A small tracker experiment

Tracker
experiments.

Trying out audio transcription and generation on a few tracker modules. These examples compare the original replay with a generated instrumental version.

Audio examples

Examples

03 TRACKS / SEPT 2026

Switch between A (reference), B (generic brief) and C (sample analysis) at the same elapsed time. C is available for Elysium and Space Debris. Arrangements can drift; playback levels are not loudness-matched.

Loading examples…

Full-length, 192 kbps MP3 listening previews from verified FLAC masters. No gain adjustment or mastering was applied to the previews.

01 / REPLAY

Reference audio

libopenmpt renders the module using its original samples and tracker playback. This provides the reference for comparison.

02 / TRANSCRIBE

Estimated score

SheetSage2 and MERT infer melody, harmony and structure from the full replay. The resulting ABC score can contain transcription errors.

03 / GENERATE

Generated audio

YuE2 takes the score and an instrumental style brief. The sample-analysis takes use native auditions, played passages and their analysis to inform that brief. YuE2 still receives no original audio or samples.

Notes and next steps

The first Elysium and WORLD takes kept recognizable melodies, but their sounds differed from the originals. Elysium and Space Debris now have takes with more specific briefs based on sample and effect analysis. Whether they are closer still needs a listening comparison.

A few things to compare
  1. 01

    Sample character

    Compare the short pitched attacks, low pulses, noisy percussion and held tones. The new brief describes those differences; it does not identify exact presets.

  2. 02

    Controlled comparisons

    Keep the ABC and seed fixed while comparing briefs, then try a few other seeds.

  3. 03

    Effects and model descriptions

    Sample offsets, volume changes, vibrato and pitch slides shape the source sounds. Music Flamingo captions remain uncertain, so waveform and pattern evidence were used to check them. Keeping original samples may still be useful for closer likeness.