import gradio as gr from faster_whisper import WhisperModel import logging import os import pysrt import pandas as pd from transformers import MarianMTModel, MarianTokenizer import ffmpeg # Configuration initiale et chargement des données url = "https://huggingface.co/Lenylvt/LanguageISO/resolve/main/iso.md" df = pd.read_csv(url, delimiter="|", skiprows=2, header=None).dropna(axis=1, how='all') df.columns = ['ISO 639-1', 'ISO 639-2', 'Language Name', 'Native Name'] df['ISO 639-1'] = df['ISO 639-1'].str.strip() language_options = [(row['ISO 639-1'], f"{row['ISO 639-1']}") for index, row in df.iterrows()] model_size_options = ["tiny", "base", "small", "medium", "large", "large-v2", "large-v3"] # Add model size options logging.basicConfig(level=logging.DEBUG) # Fonction pour formater un texte en SRT def text_to_srt(text): lines = text.split('\n') srt_content = "" for i, line in enumerate(lines): if line.strip() == "": continue try: times, content = line.split(']', 1) start, end = times[1:].split(' -> ') if start.count(":") == 1: start = "00:" + start if end.count(":") == 1: end = "00:" + end srt_content += f"{i+1}\n{start.replace('.', ',')} --> {end.replace('.', ',')}\n{content.strip()}\n\n" except ValueError: continue temp_file_path = '/tmp/output.srt' with open(temp_file_path, 'w', encoding='utf-8') as file: file.write(srt_content) return temp_file_path # Fonction pour formater des secondes en timestamp def format_timestamp(seconds): hours = int(seconds // 3600) minutes = int((seconds % 3600) // 60) seconds_remainder = seconds % 60 return f"{hours:02d}:{minutes:02d}:{seconds_remainder:06.3f}" # Fonction de traduction de texte def translate_text(text, source_language_code, target_language_code): model_name = f"Helsinki-NLP/opus-mt-{source_language_code}-{target_language_code}" if source_language_code == target_language_code: return "Translation between the same languages is not supported." try: tokenizer = MarianTokenizer.from_pretrained(model_name) model = MarianMTModel.from_pretrained(model_name) except Exception as e: return f"Failed to load model for {source_language_code} to {target_language_code}: {str(e)}" translated = model.generate(**tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)) translated_text = tokenizer.decode(translated[0], skip_special_tokens=True) return translated_text # Fonction pour traduire un fichier SRT def translate_srt(input_file_path, source_language_code, target_language_code, progress=None): subs = pysrt.open(input_file_path) translated_subs = [] for idx, sub in enumerate(subs): translated_text = translate_text(sub.text, source_language_code, target_language_code) translated_sub = pysrt.SubRipItem(index=idx+1, start=sub.start, end=sub.end, text=translated_text) translated_subs.append(translated_sub) if progress: progress((idx + 1) / len(subs)) translated_srt_path = input_file_path.replace(".srt", f"_{target_language_code}.srt") pysrt.SubRipFile(translated_subs).save(translated_srt_path) return translated_srt_path # Fonction pour transcrire l'audio d'une vidéo en texte def transcribe(audio_file_path, model_size="base"): device = "cpu" compute_type = "int8" model = WhisperModel(model_size, device=device, compute_type=compute_type) segments, _ = model.transcribe(audio_file_path) transcription_with_timestamps = [ f"[{format_timestamp(segment.start)} -> {format_timestamp(segment.end)}] {segment.text}" for segment in segments ] return "\n".join(transcription_with_timestamps) # Fonction pour ajouter des sous-titres à une vidéo def add_subtitle_to_video(input_video, subtitle_file, subtitle_language, soft_subtitle=False): video_input_stream = ffmpeg.input(input_video) subtitle_input_stream = ffmpeg.input(subtitle_file) input_video_name = os.path.splitext(os.path.basename(input_video))[0] output_video = f"/tmp/{input_video_name}_subtitled.mp4" if soft_subtitle: stream = ffmpeg.output(video_input_stream, subtitle_input_stream, output_video, **{"c": "copy", "c:s": "mov_text"}) else: stream = ffmpeg.output(video_input_stream, output_video, vf=f"subtitles={subtitle_file}") ffmpeg.run(stream, overwrite_output=True) return output_video # Initialisation de Gradio Blocks with gr.Blocks(title="Video Subtitle Creation API", description="For web use please visit [this space](https://huggingface.co/spaces/Lenylvt/VideoSubtitleCreation)") as blocks_app: with gr.Row(): video_file = gr.Video(label="Upload Video") source_language_dropdown = gr.Dropdown(choices=language_options, label="Source Language", value="en") target_language_dropdown = gr.Dropdown(choices=language_options, label="Target Language", value="en") model_size_dropdown = gr.Dropdown(choices=model_size_options, label="Model Size", value="large") # Model size dropdown transcribe_button = gr.Button("Transcribe Video") translate_button = gr.Button("Translate Subtitles") output_video = gr.Video(label="Processed Video") output_srt = gr.File(label="Subtitles File (.srt)") def transcribe_and_add_subtitles(video_file, model_size): transcription = transcribe(video_file, model_size) srt_path = text_to_srt(transcription) output_video_path = add_subtitle_to_video(video_file, srt_path, subtitle_language="eng", soft_subtitle=False) return output_video_path, srt_path def translate_subtitles_and_add_to_video(video_file, source_language_code, target_language_code, model_size): transcription = transcribe(video_file, model_size) srt_path = text_to_srt(transcription) translated_srt_path = translate_srt(srt_path, source_language_code, target_language_code) output_video_path = add_subtitle_to_video(video_file, translated_srt_path, target_language_code, soft_subtitle=False) return output_video_path, translated_srt_path transcribe_button.click(transcribe_and_add_subtitles, inputs=[video_file, model_size_dropdown], outputs=[output_video, output_srt]) translate_button.click(translate_subtitles_and_add_to_video, inputs=[video_file, source_language_dropdown, target_language_dropdown, model_size_dropdown], outputs=[output_video, output_srt]) # Lancement de l'application blocks_app.launch()