Timbral, Perceptual, And Statistical Attributes for Synthesized Sound
James McDermott, Niall J. L. Griffith, Michael O’Neill · 2006
A set of 40 timbral, perceptual, and statistical sound attributes is described and studied, with reference to machine learning applications and statistical experiments using a software synthesizer. The attributes include trajectory, vibrato, and statistical subsets, and subsets defined in the time, Fouriertransform, and partial domains. High correlations between some attributes are confirmed: this has application to future choice of attributes for machine learning applications. The synthesizer’s achievable attribute ranges give an indication of its relative flexibility and strengths, and the method described has application to synthesizer design.