AutoLyrixAlign Service

Accurate word-level alignment of lyrics to polyphonic audio

This is a service API wrapper around chitralekha18/AutoLyrixAlign. It also includes dependency checks and mostly automates the setup process. It basically makes the whole thing super easy-to-use!

All the real work is done by the system developed at chitralekha18/AutoLyrixAlign, though. Amazing work going on over there that's honestly over my head. I just wanted to use it as a foolproof API service.

Requirements

Linux. On Windows, you can probably use WSL2 (though I haven't tested that). I'm not sure about MacOS. If you're not on Linux, look into installing singularity on your OS. Maybe try this?
NodeJS. I've only tested on v16, but I imagine you can probably go as low as v10.
Singularity. If it's not installed, you'll get instructions on how to install it when you run the server. It's used to run the containerized software from chitralekha18/AutoLyrixAlign in a reproducible way. Only singularity v2.5.2 has been tested.
RAM. chitralekha18/AutoLyrixAlign recommends having 20GB of RAM. I've found that it uses closer to 13GB. Your mileage may vary!
Disk Space. You'll need about 16GB free to automatically download and extract the necessary data files. After the initial setup, about a 13GB footprint will remain used.

Installation

Clone this repository. git clone git@github.com:gazugafan/AutoLyrixAlignService.git
Go into the project folder you just cloned. cd AutoLyrixAlignService
Install JS dependencies. npm install
Run the server. node index.js

*We'll check for the required dependencies when you run the server. If you're missing singularity, we'll help you install it. We'll also help download the necessary 13GB of data.

Usage

With the server running, you should be able to open localhost:3000 in your browser (adjust the port and domain for however you set things up). This will bring you to a simple page where you can test out the API. Select a file, enter lyrics, and submit the form. You should see some simple logs output in the server console, and after a few minutes you should get the results back in the browser!

To use the service programatically, just send a POST request to /align the same way the form does. Be sure to set the Content-Type header to multipart/form-data. The POST parameters are...

audio_file The polyphonic audio file to align with. Can be MP3, WAV, etc.
lyrics The lyrics to align. Can include any sort of newline characters you like, special characters, song part identification lines (like [Chorus]), backup lines (like (woo) or *breathes*), etc. Anything you throw at it should work.
format What format you'd like the results returned in. Can be set to raw or json (defaults to json). raw gives you the results directly from the alignment process. json massages those results into an array of lines and words that should match back up to the original lyrics supplied.

You can also get the current version with a GET request to /version.

Server Options

You can specify the following command-line options at the end of node index.js...

Options:
      --version                Show version number                     [boolean]
  -p, --port                   The port to listen on             [default: 3000]
  -d, --skip-dependency-check  Skips dependency checks (lowers startup time)
                                                                       [boolean]
  -c, --concurrency            The max number of alignment processes to run at
                               the same time                        [default: 1]
      --debug                  Outputs more info, including the alignment
                               command output                          [boolean]
  -h, --help                   Show help                               [boolean]