Deposit export—publishing without publishing audio

Raw recordings from private spaces often cannot be shared; their acoustic envelope can. ambiscape deposit writes per-take TSVs of 1 Hz features:

Time    level_dbfs    centroid_hz    low_frac    high_frac
0       -39.8         881            0.786       0.189
1       -44.9         135            0.966       0.007

These files are considered safe for open deposits (Zenodo and similar) where the WAVs are withheld, since a 1 Hz loudness/spectral envelope is far below speech timescales and carries no intelligible content.

The export follows the schema of the StillStanding365 deposit (365 daily standstill sessions, fourMs/RITMO), which makes corpora that use it directly poolable. Method deltas against that deposit's original extractor are documented in the deposit module (W-channel at native rate vs an 8 kHz four-channel downmix; power vs magnitude band fractions): trends and dynamics are directly comparable, absolute fraction values differ slightly by construction.

Directional data and channel conventions

When depositing directional products, record which B-format convention the source files used (see Sessions & conventions). A convention mismatch produces azimuth distributions collapsed onto one axis, which is an artefact that survives into downstream correlations and is invisible unless you know to look for it.

What still requires raw audio: fast-level descriptors (Leq/LAeq), events at the 0.25 s criterion, diffuseness, elevation, spectra beyond the centroid, psychoacoustic indicators. Plan deposits accordingly: features for everyone, raw audio under controlled access where the ethics allow.

Publishing an excerpt: the Freesound category

A feature deposit is one way material leaves a session; a curated audio excerpt bound for a public repository is the other. Freesound has required a category from its Broad Sound Taxonomy on every upload since April 2025, and the category is also a search facet, so it decides whether anyone finds the file. Recording it beside the WAV keeps the choice reproducible across a pack of excerpts instead of retyping it into an upload form:

ambiscape background my-session/ --excerpt 600 --bst ss-i \
    --tags "ambisonic,room-tone,ventilation" --licence "CC BY 4.0"

This writes <excerpt>.wav.freesound.json next to the excerpt, carrying the category, its full name, the licence, the tags, and the excerpt's own provenance (session, wall clock, duration, selection method).

A whole-room recording is a soundscape, so --bst accepts only the ss-* categories unless you pass --any-bst:

code	use it for
ss-i	interiors: rooms, offices, halls, anywhere closed
ss-u	outdoors with human intervention: streets, stations, transit halls
ss-n	natural habitats, and interiors an open window has handed over to one
ss-s	synthesised or computer-made ambiences

Run the privacy gate before uploading and record its result in the same file:

ambiscape speechgate my-session/analysis/<excerpt>.wav

deposit.freesound_sidecar(..., speech_fraction=...) stores the fraction and sets a privacy_review flag above 0.01. It is stored rather than enforced, because what counts as an acceptable fraction is a judgement about the recording and not a property of the format.

This taxonomy classifies the file, not the sound

The Broad Sound Taxonomy cuts first by artefact type — music, instrument samples, speech, sound effects, soundscapes — which is a retrieval question, not an acoustic one. It is deliberately kept out of ambiscape.taxonomy, where Schaeffer's, Schafer's and soundscape ecology's schemes each ask a question of the same sound. For a session recorded in a room the answer here is almost always ss-i, and a field with one value is no use as a descriptor.