The researchers used Band-in-a-Box automation scripts (often archived in formats like .rar) to generate the Groove2Groove MIDI Dataset . They transformed simple ABC music files into high-quality BIAB styles to create a massive amount of "synthetic" musical training data.
The paper explores how AI can take a "groove" from one song and apply it to another in "one-shot"—meaning it learns the style almost instantly without needing thousands of examples of that specific song.
The code used to process these BIAB files is publicly available on GitHub (cifkao/pybiab) , allowing others to automate the rendering of BIAB files into MIDI or other formats. Context of BIAB ".rar" Files
The phrase commonly refers to compressed archive files containing song libraries or software for Band-in-a-Box (BIAB) , an intelligent automatic accompaniment tool . While "rar" files themselves are just containers, they are frequently used by the BIAB community to share large MIDI and "RealTrack" datasets for music production.
An interesting academic paper that utilizes these types of BIAB files for advanced music research is:
IEEE/ACM Transactions on Audio, Speech, and Language Processing Why it is interesting:
These archives typically contain .SGU (song files without melody) or .MGU (song files with melody).