Toward an estimation of the clarinet reed pulse from instrument performance

In this work, a technique is presented for estimating the reed pulse from the pressure signal recorded at the bell of a clarinet during performance. The reed pulse is a term given to the typically periodic sequence of bore input pressure pulses, a signal related to the volume flow through a vibrating reed by the characteristic impedance of the aperture to the bore. The problem is similar to extracting glottal pulse sequence from recorded speech; however, because the glottis and instrument reeds have very different masses and opening areas, the source-filter model used in speech processing is not applicable. Here, the reed instrument is modeled as a pressure-controlled valve coupled to a bi-directional waveguide, with the output pressure approximated as a linear time invariant transformation of the product of reed volume flow and the characteristic impedance of the bore. By noting that pressure waves will make two round trips from the mouthpiece to the bell and back for each reed pulse, yielding a distinct positive and negative lobe in the running autocorrelation period of the recorded signal, the round-trip attenuation experienced by pressure waves in the instrument is estimated and used to invert the implied waveguide, producing reed pulse estimates.