Computational pilot / 2026
From Birdsong
to Music.
Change the sound. Examine the model.
- Role
- Architecture, code & research
- Status
- Author manuscript
- Edition
- 13.10.2026
A question made concrete.
A blackbird phrase becomes synthetic, flute and piano renderings. This computational pilot examines how the model responds to changes in sound and to the reference used for comparison.
This pilot begins with one 2.45-second blackbird phrase. Its tracked frequencies become a transcription and sample-derived instrumental renderings. Comparing the model’s responses under different references reveals how the reference itself shapes the account of a transformation.
A single-phrase computational pilot. No human or neural measurements were collected. The renderings are transformations of one source, not independent participants.
The sound sources.
The instrument endpoints are sample-derived resyntheses. They retain the transcription’s G6–F8 register, including notes outside ordinary instrumental ranges.
The 3D view always shows the original recording. Its upper edges follow eight tracked frequency curves; depth represents linear RMS amplitude. The filled faces connect each trace to the baseline for visibility, and are not additional frequencies or harmonics. RMS is signal amplitude, not perceived loudness.
Blackbird: Oona Räisänen (2007), Turdus merula ↗. Public domain. A 2.45-second excerpt.
Flute: Versilian Studios / Sam Gossner, VSCO 2 Community Edition ↗, CC0. Pitch shifted, attack aligned and gain normalized.
Piano: Alexander Holm, Salamander Grand Piano v3 ↗, CC BY 3.0 ↗. Velocity layer 8; excerpts attack aligned, gain normalized, resampled, tuned to measured playback roots and pitch shifted for the phrase.