A clever piece of artificial intelligence makes it possible to split the vocal and instrumental part of audio tracks. Thanks to Lalal.ai you can record your favorite songs instead of the original artist.
Step 1: Music split
In the recording studio, the audio and the instruments on different tracks are mixed together. The online platform Lalal.ai works the other way around by splitting them correctly. To separate the voice from music, a neural network was set up and trained using 20 TB of data. The result of deep learning is an algorithm that recognizes and separates instrumental and vocal tracks. The web app contains 45 million neural network parameters. This sounds impressive, but still Lalal.ai is incredibly easy to use. If you have doubts about the effectiveness, just scroll down on the homepage and listen to a sample and the result of the split. Of course you want to try this yourself with a number that is on your hard drive. Lalal.ai supports almost all audio file formats such as .mp3, .mp2, .wav, .webm, .weba, .oga, .m4a, .aiff, .au, .ogg, .aac, .ac3, .dts, .opus and .wma. You can just upload your own song and Lalal.ai will split the first 70 seconds as a trial.
Step 2: Filter strength
If you do not like the result, you can adjust the processing level. You can choose from three filter levels. In the option Mild, the minimum processing level you can usually still hear errors in the track. Level Normal applies a moderate filter. Almost all errors are then corrected, but sometimes blurring of the high frequencies can occur. Bee Aggressive the filter algorithms will react very sensitively to errors. This can lead to certain sensitive elements in the mix being removed. So this is just a matter of trying to find out which setting gives the best result.
Step 3: Download
Once the files are split, you can listen to them directly in the browser or you can download the individual files in MP3 format to use them however you like. You can try this splitter for free, but to split entire songs you have to pay. To process three tracks or a total of thirty minutes costs five dollars. Processing ten input tracks or up to ninety minutes costs ten dollars, and thirty tracks or up to five hundred minutes costs thirty dollars.
