Iterative Partial Tracking in Inharmonic Audio Signals; Expanding the YIN Algorithm for Real-Time Monitoring and Treatment of Drum Harmonics

Abstract

Time-domain subtraction of partials allows for lossless amplitude scaling of an in put signal’s tonal components, without affecting the phase. A method for real-time partial tracking in inharmonic transient audio signals, such as drum recordings, was developed by adapting the YIN algorithm [1] and combining it with time-domain subtraction of synthesised sinusoids. The algorithm can be applied block-wise to any tonal audio signal with analysis blocks as short as two times the period of the fundamental frequency of the signal, and step sizes shorter than the typical looka head of a DAW. This allows for real-time application of the algorithm in a DAW, detecting fundamental frequencies down to at least 60 Hz. The study features an overview of three methods for partial- and fundamental fre quency estimation: lag domain methods, spectral methods, and transient modelling synthesis. Based on the overview, a new method called Iterative Partial Tracking (IPT) was formulated, in order to track partials via fundamental frequency detec tion alone. A YIN-based IPT implementation was adapted to the requirements set by DAW applicability, and tested on a set of floor tom recordings. The results were found to be promising: up to four partials were estimated in a typical floor tom with a gross pitch error (GPE) of 0%. The fundamental frequency of a floor tom was, in some recordings, estimated with errors that never exceeded one semitone. The accuracy of higher-order partial estimations could be improved by applying shorter, frequency based, analysis frame lengths, and the validation procedure of frequency estimations is subject to further improvements. But in its foundation, the lag-domain approach to tracking partials in inharmonic audio signals seems like a promising method, possible to implement in real-time applications where tonal atonal signal separation is required.