We are in a mission to advance music technology

Innovation in technology knows no borders. It is a worldwide endeavor to address the intricate challenges of the music industry landscape. Our scientists and engineers embody this ethos with their diverse backgrounds and unique perspectives. We are a living proof that the collective creativity of a diverse team can compose symphonies of innovation.

The team

Join the researchers behind our mission

We're pioneering the future of music-tech with cutting-edge research. Be a part of our mission to empower creative potential through technology.

Datasets

MoisesDB

MoisesDB is a comprehensive multitrack dataset for source separation beyond 4-stems, comprising 240 previously unreleased songs by 47 artists spanning twelve high-level genres. The total duration of the dataset is 14 hours, 24 minutes and 46 seconds, with an average recording length of 3:36 seconds. MoisesDB is offered free of charge for non-commercial research use only and includes baseline performance results for two publicly available source separation methods.

SDXDB23_LabelNoise

The purpose of this dataset is to provide the research community with a set of songs that can be used to design and evaluate source separation system under robust separation settings (i.e., when the ground truth data contains errors and inconsistencies). This dataset contains simulated errors regarding the identity of the musical instruments included in each track: we call these errors Label Noise.

SDXDB23_Bleeding

The purpose of this dataset is to provide the research community with a set of songs that can be used to design and evaluate source separation system under robust separation settings (i.e., when the ground truth data contains errors and inconsistencies). This dataset contains simulated errors for which each source is also present in the recording of all others: we call this phenomenon Bleeding.