[!NOTE] This is a fork of kaegi/alass.
ilass is a command line tool to synchronize subtitles to movies.
It can automatically correct
- constant offsets
- splits due to advertisement breaks, directors cut, ...
- different framerates
The alignment process is not only fast and accurate, but also language-agnostic. This means you can align subtitles to movies in different languages.
ilass stands for "Automatic Language-Agnostic Subtitle Synchronization". The theory and algorithms
are documented in my bachelor's thesis
and summarized in my bachelor's presentation.
Executable for Windows (64-bit)
Get the latest executable from here! Just download and extract the archive. The file ilass.bat is the command line tool.
Executable for Linux (64-bit)
Get the latest executable from here! To run the executable, ffmpeg and
ffprobe have to be installed.
You can change their paths with the environment variables
ILASS_FFMPEG_PATH (default ffmpeg) and ILASS_FFPROBE_PATH (default ffprobe).
Usage
The most basic command is:
You can also use ilass to align the incorrect subtitle to a different subtitle:
You can additionally adjust how much the algorithm tries to avoid introducing or removing a break:
# split-penalty is a value between 0 and 1000 (default 7)
Values between 5 and 20 are the most useful. Anything above 20 misses some important splits and anything below 5 introduces many unnecessary splits.
If you only want to shift the subtitle, without introducing splits, you can use --no-splits:
# synchronizing the subtitles in this mode is very fast
Currently supported are .srt, .ssa/.ass and .idx files. Every common video format is supported for the reference file.
Performance and Results
The extraction of the audio from a video takes about 10 to 20 seconds. Computing the alignment usually takes between 5 and 10 seconds.
The alignment is usually perfect -
the percentage of "good subtitles" is about 88% to 98%, depending on how strict you classify a "good subtitle".
Downloading random subtitles
from OpenSubtitles.org had an error rate of about 50%
(sample size N=118).
Of all subtitle lines (not subtitle files) in the tested database,
after synchronization
- 50% were within 50ms of target position
- 80% were within 100ms of target position
- 90% were within 400ms of target position
- 95% were within 800ms of target position
compared to a (possibly not perfect) reference subtitle.
How to compile the binary
Install Rust and Cargo then run:
# this will create the lastest release in ~/.cargo/bin/ilass
The voice-activity module this project uses is written in C. Therefore a C compiler (gcc or clang) is needed to compile this project.
To use ilass with video files, ffmpeg and ffprobe have to be installed. It is used to extract the raw audio data. You can set the paths used by ilass using the environment variables ILASS_FFMPEG_PATH (default ffmpeg) and ILASS_FFPROBE_PATH (default ffprobe).
Building from Source
If you want to build and run the project from source code:
Configuration
All parameters are shown for cargo build can also be used for cargo install and cargo run.
FFmpeg as a library
You can also link ffmpeg as a dynamic library during compile time. The library implementation can extract the audio about 2 to 3 seconds faster. Unfortunately it is harder to compile, the error handling is only very basic and might still have bugs.
You have to remove "# FFMPEG-LIB" from every line that starts with it in ilass-cli/Cargo.toml. Then use:
# Important: you have to be inside `ilass-cli`! Otherwise the parameters get ignored.
Alias Setup
For Linux users: It is recommended to add the folder path to your system path. Add this to your ~/.bashrc (or the setup file of your favorite shell):
Folder structure
This cargo workspace contains two projects:
-
ilasswhich provides the algorithmIt is targeted at developers who want to use the same algorithm in their project.
-
ilass-cliwhich is the official command line toolIt is target at end users who want to correct their subtitles.
Library Documentation
Open README from ilass.
Notes
This program was called aligner in the past. This made it nearly impossible to find on a search engine, so ilass was chosen instead.