Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import moviepy.editor as mp | |
| import librosa | |
| import numpy as np | |
| from transformers import pipeline | |
| # Load Whisper model for speech-to-text | |
| asr = pipeline("automatic-speech-recognition", model="openai/whisper-large") | |
| # MarianMT or M2M100 for translation (multi-language) | |
| translator = pipeline("translation", model="facebook/m2m100_418M") | |
| # Supported languages with their codes | |
| languages = { | |
| "Persian (fa)": "fa", | |
| "French (fr)": "fr", | |
| "Spanish (es)": "es", | |
| "German (de)": "de", | |
| "Chinese (zh)": "zh", | |
| "Arabic (ar)": "ar", | |
| "Hindi (hi)": "hi", | |
| "Russian (ru)": "ru" | |
| } | |
| def generate_subtitles(video_file, language_name): | |
| try: | |
| # Extract the target language code from the selected language name | |
| target_language = languages[language_name] | |
| # Check if video_file is a file object or a file path string | |
| if isinstance(video_file, str): | |
| video_path = video_file # It's a file path | |
| else: | |
| video_path = video_file.name # It's a file object | |
| print(f"Processing video from path: {video_path}") | |
| # Extract audio from video using moviepy | |
| video = mp.VideoFileClip(video_path) | |
| audio_path = "temp_audio.wav" | |
| audio = video.audio | |
| audio.write_audiofile(audio_path, codec='pcm_s16le') | |
| print("Starting speech-to-text transcription") | |
| # Load the audio file as a waveform using librosa | |
| waveform, sr = librosa.load(audio_path, sr=16000) # sr=16000 for Whisper | |
| # Process audio in chunks | |
| chunk_duration = 30 # seconds | |
| chunk_size = sr * chunk_duration # number of samples per chunk | |
| transcriptions = [] | |
| for i in range(0, len(waveform), chunk_size): | |
| chunk = waveform[i:i + chunk_size] | |
| if len(chunk) == 0: | |
| break # Avoid processing empty chunks | |
| # Pass the chunk to Whisper's ASR model | |
| transcription = asr(chunk)["text"] | |
| transcriptions.append(transcription) | |
| # Combine all transcriptions into a single string | |
| full_transcription = " ".join(transcriptions) | |
| print("Starting translation") | |
| # Translate transcription to the target language using M2M100 | |
| translation_pipeline = pipeline('translation', model='facebook/m2m100_418M') | |
| translated_subtitles = translation_pipeline( | |
| full_transcription, | |
| forced_bos_token_id=translation_pipeline.tokenizer.get_lang_id(target_language) | |
| )[0]["translation_text"] | |
| # Return subtitles | |
| subtitles = f"Original: {full_transcription}\nTranslated: {translated_subtitles}" | |
| return subtitles | |
| except Exception as e: | |
| # Catch and log the error | |
| print(f"Error occurred: {e}") | |
| return f"Error occurred: {e}" | |
| # Define Gradio interface | |
| def subtitle_video(video_file, language_name): | |
| try: | |
| # Handle both file-like objects and file paths | |
| return generate_subtitles(video_file, language_name) | |
| except Exception as e: | |
| print(f"Error in processing video: {e}") | |
| return f"Error in processing video: {e}" | |
| # Gradio app layout | |
| interface = gr.Interface( | |
| fn=subtitle_video, | |
| inputs=[ | |
| gr.Video(label="Upload Video"), | |
| gr.Dropdown( # Dropdown for language selection | |
| label="Choose Target Language", | |
| choices=list(languages.keys()), # Display language names in the dropdown | |
| value="Persian (fa)" # Default language | |
| ) | |
| ], | |
| outputs="text", | |
| title="Automatic Video Subtitler & Translator" | |
| ) | |
| interface.launch() | |