Back in 2020, LALAl.AI was a fledgling project that quickly rose to prominence and made a name for itself in the music source separation game. In 2021, LALAL.AI has got a rich feathering of features, spread its wings with confidence and conquered incredible heights.

It was truly an outstanding and special year in the life of the project and our team is looking forward to 2022 to realize more ideas and take LALAL.AI to the next level. While the new year is drawing near, let’s take a look at everything we've done and accomplished in 2021.

6 New Supported Stems

Throughout the year we’ve been gradually implementing the possibility to extract various musical instruments from audio and video files. In addition to vocals and accompaniment, our service became capable of detecting the sound of drums, bass, acoustic guitar, electric guitar, piano, synthesizer, and accurately extracting each of the instruments as stems.

The First 8-Stem Splitter Status

After the sixth musical instrument, the synthesizer has been added to the list of our officially supported stems, LALAL.AI has become the world’s first AI-powered 8-stem splitting service. We’ve launched our campaign on ProductHunt as LALAL.AI 2.0 with great success and many positive reviews from the community.

Users & Usage Growth

As LALAL.AI was growing, the audience was exponentially growing as well. This year we’ve seen an influx of users trying our service, 476,213 new users registered to utilize the features in full. In 2021 LALAL.AI did 2,443,049 previews, 917,506 full splits, processed 85,694,460 seconds of audio - that's 23,804 hours!

2 Neural Network Versions

Initially LALAL.AI isolated stems with Rocknet, a unique neural network we created using 20TB of training data for extraction of instrumentals and voice tracks. In search of ways to improve the vocal and accompaniment isolation techniques and expand to support individual musical instruments, our team developed a new neural network, Cassiopeia. It was a significant leap forward in terms of the stem separation process and the quality of results - you can read the detailed report here.

In the first half of 2021, LALAL.AI users were able to choose between the networks before uploading the file. Later, we’ve removed the neural network switch and all files started to get processed only through Cassiopeia as it was superior to Rocknet in all aspects and for all stem isolation tasks.

2 Business Solution Options

The music source separation algorithm of LALAL.AI has become available for integration into any service, application or website to allow businesses to provide their clients with high-quality stem isolation.

We started to offer the utilization of our infrastructure and enable businesses to use their own servers and AWS instances for the solution deployment. Our business solution is available for purchase in two ways - a monthly subscription (for regular large data volume processing) and a pay-as-you-go model (for quick and hassle-free processing of a specific number of minutes).

7 New Languages Support

Previously available exclusively in English, in 2021 LALAL.AI became more accessible to audiences from around the world. The service was translated into 7 additional languages - French, German, Italian, Spanish, Korean, Japanese, and Chinese (both traditional and simplified versions).

The departing year has been fruitful and amazing, the LALAL.AI team is looking back on it with joy, gratitude and a sense of accomplishment. The upcoming 2022 will be a year of new possibilities and achievements, both for us and our users!