Model Customisation

Transkr interprets pitch through a parametrised model of your instrument rather than matching patterns against a large training dataset. The model describes how your instrument distributes energy across its harmonic series. You configure it; the algorithm uses it to interpret incoming audio.

This page explains the acoustic phenomena behind each parameter and how to build a model that works reliably for your specific instrument and playing style.

• Understanding the Instrument Model
• Harmonic Structure
• Tone & Equalisation
• Ghost Notes & Octave Doubling
• Interference & Resonance
• Sensitivity
• Building a Custom Preset

Understanding the Instrument Model

No two instruments sound alike. The acoustic character of a guitar is shaped by its body, woods, strings, pickup position and the player's technique — all of which affect the harmonic content of every note. This variation is too specific to capture through a generalised training dataset and too musically significant to ignore.

Rather than approximating a universal pitch detector, Transkr gives you a model you define for your instrument. The algorithm attempts to explain the sound it hears in terms of that model. A well-calibrated model makes pitch detection more accurate, more consistent and less prone to artefacts specific to your instrument's acoustics.

The same approach also means the system is interpretable. Each parameter has a direct physical correlate. When detection behaviour changes, you can trace it to a specific acoustic property rather than treating the result as a black box.

Harmonic Structure

When a string vibrates, it produces a fundamental frequency accompanied by a series of overtones — harmonics — at integer multiples of that frequency. The relative strength of these harmonics is one of the primary acoustic signatures of an instrument. A classical guitar, an electric guitar played clean and the same guitar played through distortion each have a distinct harmonic profile.

The Harmonics parameter controls how prominently the instrument model represents these overtones. For most string instruments, the optimal value falls between 1.1 and 1.7. The correct value for your instrument is the lowest one at which all expected pitches are detected cleanly across the full playable range.

Setting Harmonics too low: overtones are strong enough to register as independent pitch events — extra notes appearing above the played pitch, particularly in the mid and upper range.

Setting Harmonics too high: the model under-represents overtones, causing notes in the lower register to absorb energy that should resolve to notes in the higher register — missed upper pitches in chords are a common symptom.

For instruments with an uneven overtone profile, you can adjust the strength of individual harmonics by holding the mouse button and moving the cursor up or down over specific harmonic positions in the Harmonics control.

Model Tuner

Model Tuner automates the Harmonics calibration described above. Click Record, then play a single note and let it ring out. Lower-pitched notes are recommended because they contain more harmonics and decay more slowly, giving a better estimate. Once about a second of clean signal is captured, click Tune to re-derive the Harmonics value from the recorded signal rather than searching for it by ear.

Tone & Equalisation

Before pitch analysis, the signal passes through a 3-band leveller followed by a 3-band parametric equaliser. These tools serve the analysis as much as they serve the sound.

The Leveller stabilises harmonic energy across the bass, mid and treble ranges independently. This is useful when the instrument has an uneven frequency response, strong room coloration or when certain harmonic registers behave unpredictably under different playing dynamics.

The Equaliser allows finer reshaping of the input signal: boosting a range where detection is sparse, or attenuating resonant peaks that produce false pitch events. The leveller precedes the equaliser in the signal chain, and all time-domain pre-processing takes place before the signal enters the pitch detection stage.

To view the frequency response of each filter individually, click on the Bode diagram and hold the mouse button.

Ghost Notes & Octave Doubling

Two categories of artefact appear most commonly during preset development.

Ghost notes are unintended pitch events produced by picking noise, fret-hand movement and wave interference between harmonics. They typically appear as brief, low-velocity note events that do not correspond to any deliberately played pitch. The Melody parameter attenuates these. Note that slightly increasing the Harmonics value can also reduce ghost notes when playing single-note passages, because a stronger instrument model leaves less ambiguity for interference to exploit.

Octave doubling occurs when the second harmonic is strong enough to register as a separate note one octave above the fundamental. It tends to appear in instruments where the second harmonic is relatively prominent, particularly in certain pickup configurations or playing positions. The Unison parameter attenuates this.

Interference & Resonance

Chords present a harder problem than single notes, and understanding why helps set realistic expectations for what the model can achieve.

When multiple strings ring simultaneously, their harmonics interact. Constructive interference amplifies certain frequencies; destructive interference suppresses others. The resulting signal differs from what a simple sum of individual note spectra would predict. In instruments with a resonant body — acoustic guitars in particular — the body cavity reinforces specific frequencies regardless of which notes are being played, adding acoustic energy that the analysis stage must account for.

Transkr reports what the instrument produces acoustically, not an idealised interpretation of what was intended. A chord played on a heavily resonant acoustic guitar may produce more detected events than the number of strings pressed, because the instrument itself is acoustically active in ways that notation does not represent. This is the system behaving correctly.

Sensitivity

The Sensitivity parameter controls the onset detection threshold: how strong a transient must be before the algorithm registers a new note event.

Lower sensitivity reduces false positives from quiet or ambient signals and stabilises detection in noisy environments. Higher sensitivity catches quieter onsets more reliably but also responds more readily to unintended transients — fret noise, pick attack on adjacent strings, or mechanical vibration.

The Loudness and Dynamic Range parameters affect MIDI note velocity only — not the pitch detection model. Loudness sets the maximum velocity. A Dynamic Range of zero produces fixed-velocity notes equal to the Loudness value.

Building a Custom Preset

A recommended workflow for calibrating a model to a new instrument:

1. Select your instrument from the instrument menu. This loads the default preset for that type as the starting point.
2. Set Melody, Unison and Legato to zero.
3. Play each open string separately. Adjust Harmonics until each string produces a single clean note event.
4. Play basic chords. Confirm that the expected pitches appear without excessive additional events.
5. Set Harmonics to the lowest value that still achieves clean detection. Lower values reduce the risk of low-register notes masking high-register ones.
6. Use the EQ to address consistent gaps or peaks in the detected frequency range.
7. Reintroduce Melody, Unison and Legato incrementally to reduce artefacts.
8. Adjust Sensitivity if notes are consistently missed or triggered unintentionally.
9. Save the result — Save state in the standalone application, Save preset as in a DAW.

For instruments with a notably different chord and single-note response, consider maintaining two presets: one optimised for melody, one for chords.