Musical interpretation is a human act. The note on the page, the sound in the room and the intent behind the performance are three different things. Transkr is built around that distinction.
We model pitch symbolically rather than statistically. A parametrised instrument model describes how an instrument distributes energy across its harmonic series in terms that remain interpretable to the musician. The output is something you can read, edit and reason about.
When a parameter changes, you can trace the effect to a physical property of your instrument. That interpretability is not a secondary feature. It is the design.
Every instrument is physically unique. Every player shapes its sound through technique, touch and the particulars of their setup. This variation is not noise to be averaged out — it is the musical content.
A model should fit the instrument. Not the other way around.
We are interested in extending what musicians can work with, not in substituting for musical decision-making. The composer, arranger and performer remain the authors of the music. Transkr converts acoustic performance into symbolic form so that the musical ideas stay accessible, editable and intentional.
The musician is in the loop. That is not a constraint — it is the point.
Generative models can produce convincing audio. What they cannot reliably produce is intention. When the specific voicing, harmonic direction or melodic contour matters, symbolic representation gives the musician direct access to what was played — not a statistical approximation of what might plausibly have been played.
These are different tools for different ends. We are not building an alternative to generative audio. We are building something that makes musical content editable regardless of how it was produced.
Before a musical gesture can be interpreted harmonically or symbolically, it must first exist as physical action unfolding through time.
Transkr operates at this foundational layer: analysing instrumental behaviour, temporal structure, articulation and pitch formation directly from the audio signal. Rather than treating music as abstract note data alone, the system interprets the physical characteristics that give rise to musical events in the first place.
Within the Algoriffix ecosystem, this forms a complementary perspective to Pitch Perfekt's tonal and expressive interpretation layers.
Musical interpretation requires symbolic understanding. Tools that support that understanding respect the musician's role. Tools that obscure it do not.
No big data. No cloud processing. No copyright dependency. No black-box outputs. The model is yours, the decisions are yours and the music is yours.
These principles are applied in practice in Model Customisation and Audio-to-MIDI Conversion.
© 2020–2026 Algoriffix AB