You are here

Music signal analysis

The goal of this research is the automatic extraction of relevant cues from audio streams, to enable effective solutions of music information retrieval. This is a field on which there was an increasing interest during the last years, not only at scientific level (see for instance, the apps that were realized by Shazam and SoundHound).

For more details on research and evaluation actions conducted at international level we suggest to read the MIREX (Music Information Retrieval Evaluation eXchange) site.

In this field, the SHINE unit is active on a limited number of tasks. The main motivation for us to start some years ago a research on these issues was the synergy that can be exploited with the resources and expertise available in our unit for speech processing and recognition. As a consequence, our main interest is on the automatic processing of music sequences that include speech information (e.g., singing voice) and, in general, on techniques that can support the development of solutions regarding that goal.

Between 2008 and 2012, state-of-the-art solutions were realized for beat tracking and automatic chord detection tasks. In the latter case, during the last four years the participations to international MIREX evaluation campaigns always ranked our technology in one of the first two positions. More details on those activities can be found here. During 2013, we started to address the automatic analysis of singing voice. The final goal is to develop in a few years a first complete system for cover song identification and, in general, for an aimed automatic access to music chunks extracted from a very large database of songs.