Violin Acoustic Signal Analysis

Advisor: Dr. Jeffrey Krolik

Despite research dating back to the famous 19th century physicist Helmholtz, remarkably little is known about what physically produces a beautiful violin sound. Helmholtz’s theory of the bowed string posits that the string “sticks” and “slips” on the rosined bow hair producing a sawtooth-shaped excitation. This project is aimed at estimating the deviations from Helmholtz’s theoretical sawtooth excitation which are thought to make the difference between an artificial versus a more natural violin sound.

Through a collaboration with string maker J. D’Addario & Company, the features of a bowed excitation waveform necessary to produce a realistic violin sound were estimated using blind multichannel system identification techniques and a violin bridge driver developed by Norman Pickering. Modeling the violin as an unknown linear-time-invariant system, a set of recordings of the violin being naturally bowed and stimulated by the bridge driver with known waveforms were created with the aim of developing a least squares estimate for the naturally bowed excitation signal.

Rather than setting up the bridge driver at Duke, D’Addario agreed to run the experiments at their laboratory and then send the signal data to Duke for processing. Frequency sweeps of the violin response and stepped inputs using sawtooth waves were made at the D’Addario laboratory. From the sweeps, sets of known input and output reference waveforms were created in order to test the least squares estimation algorithm. Waveform estimates were fairly accurate indicating that this estimation technique could be useful for further human bowing tests. The next step of the project would be to obtain both artificially stimulated and naturally bowed recordings of the violin in a controlled configuration that would allow least squares estimation of the naturally bowed waveforms. Ultimately, an estimate for the naturally bowed excitation signal could be used as input to the bridge driver to create a realistic violin sound.

The full final report is available here in Adobe Acrobat Reader (pdf) format.