The method

This page exists for one precise reason: anyone must be able to trace a sung line back to the source that justifies it, and know exactly what I did and what a machine did.

Research comes first

Before a single line is written, I gather the village's documentation and sort it into three levels: fact when it is documented with a date and an institutional source, uncertain when sources disagree, legend when it is oral tradition. The level does not decide what goes into the song: it decides how it is said. A legend can be sung, as long as it is announced as a legend.

The language is not decoration

I write in French first, then adapt — never line by line: I start again from the meaning and rebuild the metre in the local variety. The text then goes to people who have spoken that language since childhood. For Su Votu Abarrat the corrections ran into the dozens, often contradicting one another, and each had to be arbitrated. No song is released before that review.

Where artificial intelligence fits

I use audio generation tools, and I say so. They produce material: takes, timbres, motifs. What comes after is not automatic — selection, editing, stem and MIDI extraction, re-instrumentation, arrangement, mixing and mastering in Ableton Live. The compositional, lyrical and arrangement decisions are mine, and they are documented: every song has a statement of intent written before any generation, and a log recording the takes I discarded and why.

Why I write this here

Because a village of five hundred people has the right to know who is singing it, and on what basis. And because the record of this work is the only thing that separates a work from content.