On January 20th 2012 and near Delhi we had the first workshop of the CompMusic project. It was organized as a satellite event of FRSM 2012 and it was an opportunity to bring together the two Indian CompMusic teams and to present the project to the Indian research community. FRSM and the workshop took place at the College of Engineering of the Kamrah International Institute of Technology (KIIT) in Gurgaon, India. KIIT is a private engineering College that despite not being active in music research was very generous to host our workshop.
FRSM is an annual event with 20 years of history. It focuses on speech and music and has a strong emphasis on young researchers. The main person behind the conference is Prof. Ashoke Datta, a very senior and active researcher that has been collaborating with ITC-SRA and doing research in Indian music for a long time.
The CompMusic workshop program and the slides of the presentations are available here. We also videotaped the talks and put them on our YouTube channel, however the quality is not so good. We will try to do better next time.
As part of FRSM I gave a keynote speech on CompMusic (video of the talk). That was a good opportunity to introduce the workshop, which took place right after FRSM, and present an updated version of the overall vision and research aims of the project.
In my opening talk of the workshop (video of the talk) I covered some of the initial research efforts in the project. We are working on the computational modeling within five main topics: intonation, melody, rhythm, on-line communities, and music discovery tools. These topics have to be understood within each particular music culture and each one requires a specific research approach. We have realized that most of the definitions of relevant music concepts, like melody and rhythm, found in the literature have a western bias and we are trying to redefine them within our particular cultures and for our research goals. In the talk I also showed a mockup and a very preliminary version of the CompMusic Browser, a tool for exploring the music collections of the different repertoires. The Browser's intended functionality is quite indicative of our research objectives and it will be used as the major demonstrator of our technological results. We intend to put the browser online as soon as it offers some useful functionality, which should be before the next workshop that we are organizing in Istanbul in the month of July.
Preeti Rao, who leads the research team at IIT-Bombay, gave a talk on "Computational modeling for Hindustani music" (video of the talk). She talked about some of the tools that can be used to analyze Hindustani music and then she described the use of pitch contours to understand the melodic characteristics of the music. She presented her team's work on melodic transcription.
Hema Murthy, who leads the research team at IIT-Madras, gave a talk on "Carnatic Music: Signal Processing Perspective" (video of the talk). She started by introducing the structure of a Carnatic music concert and then focused on the issue of ptich analysis of Carnatic music. She presented her team's work on the analysis and synthesis of pitch contours, specifically on Gamakaas. She also introduced their first results on time-frequency representations aimed as separating the voice from the accompaniment instruments, work that could be used to better analyze the gamakaas and phrases sung by the lead voice.
Pandit Buddhadev Das Gupta, a well recognized Hindustani musician and sarod maestro, gave a talk about how technology has been used in Hindustani music (video of the talk). He talked about electronic shruti boxes, synthesizers and proposed the use of computers for education, generating melodies to be learned or for slowing down audio recordings so details of a performance are better perceived.
T. M. Krishna, one of the leading Carnatic music vocalists, gave a talk on Carnatic music from a historial perspective, focusing on melodic issues (video of the talk). He pointed out that the indian music of the early times, around 2nd and 3rd centrury AC, did not use a fixed pitch (fixed tonic), instead, the common practice was to use fixed frequencies for the notes, which is the inverse of what is used now. From this view he argued that the use of 22 shrutis as a way to divide an octave is not relevant today, that it is a concept of the past. He also emphasized that a note is not a specific frequency position, instead it is a frequency region. Also he made the point that the original ragas were collections of motives, or melodic phrases and that it was later when new ragas started to be defined as collections of notes, scales. This is why we currently find both phrase-based and scale-based ragas. Another topic that he covered was the issue of gamakas and their importance in defining ragas and he also talked about the different compositional forms used in Carnatic music. He finished his presentation by talking about the treatise "Sangita Sampradaya Pradarshini" by Subbarama Dikshitar, published in Telugu in 1904. He has been involved in the rendering of the compositions notated there following the descriptions included in the treatise, thus following to the performance practice of the time.
Bariş Bozkurt, who leads the team in Istanbul working on Turkish-makam music, gave a talk on "Distribution Based Computational Analysis of Makam Music" (video of the talk). His talk was mainly an introduction to the makam music of Turkey and to the work that his team has done on the analysis of that music. He made special emphasis on the issues of notation and tuning and on his signal processing work to automatically describe intonation and to recognize the makam used in a given piece. This work is quite relevant for a number of the issues that the study of indian music is faced with.
The last talk of the workshop was given by Joan Serrà, a post-doc researcher working in Barcelona and collaborating with the project. His talk was entitled "Machine Learning for Music Discovery" (video of the talk) and he gave a quick overview of the field of machine learning as it is used for music description, currently applied to western music. He made a lot of emphasis on how to best use of the different machine learning methods in specific problems. Most of the presented methods and approaches should be of relevance to indian music.
The workshop ended with a panel discussion with the participation of Preeti Rao, Hema Murthy, Bariş Bozkurt, T. M. Krishna, and Mallika Banerjee (a hindustani vocalist from Delhi) (video of the panel: part 1, part 2). The aim of the panel was to talk about melodic and rhythmic characteristics of both Hindustani and Carnatic music, trying to focus in the differences between the two cultures. It was clear that is very difficult to formalize the characteristics of such rich musical cultures; there are many styles, approaches to improvisation and many factors that influence a given musical performance.
The workshop was very succesful. We were able to have a fruitful dialog between engineers and musicians. We will definitely continue in that direction.