Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from faster_whisper import WhisperModel | |
| import logging | |
| from transformers import MarianMTModel, MarianTokenizer | |
| import pandas as pd | |
| import requests | |
| # Configure logging for debugging purposes | |
| logging.basicConfig() | |
| logging.getLogger("faster_whisper").setLevel(logging.DEBUG) | |
| # Fetch and parse language options from the provided URL | |
| 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() | |
| # Prepare language options for the dropdown | |
| language_options = [(row['ISO 639-1'], f"{row['ISO 639-1']}") for index, row in df.iterrows()] | |
| def transcribe_and_optionally_translate(audio_file, source_language, target_language, model_size, change_transcript): | |
| # Transcription | |
| device = "cpu" # Use "cuda" for GPU | |
| compute_type = "int8" # Use "float16" or "int8" for GPU, "int8" for CPU | |
| model = WhisperModel(model_size, device=device, compute_type=compute_type) | |
| segments, _ = model.transcribe(audio_file) | |
| transcription = " ".join([segment.text for segment in segments]) | |
| if change_transcript: | |
| # Assume user will modify the transcript manually before translation | |
| return transcription, True | |
| # Translation | |
| if source_language != target_language: | |
| model_name = f"Helsinki-NLP/opus-mt-{source_language}-{target_language}" | |
| tokenizer = MarianTokenizer.from_pretrained(model_name) | |
| model = MarianMTModel.from_pretrained(model_name) | |
| translated = model.generate(**tokenizer(transcription, return_tensors="pt", padding=True, truncation=True, max_length=512)) | |
| transcription = tokenizer.decode(translated[0], skip_special_tokens=True) | |
| return transcription, False | |
| def add_hard_subtitle_to_video(input_video, transcript): | |
| """Add hard subtitles to video.""" | |
| temp_subtitle_path = '/tmp/subtitle.srt' | |
| with open(temp_subtitle_path, 'w', encoding='utf-8') as file: | |
| file.write(transcript) # Assuming transcript is in SRT format | |
| output_video_path = f"/tmp/output_video.mp4" | |
| ffmpeg.input(input_video).output(output_video_path, vf=f"subtitles={temp_subtitle_path}").run(quiet=True) | |
| return output_video_path | |
| def process_video(video, source_language, target_language, model_size='base', change_transcript=False, modified_transcript=None): | |
| audio_file = video # Directly use the video file as the audio input | |
| transcript, can_modify = transcribe_and_optionally_translate(audio_file, source_language, target_language, model_size, change_transcript) | |
| if can_modify and modified_transcript: | |
| # Use the modified transcript for translation if allowed and provided | |
| transcript = modified_transcript | |
| # Perform translation here if necessary (similar to the previous step) | |
| output_video = add_hard_subtitle_to_video(video, transcript) | |
| return output_video | |
| # Setup the Gradio app | |
| app = gr.Interface( | |
| fn=process_video, | |
| inputs=[ | |
| gr.Video(label="Upload Video"), | |
| gr.Dropdown(choices=language_options, label="Source Language"), | |
| gr.Dropdown(choices=language_options, label="Target Language"), | |
| gr.Dropdown(choices=["base", "small", "medium", "large", "large-v2", "large-v3"], label="Model Size"), | |
| gr.Checkbox(label="Change Transcript before Translation?", value=False), | |
| gr.TextArea(label="Modified Transcript (if allowed)") | |
| ], | |
| outputs=gr.Text(label="Transcript"), | |
| title="Video Transcription and Translation Tool", | |
| description="Transcribe or translate your video content. Optionally, edit the transcription before adding hard subtitles." | |
| ) | |
| if __name__ == "__main__": | |
| app.launch() | |