subaligner

Supported Formats

Subtitle: SubRip, TTML, WebVTT, (Advanced) SubStation Alpha, MicroDVD, MPL2, TMP, EBU STL, SAMI, SCC and SBV.

Video/Audio: MP4, WebM, Ogg, 3GP, FLV, MOV, Matroska, MPEG TS, WAV, MP3, AAC, FLAC, etc.

:information_source: Subaligner relies on file extensions as default hints to process a wide range of audiovisual or subtitle formats. It is recommended to use extensions widely acceppted by the community to ensure compatibility.

Dependant package

Required by the basic installation: FFmpeg

Install FFmpeg

$ apt-get install ffmpeg
$ brew install ffmpeg

Basic Installation

Install from PyPI

$ pip install -U pip && pip install -U setuptools wheel
$ pip install subaligner

Install from source

$ git clone [e-mail]:baxtree/subaligner.git && cd subaligner
$ pip install -U pip && pip install -U setuptools
$ pip install .

:information_source: It is highly recommended creating a virtual environment prior to installation.

Installation with Optional Packages Supporting Additional Features

Install dependencies for enabling translation and transcription

$ pip install 'subaligner[llm]'

Install dependencies for enabling forced alignment

$ pip install 'setuptools<65.0.0'
$ pip install 'subaligner[stretch]'

Install dependencies for setting up the development environment

$ pip install 'setuptools<65.0.0'
$ pip install 'subaligner[dev]'

Install all extra dependencies

$ pip install 'setuptools<65.0.0'
$ pip install 'subaligner[harmony]'

Note that subaligner[stretch], subaligner[dev] and subaligner[harmony] require eSpeak to be pre-installed:

Install eSpeak

$ apt-get install espeak libespeak1 libespeak-dev espeak-data
$ brew install espeak

Container Support

If you prefer using a containerised environment over installing everything locally:

Run subaligner with a container

$ docker run -v `pwd`:`pwd` -w `pwd` -it baxtree/subaligner bash

For Windows users, you can use Windows Subsystem for Linux (WSL) to install Subaligner. Alternatively, you can use Docker Desktop to pull and run the image. Assuming your media assets are stored under d:\media, open built-in command prompt, PowerShell, or Windows Terminal:

Run the subaligner container on Windows

docker pull baxtree/subaligner
docker run -v "/d/media":/media -w "/media" -it baxtree/subaligner bash

Usage

Single-stage alignment (high-level shift with lower latency)

$ subaligner -m single -v video.mp4 -s subtitle.srt
$ subaligner -m single -v https://example.com/video.mp4 -s https://example.com/subtitle.srt -o subtitle_aligned.srt

Dual-stage alignment (low-level shift with higher latency)

$ subaligner -m dual -v video.mp4 -s subtitle.srt
$ subaligner -m dual -v https://example.com/video.mp4 -s https://example.com/subtitle.srt -o subtitle_aligned.srt

Generate subtitles by transcribing audiovisual files

$ subaligner -m transcribe -v video.mp4 -ml eng -mr whisper -mf small -o subtitle_aligned.srt
$ subaligner -m transcribe -v video.mp4 -ml zho -mr whisper -mf medium -o subtitle_aligned.srt

Pass in a global prompt for the entire audio transcription

$ subaligner -m transcribe -v video.mp4 -ml eng -mr whisper -mf turbo -ip "your initial prompt" -o subtitle_aligned.srt 

Use the full subtitle content as a prompt

$ subaligner -m transcribe -v video.mp4 -s subtitle.srt -ml eng -mr whisper -mf turbo -o subtitle_aligned.srt

Use the previous subtitle segment as the prompt when transcribing the following segment

$ subaligner -m transcribe -v video.mp4 -s subtitle.srt --use_prior_prompting -ml eng -mr whisper -mf turbo -o subtitle_aligned.srt

(For details on the prompt crafting for transcription, please refer to Whisper prompting guide.)

Alignment on segmented plain texts (double newlines as the delimiter)

$ subaligner -m script -v video.mp4 -s subtitle.txt -o subtitle_aligned.srt
$ subaligner -m script -v https://example.com/video.mp4 -s https://example.com/subtitle.txt -o subtitle_aligned.srt

Generate JSON raw subtitle with per-word timings

$ subaligner -m transcribe -v video.mp4 -ml eng -mr whisper -mf turbo -ip "your initial prompt" --word_time_codes -o raw_subtitle.json
$ subaligner -m script -v video.mp4 -s subtitle.txt --word_time_codes -o raw_subtitle.json

Alignment on multiple subtitles against the single media file

$ subaligner -m script -v video.mp4 -s subtitle_lang_1.txt -s subtitle_lang_2.txt
$ subaligner -m script -v video.mp4 -s subtitle_lang_1.txt subtitle_lang_2.txt

Alignment on embedded subtitles

$ subaligner -m single -v video.mkv -s embedded:stream_index=0 -o subtitle_aligned.srt
$ subaligner -m dual -v video.mkv -s embedded:stream_index=0 -o subtitle_aligned.srt

Translative alignment with the ISO 639-3 language code pair (src,tgt)

$ subaligner --languages
$ subaligner -m single -v video.mp4 -s subtitle.srt -t src,tgt
$ subaligner -m dual -v video.mp4 -s subtitle.srt -t src,tgt
$ subaligner -m script -v video.mp4 -s subtitle.txt -o subtitle_aligned.srt -t src,tgt
$ subaligner -m dual -v video.mp4 -s subtitle.srt -tr helsinki-nlp -o subtitle_aligned.srt -t src,tgt
$ subaligner -m dual -v video.mp4 -s subtitle.srt -tr facebook-mbart -tf large -o subtitle_aligned.srt -t src,tgt
$ subaligner -m dual -v video.mp4 -s subtitle.srt -tr facebook-m2m100 -tf small -o subtitle_aligned.srt -t src,tgt
$ subaligner -m dual -v video.mp4 -s subtitle.srt -tr whisper -tf small -o subtitle_aligned.srt -t src,eng

Transcribe audiovisual files and generate translated subtitles

$ subaligner -m transcribe -v video.mp4 -ml src -mr whisper -mf small -tr helsinki-nlp -o subtitle_aligned.srt -t src,tgt

Shift subtitle manually by offset in seconds

$ subaligner -m shift --subtitle_path subtitle.srt -os 5.5
$ subaligner -m shift --subtitle_path subtitle.srt -os -5.5 -o subtitle_shifted.srt

Run batch alignment against directories

$ subaligner_batch -m single -vd videos/ -sd subtitles/ -od aligned_subtitles/
$ subaligner_batch -m dual -vd videos/ -sd subtitles/ -od aligned_subtitles/
$ subaligner_batch -m dual -vd videos/ -sd subtitles/ -od aligned_subtitles/ -of ttml

Run alignments with pipx

$ pipx run subaligner -m single -v video.mp4 -s subtitle.srt
$ pipx run subaligner -m dual -v video.mp4 -s subtitle.srt

Run the module as a script

$ python -m subaligner -m single -v video.mp4 -s subtitle.srt
$ python -m subaligner -m dual -v video.mp4 -s subtitle.srt

Run alignments with the docker image

$ docker pull baxtree/subaligner
$ docker run -v `pwd`:`pwd` -w `pwd` -it baxtree/subaligner subaligner -m single -v video.mp4 -s subtitle.srt
$ docker run -v `pwd`:`pwd` -w `pwd` -it baxtree/subaligner subaligner -m dual -v video.mp4 -s subtitle.srt
$ docker run -it baxtree/subaligner subaligner -m single -v https://example.com/video.mp4 -s https://example.com/subtitle.srt -o subtitle_aligned.srt
$ docker run -it baxtree/subaligner subaligner -m dual -v https://example.com/video.mp4 -s https://example.com/subtitle.srt -o subtitle_aligned.srt

The aligned subtitle will be saved at subtitle_aligned.srt. To obtain the subtitle in raw JSON format for downstream processing, replace the output file extension with .json. For details on CLIs, run subaligner -h or subaligner_batch -h, subaligner_convert -h, subaligner_train -h and subaligner_tune -h for additional utilities. subaligner_1pass and subaligner_2pass are shortcuts for running subaligner with -m single and -m dual options, respectively.

Advanced Usage

You can train a new model with your own audiovisual files and subtitle files,

Train a custom model

$ subaligner_train -vd VIDEO_DIRECTORY -sd SUBTITLE_DIRECTORY -tod TRAINING_OUTPUT_DIRECTORY

Then you can apply it to your subtitle synchronisation with the aforementioned commands. For more details on how to train and tune your own model, please refer to Subaligner Docs.

For larger media files taking longer to process, you can reconfigure various timeouts using the following:

Options for tuning timeouts

-mpt [Maximum waiting time in seconds when processing media files]
-sat [Maximum waiting time in seconds when aligning each segment]
-fet [Maximum waiting time in seconds when embedding features for training]

Anatomy

Subtitles can be out of sync with their companion audiovisual media files for a variety of causes including latency introduced by Speech-To-Text on live streams or calibration and rectification involving human intervention during post-production.

A model has been trained with synchronised video and subtitle pairs and later used for predicating shifting offsets and directions under the guidance of a dual-stage aligning approach.

First Stage (Global Alignment):

Second Stage (Parallelised Individual Alignment):

Acknowledgement

This tool wouldn't be possible without the following packages: librosa tensorflow scikit-learn pycaption pysrt pysubs2 aeneas transformers openai-whisper.

Thanks to Alan Robinson and Nigel Megitt for their invaluable feedback.