SOUND LEVELS

The levelled sound catalogue

Every sound RISE offers (the atmospheres, the music and the tones in SOUND_GROUPS, src/audio/sound-list.js) plays at one loudness: −29 dBFS RMS, ±2 dB, measured at the output of the real audio engine at its default levels (soundscape layer 0.85, master 0.7). Before SND-001 the catalogue spanned 42 dB, from the music at about −53 dBFS to the Gateway tone at about −11.5 dBFS, so choosing a sound also chose a volume.

Each sound reaches the band by a trim in dB, SOUND_TRIM_DB in src/audio/sound-levels.js. The engine multiplies a layer's gain by the trim of the sound it is playing, so a reader's own volume and the voice ducking work from the trimmed level. Sounds that are not offered (a personal bed, a chant bed, the parked Feelings sounds) have no trim.

The manifest

levels.json is the last measurement:

• measuredAt: when it was taken.
• settings: the engine's levels, its fade-in, the sample rate, and the window measured: 25 seconds from 3 seconds after the start, past the 2-second fade-in.
• band: the target and its tolerance, the same as LEVEL_BAND in sound-levels.js.
• sounds: for each offered sound, its trimDb and the measured rmsDbfs and peakDbfs.

src/audio/sound-levels.test.js holds the manifest to the code: every offered sound has a trim and nothing else does, each entry was measured with the trim the code ships, and each sits inside the band. A sound that changes level fails the build once it is measured again.

Measuring again

After changing a sound, or the engine's gain stages:

node scripts/measure-sound-levels.mjs            # every sound; writes levels.json
node scripts/measure-sound-levels.mjs piano jazz # only these; prints, writes nothing

The script starts a Vite dev server, opens headless Chromium with Playwright, starts each sound the way the Chamber does on a fresh engine, and records the engine's last node through an AudioWorklet in real time, three sounds at once (about two and a half minutes for all sixteen). It prints a table with the trim each sound would need to sit at the centre of the band. Copy those trims into SOUND_TRIM_DB, measure again, and commit the manifest once every sound is in band. It is a development tool and does not run in CI.

---

Published from docs/evals/sound/SOUND-LEVELS.md. Edits made here are overwritten on the next push to main.