gptsubtitler 0.0.2

Automatically subtitle any video spoken in any language to a language of your choice.

Subtitler

Automatically subtitle any video spoken in any language to a language of your choice.

Models used:

• OpenAI whisper - for text-to-audio
• Facebook M2M10 - for translation

Tools used:

• ffmpeg

Please don't forget to star the repository if you find it useful or educational!

Before:

https://github.com/extremq/subtitler/assets/45830561/49f6ecce-cfdc-4f1c-97eb-07a36ac841c9

After (in Romanian - model_type=medium, language_model_type=base):

https://github.com/extremq/subtitler/assets/45830561/20bc5169-0ce3-47cd-adb7-15d75daf27f4

Setup

Install the requirements.txt file.

pip install -r requirements.txt

Install ffmpeg:

# Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg

# MacOS
brew install ffmpeg

# Windows using Chocolatey https://chocolatey.org/
choco install ffmpeg

Quick guide

Example usage for adding subtitles and translating them in Romanian:

You only need to specify the language you want the subtitles to be in, the program will handle the rest of the work.

from src.transcriber import Transcriber

# I strongly recommend using the "medium" model_type.
Transcriber.transcribe("soldier.mp4", target_language="ro", model_type="medium", language_model_type="base", device="cuda")

You can also use the Translator class from translator.py if you just want to translate some text.

Example usage for translating from English to Romanian:

from src.translator import Translator

print(Translator.translate("Hi!", target_language="ro", source_language="en", device="cuda"))

If you have generated a .srt file and just want to add subtitles:

from src.video_utils import create_video_with_subtitles
create_video_with_subtitles("video.mp4", "output.srt", "video_subtitled.mp4")

Options

Device (if you have a gpu and have installed pytorch, use "cuda"):

• cpu - default
• cuda

Available options for model_type (the audio to text model):

• tiny
• base - default
• small
• medium
• large

Available options for language_model_type (the language translator model):

• base - default
• large