Pitchsplitting

Pitchsplitting is a concept used in audio signal processing, music information retrieval, and digital sound synthesis to describe techniques that handle pitch information by separating or duplicating it into multiple concurrent pitch components. It can refer to both splitting a monophonic pitch into several pitch streams for synthesis and separating a complex signal into multiple pitch tracks for analysis.

In synthesis, pitchsplitting can help create richer timbres by distributing energy across adjacent pitches or subharmonics, enabling more expressive or evolving textures. In analysis, it supports polyphonic pitch tracking by producing several pitch estimates that may correspond to different voices or instruments within a recording. This dual use makes pitchsplitting relevant to tasks such as polyphonic transcription, vocal and instrument separation, and advanced pitch correction in complex musical textures.

Techniques associated with pitchsplitting include sub-band pitch estimation, where the spectrum is divided into frequency bands and a pitch is estimated per band; sub-harmonic or harmonic decomposition, which models a pitch as a combination of sub- or super-harmonics and allocates them across multiple streams; and multi-pitch estimation approaches that employ probabilistic models or neural networks to output several simultaneous pitch candidates. In some contexts, pitchsplitting may also refer to deliberately separating a single note into parallel pitch tracks to model vibrato, detuning, or other expressive phenomena.

Common challenges include pitch aliasing and octave ambiguities, rapid note changes, noisy or overlapping sources, and computational complexity. Evaluations often use metrics for the accuracy of detected pitches, such as precision, recall, and F-measure, across polyphonic situations. See also pitch detection, polyphonic transcription, and multi-pitch estimation.