Mbira dzavadzimu music is better represented as tuning-specific, embodied cyclic counterpoint than as a melody with accompaniment. A player repeats an already-polyphonic hand pattern, another player may interlock a related part, hosho and voices articulate further layers, and listeners can follow melodic lines that no single hand performs. A piano roll can show events, but it does not settle which of those simultaneous structures is the piece.
The common cycle in Paul Berliner and Cosmas Magaya’s documented repertory is 16 ternary beats, or 48 pulses: four large segments, each containing four beats of three pulses. That is a strong default, not a universal meter. Berliner’s account also includes 36-pulse ternary and 32-pulse duple pieces. MBIRA’s schema therefore stores an arbitrary pulse count and pulses per beat even though much of the present source set uses 48.
A cycle does not require one absolute beginning
Garfias divides Nyamaropa, Mandarindare, and Nhemamusasa into four 12-pulse sections while describing how players may enter at different points. Synchronization and section order matter, but the first printed column is not automatically a universal bar one. Erica Azim makes the practical consequence vivid: a student who identifies a tune only by the teacher’s first entry point may fail to recognize the same cycle when it starts elsewhere.
Rotation is therefore a first-class parameter. Software may display pulse one for orientation, store a source entry point, or align a hosho transcription, but it should not turn that convenience into a claim that every listener and performer locates the downbeat there. Comparing cycles under rotation can reveal that two printed harmonic sequences differ by starting segment rather than substance.
Three hearings coexist
Martin Scherzinger’s analysis distinguishes ways of parsing the same texture. The distinctions matter for transcription because a line heard in the aggregate may cross between performers or hands. Assigning that line back to one source stream would produce a neat score and a false motor pattern.
| Projection | What the listener follows | What software must not infer |
|---|---|---|
| Kinesthetic or source pattern | One performer’s hand and motor sequence | That the stream is the only melody being heard |
| Inherent pattern | Register-near notes regrouped across source streams | That any one performer executes the resulting line |
| Resultant pattern | The complete interwoven sounding surface | That aggregate order reveals fingering or part ownership |
These hearings can imply different accents and metric centers. Scherzinger shows registrally formed lines and the total resultant supporting multiple rotations of binary and ternary groupings. The point is not that meter disappears. Hosho, harmonic rhythm, bass motion, articulation, entry conventions, and embodied patterns all provide orientation. The point is that the sounding lattice affords several coherent organizations at once.
Variation changes relationships, not random notes
Berliner and Magaya describe a composition as a family of complete repeatable parts and variations. Kushaura commonly carries an identifying ground. Kutsinhira follows and interweaves, but it is not defined as a mechanical one-pulse copy. Conventional patterns can shift, rotate, change role, or take distinct shapes. The relationship to the composite matters more than a single transformation formula.
Grupe’s comparisons suggest a useful hierarchy. Small additions, omissions, and harmonically equivalent substitutions disturb the whole least. Larger changes reshape a note pair or passage, then one hand while the other remains stable, and only later the complete motor-rhythmic pattern. Abrupt movement between distant versions is possible, but gradual establishment and reshaping across cycle boundaries is characteristic of the performances he analyzes.
Azim supplies the acceptance condition that a note-only generator misses. Two variations taught as valid in isolation may sound wrong when combined because one interrupts the melodic thread a player is following. A model that checks only whether each hand contains legal pitches can still destroy the aggregate. This is why MBIRA’s current algorithm admits only whole aligned phrase cells from a project-registered sibling and labels even that narrow operation as research.
Tuning belongs to a particular instrument
Named tunings help people search and orient themselves, but they do not fully determine pitch. Corresponding parts may differ by more than a fifth in absolute register across documented instruments. Keys that function as corresponding degrees on different manuals can form stretched or compressed near-octaves and near-unisons. Buzz, resonator, overtone treatment, forging, register layout, and decay contribute to what players describe as tuning.
Andrew Tracey’s historical report gives Hugh Tracey’s octave-normalized average as a near-equiheptatonic sequence: cumulative positions around 0, 162, 343, 516, 684, 852, 1,030, and 1,200 cents. That average summarizes multiple instruments. It is not a measurement of the instrument used for a particular tune, and it should not be installed as a universal mbira scale. The MBIRA tuning guide likewise emphasizes relationships within a specific instrument and variation among makers and musicians.
Ryan Pratt’s interval study compares documented mbira tunings while resisting the urge to squeeze their unequal and sometimes octave-evading relationships into simpler ratios. Acoustic measurements by Laurie McNeil and Sorin Mitran find strong overtones around five and fourteen times a key’s fundamental. A useful digital tuning record therefore needs measured key frequencies, uncertainty, instrument provenance, and an explicit playback projection; a seven-note mode label alone is insufficient.
What survives a Western pitch projection
A fixed piano, fretted guitar, or ordinary General MIDI synth cannot reproduce an undocumented instrument. Snapping keys to 12-tone equal temperament can preserve pulse order, cyclic phase, register assignment, hand distribution, approximate contour, and broad dyad motion. It changes unequal scale steps, manual-specific octave relations, spectral identity, and sometimes the harmonic pull a listener perceives.
- For fixed Western instruments, preserve cycle topology, phase, register, hand streams, and relative motion; label the result a structural adaptation.
- For MIDI 1.0, map each physical key separately and retain its cents error. Pitch bend is safe only when channel allocation makes the intended detuning unambiguous.
- For MPE, samplers, or microtuned synths, use a documented key-frequency map and preserve spectra, decay, resonator, and buzz when the source actually supplies them.
- For exact transposition of a measured instrument, multiply every key frequency by one ratio instead of snapping the result to a named scale.
The software model follows from these limits
MBIRA’s canonical events always keep cyclic pulse and part identity before applying a view-only semitone offset. They keep physical-key identity when the source supplies it; otherwise the field remains explicitly null. A tuning snapshot can map each key to a measured frequency and a separate MIDI projection. Source reductions that contain only Western pitches stay labeled as reductions. Catalog records without measured frequencies stay labeled as editorial 12-tone playback. The player never back-fills a plausible tuning and presents it as observed.
This representation does not solve mbira analysis. It preserves enough distinctions to avoid foreclosing it. Versioned annotation schemas now represent cross-stream selected focus lines, tune-relative harmonic areas, measured instruments, reviewed variation examples, and contiguous multi-cycle performances without rewriting canonical events. Until those schemas contain source-specific evidence, the analyzer exposes candidates and unavailable perspectives rather than filling gaps with Western inference. The result remains a structural adaptation with provenance—not a digital substitute for an instrument, a teacher, or a performance.