Deep Machine Learning for Parameter Estimation in Physical Modelling
One of the most challenging tasks in physically-informed sound synthesis is the estimation of model parameters to produce a desired timbre. Automatic parameter estimation procedures have been developed in the past for some specific parameters or application scenarios but, up to now, no approach has been proved applicable to a wide variety of use cases. A general solution to parameters estimation problem is provided along this paper which is based on a supervised convolutional machine learning paradigm. The described approach can be classified as end-to-end and requires no specific knowledge of the model itself.
The proposed examples are taken from a flue pipe organ physical model [1] and from dry recordings of a Montre 8 stop.
Sound examples B3
Original contrived B3Estimated with STFT mag+phase
Estimated with MEL
Estimated with STFT, Trained with random dataset
Sound examples G5
Original contrived G5Estimated with STFT mag+phase
Estimated with MEL
Estimated with STFT, Trained with random dataset
Sound Examples - Non-Contrived
Original dry recording (B3) from a Montre 8 pipeEstimated with STFT, Trained with contrived dataset
Original dry recording (G5) from a Montre 8 pipe
Estimated with STFT, Trained with contrived dataset
Sound examples A4
Original A4 tone "Stentor" Generated A4 tone "Stentor" with STFT spectrogramOriginal A4 tone "HW-DE" Generated A4 tone "HW-DE" with STFT spectrogram
Figures
Compressed package containing spectra, attack and error plots for the B3 and G5 tones. The spectral plots show black lines and balls for te target and gray lines and crosses for the estimated tones.
figures-pipe.tar.gz
Spectra and harmonic content for two A4 tones from (a) Principal stop named ``Stentor'', and (c) Principal stop named ``HW-DE''. The gray lines and crosses show the spectrum and the harmonic peaks of s(n), while the black line and dots show the spectrum and the harmonic peaks of s^(n).