# Light rain ambience

A diffuse rain bed and nearby drops create quiet atmosphere.

Source checkpoint: 8dc3dd143f8eb56223e62d40c694689330a94853. Original procedural synthesis by Renaissance Solutions with Codex.
Local review candidate: public distribution and reuse terms have not yet been selected.
Numbered audio, source and presets are preserved unchanged. Listening preferences and use under narration remain pending.

## Make it with your AI assistant

Ask whether your assistant can run Python and return WAV files. Give it this complete folder (or upload this ZIP and ask it to extract the files).
No account keys or Python packages are needed by the recipe. Use the supplied generator to reproduce our takes.
A prompt alone can lead to a different implementation.

Create an original light-rain atmosphere for a podcast scene. Make a
soft, diffuse wash with sparse small drops close to the listener. No
thunder, speech, music, recorded samples, or sharp dramatic splashes.

Use only Python's standard library and seeded procedural synthesis.
Make an eight-second finite bed with gentle opening and closing fades.
Do not claim it is a seamless loop. Keep separate aligned stems for
the rain bed and nearby drops. Leave room to place narration above it.

Save a 48 kHz, 24-bit stereo master, mono fold-down, both stereo stems,
and a 16-bit stereo preview. Target -6 dBFS stereo sample peak. Preserve
the code, every setting, tool versions, commands, and SHA-256 hashes.
Save this as v001; never overwrite a previously shared take.

If you can execute the recipe, generate and verify the files. If not,
provide complete local instructions and label execution as pending.
Keep human listening and podcast use separate from technical checks.

## Run locally

Unzip the complete kit and open Terminal in the folder containing this README, effects/, scripts/ and library/.
Python 3.10 or later is required; reference bytes were generated with Python 3.14.3 on macOS.
Run the following commands from that folder. Use a fresh output folder for every attempt; existing takes will not be overwritten.

python3 effects/010-light-rain/src/render.py --preset effects/010-light-rain/presets/v001.json --output tmp/010-light-rain-v001

python3 effects/010-light-rain/src/render.py --preset effects/010-light-rain/presets/v002.json --output tmp/010-light-rain-v002

Open each new preview.wav in an audio player. stereo.wav is the 48 kHz/24-bit master, mono.wav is its mono fold-down, and stems/ contains aligned layers.
Check the retained files and reproduce all WAVs with:

python3 scripts/verify_library.py --reproduce

Byte equality is verified in the recorded runtime; check other runtimes. This command does not judge how a sound feels.

## What we did

1. Make a diffuse bed from filtered noise.
2. Add 24 seeded nearby drops.
3. Fade the boundaries. This excerpt is not a seamless loop.

## Change one thing

Preserve v001. Copy its preset to v002 and change only drop_gain from
0.22 to 0.10, plus the take label. Keep the generator, seed, drop times,
pitch range, bed, panning, fades, duration, and peak target unchanged.
Render both takes separately and compare the full mixes and drop stems.
Report the shared normalization gain and actual exported RMS changes.
Do not say that the revision is better before someone has heard it.

Use the supplied source for these parameter names. With newly written code, first identify the equivalent control.
The drop setting falls from 0.22 to 0.10. Exported drops measure 6.13 dB lower; the bed rises by 0.72 dB after shared peak adjustment.
Both mixes share a sample-peak target; that does not mean they seem equally loud.

## Listen and decide

Can the rain establish the scene without pulling attention from the voice?
Put rain under an indoor scene. Lower it until every word is clear, and fade it around the passage. Compare both takes and the passage without the effect.
Record what you heard, your preference (including neither or undecided), and one next change.
Technical checks passed for the reference files; human listening and use in context remain pending.

## Source identity

The original chapter SHA-256 is eebe17378b01763077877ea187d1388f53b36bef8386b0a2bff52c60953e804c.
Each library manifest records source, preset, dependencies and WAV hashes.
The included catalog is the matching single-recipe subset of the pinned Studio catalog.