        gr.Textbox(label="YouTube Video URL (optional)", placeholder="https://www.youtube.com/watch?v=..."),
        gr.Dropdown(label="Select Original Video Language", choices=["en", "es", "fr", "de", "it", "pt"], value="en"),
        gr.Dropdown(label="Select Subtitle Translation Language", choices=["en", "fa", "es", "de", "fr", "it", "pt"], value="fa"),
        gr.Radio(label="Choose Output Format", choices=["SRT", "Video with Hardsub", "Word", "PDF", "PowerPoint"], value="Video with Hardsub")
    ],
    outputs=gr.File(label="Download File"),
    title="Video Subtitle Generator with Translation & Multi-Format Output (Supports YouTube)",
    description=(
        "This tool allows you to generate subtitles from a video file or YouTube link using Whisper, "
        "translate the subtitles into multiple languages using M2M100, and export them "
        "in various formats including SRT, hardcoded subtitles in video, Word, PDF, or PowerPoint."
    ),
    theme="compact",
    live=False
)

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