Automated analysis and computationally efficient synthesis of acoustic guitar strings and body

Abstract

Synthesis of stringed instrument sounds that match an acoustic prototype requires models for the string oscillation and the instrument body. Cost-effective digital filters that implement physical models of oscillating strings have been extensively studied; the current challenge is in the inexpensive replication of the spectral changes induced by the body of a physical instrument. Currently reported synthesis efforts suggest that a linear body model based on sampled impulse responses of the guitar body cannot be designed to be computationally efficient. Using an advanced IIR filter design algorithm and sampled impulse responses of the guitar body, a computationally efficient model of the guitar body is determined and can be used for synthesis of an acoustic guitar sound. Digital filters model the horizontal and vertical guitar string vibrations. A sampled plucked guitar string is analyzed to extract a linear prediction error and a Fourier series representation of the string response. A segment of the linear prediction error, which captures transient plucking effects, is used as the input to the vertical string model. The Fourier series coefficients are used to synthesize one period of the steady-state response, which is used as the input to the horizontal string model. Horizontal and vertical decay rates are analyzed from the beginning and end of the sampled string data, respectively. The guitar body model is used to provide resonance and amplify the frequencies of interest.