Two Albums from One Python Studio
Sign-Off and The Quiet Hours came out of the same renderer and sound nothing alike, which was the whole idea. Sign-Off is a worn late-night broadcast — the last hour before the transmitter gives up. The Quiet Hours is piano-led night music with strings that only turn up when they’re needed. Shared engine, separate records.
What the studio keeps of each one is the reasoning: scores, performed audio, mix stems and delivery records, saved at every stage so any decision can be found again and argued with.
Write the identity into the score
Sign-Off varies its chord banks, motifs, section lengths and production settings from track to track. Rabbit Ears and Vertical Hold carry the heaviest warble; elsewhere the damage backs off and leaves room for something softer. Its drums stay on a steady final-tempo grid while the pitched material goes through slowdown and tape movement, so the beat holds its shape while everything around it sags.
The Quiet Hours varies piano patterns, register, meter and phrase length instead. A smooth performance timeline moves both ends of every note, and voice IDs keep same-pitch overlaps correctly paired. The piano stays in front throughout — no equally loud accompaniment elbowing it out of the way in every passage.
Prepare sources with records attached
The piano is Salamander Grand Piano; the strings are VSCO Community Edition. The preparation script downloads pinned source revisions, keeps their licenses, and builds the zone maps the renderer plays. It adjusts piano onsets and root tuning, and extends selected string sustains with crossfades so they can hold a long note without drifting. Source and attribution details are on the listening page.
Sign-Off uses synthesized pitched sources and three separately supplied LM-2 drum samples. The public repository holds the code and the list of what’s required — not copies of the studio’s installed plugins or sample libraries.
Save a boundary before each expensive change
Every session contains score data, dry performed buses, processed stems, a float mix and a manifest. A room-balance change reuses the dry audio and takes seconds. A note, tempo or instrument change needs another source render and takes as long as it takes. Code and asset hashes make it possible to ask which existing renders still match their inputs, which matters more than it sounds like when you come back to a record after a month away.
Processed float stems reconstruct the float mix. The nonlinear shared music bus stays a group stem, because processing each input separately is simply a different mix and pretending otherwise would be a lie told in WAV format. Final delivery applies the chosen gain and exports PCM WAV and MP3, both measured after encoding rather than before.
Listen as a record
The album players show the artwork and the complete track lists. Pick a song, or let the sequence run. Each player moves to the next track on its own and stops at the end, and starting one pauses the other.
The technical checks confirm that the expected files were produced — nothing more than that. Whether the sequence works is a question about transitions, density, how often a phrase repeats before it wears out, and whether the last track earns the ending. That part only answers to listening. The album chapter goes into the choices in detail, and the repository has the runnable scores and renderer if you’d rather read the source.
To explore the system without tracking down external assets, start with Afterimage, Open Window or Night Transit. They’re compact sketches built from synthesized tones and noise, and they go through exactly the same stages.