Bertsobot, an engineered humanoid, can sing and categorize music

You have seen the mighty Jarvis of Ironman, one of the most intelligent systems (on Marvel earth) that can, in fact, save you from a disaster. But what if, you get to see a humanoid is singing at the top of its mechanical lungs? Robotics and Autonomous Systems (RSAIT) research group at the UPV/EHU's Faculty of Informatics are interested in a unique line up of research projects among which significant development of interaction between humans and machines or robots is of primary importance. According to Basilio Sierra, the central idea of "BertsoBot Project" is to make robots sing verses. To reach the ultimate goal the team has improved a bunch of sections including voice signal comprehension, computer vision, browsing, generating of new musical melodies etc.

The automatic generation of music and the automatic classification of music are the two areas where Izaro Goienetxea, a researcher and Ph.D. student is involved. According to her, many of the contemporary statistical models to automatically generate music the abstract melodic levels are missing where the coherence of the melodies is needed. Although her work started with the ways of generating melodies, later on, she has considered the possibilities of classifying music. With a gigantic storage of music, it is now an important aspect to automatically sort the music as per their correct genre.


The published research talks about two new methods, among which the first one deals with classifying music according to its genre. It basically clusters similar bertso tunes in a single stack. Not just that, these clusters are further used to automatically generate new tunes following the style of the genre. Plus including more music model also creates different versions.

In short, at first, the research team collected a group of bertso tunes and analyzed their method, on it. Unlike other ways of categorizing music, this one is robust and can work on numerous datasets. The method can also be used in different applications including music therapy applications, musical composition learning or composing programs. The research has been published in the PLOS ONE journal.

Source: The analysis of bertso tunes has resulted in new automatic methods for generating and classifying music - UPV/EHU