Filter design for synthesis of musical notes: A multidimensional feature-based approach

We present a fast and efficient method for feature-based synthesis of musical sounds that are perceptually close to an actual instrument, by designing filters that spectrally shape a chosen excitation signal. We have proposed a novel, multi-dimensional set of features for the Indian bamboo flute which has been used for the filter design. The filter design approach aims at providing a generic framework that can be used to obtain filters of appropriate orders to synthesize musical sounds of melodic instruments characterized by continuity in sound production, e.g., the bowed and woodwind classes of instruments. The designed filters are used for real time synthesis along with a harmonically rich excitation that can be generated with low complexity. The proposed approach of modeling and filter design is scalable in complexity and quality.