See all of Mihir's current research.
Building a Perceptually Based Sound Synthesizer
Musicians and listeners often describe sounds using adjectives such as "warm" or "rich", "ethereal" or "metallic". But how do we assign these words-do they correspond to a particular timbre quality (in terms of spectrum or envelope), or do they relate to the physical method of sound generation (like plucking a string, or hitting a sheet of metal with a hammer)? Moreover how do these words correlate with our perception of sounds-could we actually create a sound from a verbal description of our intuitive expectation ("a sound that is sharp", "make it brighter?") We are designing a sound synthesis engine and interface by which descriptive words, instead of technical parameters, can be used to generate and modify sounds.
Modeling Indian Music
Synthesizing music from non-Western cultures is a problem for current music technology, which is based on constant pitch and note-oriented concepts such as MIDI. This is an issue for Indian music because time-varying pitch inflections, called gamakas, are an essential part of its construct. We are analyzing songs and instrumental pieces from the South Indian tradition, and developing software that enables musicians to synthesize Indian music with the required ornamentations. Such innovations will allow the music industry to cross cultural boundaries and provide appropriate representations for the expressive artifacts of non-Western music, in particular from Asia and the Middle-East.
Modeling Indian Percussion Instruments
"I am learning to play the Indian tabla drums in my village, but my teacher lives in a far-away town and I cannot meet him regularly. I would also like to play the tabla with my friend, but her home is two miles away from mine." Scenarios like these illustrate the need for software that "understands" rhythmic patterns in order to efficiently represent and transmit the musical signal. Playing Indian percussion instruments takes a variety of forms, such as rhythmic accompaniment, call and response, and vocalization of onomatopoeic syllables representing drum strokes. In this context we are creating a system (with sensors and software) that recognizes individual drum sounds, extracts higher-level features such as the pitch and the tempo, and identifies relevant motives like rhythmic phrases.