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 # Define available Whisper models whisper_models = { "Tiny (Fast, Less Accurate)": "tiny", "Base (Medium Speed, Medium Accuracy)": "base", "Small (Good Speed, Good Accuracy)": "small", "Medium (Slow, High Accuracy)": "medium", "Large (Very Slow, Highest Accuracy)": "large" } # 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 write PDF documents def write_pdf(transcription, output_file, tokenizer=None, translation_model=None): # Create PDF with A4 page size c = canvas.Canvas(output_file, pagesize=A4) app_dir = os.path.dirname(os.path.abspath(__file__)) # Register fonts nazanin_font_path = os.path.join(app_dir, 'B-NAZANIN.TTF') arial_font_path = os.path.join(app_dir, 'Arial.ttf') if os.path.exists(nazanin_font_path): pdfmetrics.registerFont(TTFont('B-Nazanin', nazanin_font_path)) if os.path.exists(arial_font_path): pdfmetrics.registerFont(TTFont('Arial', arial_font_path)) y_position = A4[1] - 50 line_height = 20 for i, segment in enumerate(transcription['segments']): text = segment['text'] if translation_model: text = translate_text(text, tokenizer, translation_model) line = f"{i + 1}. {text.strip()}" target_language = tokenizer.tgt_lang if translation_model else None if target_language in ['fa', 'ar']: reshaped_text = arabic_reshaper.reshape(line) bidi_text = get_display(reshaped_text) c.setFont('B-Nazanin', 12) c.drawRightString(A4[0] - 50, y_position, bidi_text) else: c.setFont('Arial', 12) c.drawString(50, y_position, line) if y_position < 50: c.showPage() y_position = A4[1] - 50 y_position -= line_height c.save() return output_file # Helper function to write PowerPoint slides def write_ppt(transcription, output_file, tokenizer=None, translation_model=None): ppt = Presentation() slide = ppt.slides.add_slide(ppt.slide_layouts[5]) text_buffer = "" max_chars_per_slide = 400 for i, segment in enumerate(transcription['segments']): text = segment['text'] if translation_model: text = translate_text(text, tokenizer, translation_model) line = f"{i + 1}. {text.strip()}\n" if len(text_buffer) + len(line) > max_chars_per_slide: slide.shapes.title.text = "Transcription" textbox = slide.shapes.add_textbox(left=0, top=0, width=ppt.slide_width, height=ppt.slide_height) textbox.text = text_buffer.strip() slide = ppt.slides.add_slide(ppt.slide_layouts[5]) text_buffer = line else: text_buffer += line if text_buffer: slide.shapes.title.text = "" textbox = slide.shapes.add_textbox(left=0, top=0, width=ppt.slide_width, height=ppt.slide_height) textbox.text = text_buffer.strip() ppt.save(output_file) # Transcribing video and generating output def transcribe_video(video_file, language, target_language, model_name, output_format): actual_model_name = whisper_models[model_name] # Map user selection to model name model = whisper.load_model(actual_model_name) # Load the selected model if video_file is not None: # Ensure the video_file is not None video_file_path = video_file.name else: raise ValueError("No video file provided. Please upload a video file.") 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 else: raise ValueError("Invalid output format selected.") # Gradio Interface setup iface = gr.Interface( fn=transcribe_video, inputs=[ gr.File(label="Upload Video File"), 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.Dropdown(label="Select Whisper Model", choices=list(whisper_models.keys()), value="Tiny (Fast, Less Accurate)"), 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", description=( "This tool allows you to generate subtitles from a video file, 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 ) # Run the interface iface.launch(share=True)