LALAL.AI Introduces Cassiopeia, a New Neural Network Enabling Superior Music Source Separation

The novel AI solution is designed to enhance the stem splitting process due to its advanced capabilities of tracking the input and output signal phase components.

Lalal.ai

Lalal.ai, an AI-powered music source separating service, presents a new neural network - Cassiopeia. The novel AI solution is designed to enhance the stem splitting process due to its advanced capabilities of tracking the input and output signal phase components. 

Unlike the previous Lalal.ai neural network Rocknet, it considers both the amplitude and the phase component when separating the stems, enabling more precise and high-quality sound results. 

"Our new audio-track splitting network was trained, tested, and proven to be superior to the leading audio separation solutions on the market," says Nikolay Pogorskiy, Lead Engineer. "Cassiopeia makes for improved splitting results with significantly fewer audio artefacts and unnatural sounds."

The Lalal.ai development team tested Cassiopeia and compared it with other splitting solutions calculating SDR (Source-to-Distortion Ratio), SIR (Source-to-Interference Ratio), and SAR (Source-to-Artefact Ratio) metrics for each track and solution. The SDR, SIR, and SAR are considered to be the objective criteria of separation quality. 

The test results show that both instrumental and vocal stems separated by Cassiopeia sound more natural and precise, thus outperforming other splitting solutions' results. 

Users can split up to three tracks for free to try Cassiopeia's performance and even compare it with the previous solution Rocknet on the Lalal.ai website

Source: LALAL.AI