Conditioned Source Separation for Music Instrument Performances
This work addresses music source separation for variable instrument sets, offering a domain-specific improvement that is incremental by integrating multimodal data.
The paper tackled the challenge of separating multiple musical instruments in mixtures where sources vary in number and share timbral characteristics, by proposing a method that uses conditioning with instrument presence and video data, achieving improved separation quality as indicated by concrete metrics like SDR gains.
In music source separation, the number of sources may vary for each piece and some of the sources may belong to the same family of instruments, thus sharing timbral characteristics and making the sources more correlated. This leads to additional challenges in the source separation problem. This paper proposes a source separation method for multiple musical instruments sounding simultaneously and explores how much additional information apart from the audio stream can lift the quality of source separation. We explore conditioning techniques at different levels of a primary source separation network and utilize two extra modalities of data, namely presence or absence of instruments in the mixture, and the corresponding video stream data.