In this presentation I will go over some of the research I have been involved with in my thirty-year career within the field of Sound and Music Computing, emphasizing the goals I aimed for and identifying some of the results obtained.

 

My personal research, and the one I have directly supervised, has been mainly focused on the analysis, description and synthesis of sound and music signals. My initial efforts were dedicated to analyze and transform complex and musically relevant sounds; sounds that were not well captured by the audio processing techniques used at that time. My approach was to use spectral analysis and synthesis techniques to develop a deterministic plus stochastic model with which to obtain sonically and musically meaningful parameterizations and descriptions. That work had practical applications for synthesizing and transforming a wide variety of sounds, including the voice.

 

As a natural evolution of that initial research I became interested in going from single sounds to collections of sounds, thus being able to describe and model the relationships between sound entities. To tackle that I had to incorporate methodologies coming from disciplines such as machine learning and semantic technologies in order to complement the signal processing approaches I was using. A major bottleneck for carrying out that research was the availability of large and adequate audio collections. To solve that we developed a platform for collecting and sharing sounds and then using the collected sounds we carried out research on the automatic description of sound signals. We now have large publicly available sound collections and we have developed information retrieval tools of relevance for several sound and music applications.

 

In the last few years, and in the context of the music information retrieval field, it has become evident that to make sense of sound and music data we need to incorporate domain knowledge. In order to improve the distance measures used in exploration and retrieval tasks we need to incorporate the knowledge that exist around sound and music. To that aim, most of my current projects focus on the study of specific sound and music repertories and on targeting well-defined tasks, trying to formalize and represent user knowledge with which to develop applications of relevance to those users. We are putting together coherent corpora; we are working on audio analysis techniques specific for these corpora and chosen tasks; we are gathering and analyzing user and contextual information to train our data models; and we are developing task-oriented tools to interact with particular sound and music collections. Our results show the benefit of this type of information processing research in which bottom up and top down approaches are combined.

 

My research has always been motivated by music, by the interest of developing musical tools that can be socially and culturally relevant. In this talk I want to emphasize this aspect while talking about my thirty-year research journey.