Three wooden knocks

Three close impacts suggest someone just outside the scene.

Arrivals, interruptions, and an offscreen presence · 1.65 seconds

1 Hear it

Start with the whole sound. What action does it suggest before you read how it was made?

The first take

Hear the separate layers

A layer, also called a stem, is one ingredient saved on its own. These share a timeline; keep their leading silence when combining them.

wood body

contact tick

2 Make it with your AI

Ask your assistant: “Can you run Python and return WAV files here?” If it can, use the starter prompt below and ask it to return the audio, source and settings. Download the files it makes.

If it only returns text, ask for the complete script and follow the local steps below. A written response is not an audio file.

Read or copy the starter prompt

Prompt text — select and copy, or use the buttons below

Create three original wooden knocks for someone arriving at a door in a
podcast. Make the knocks dry, close, short, and slightly unequal in
strength. Use a gentle contact followed by irregular, rapidly damped
body resonances. Avoid a long bell-like ring or a sharp snare-like burst.

Use Python's standard library only. Do not use a recording, stock sample,
voice, or paid audio service. Place the three hits at 0.25, 0.63, and
1.09 seconds in a 1.65-second file. Keep room sound out of the master.

Save separate aligned body and contact stems, a 48 kHz/24-bit stereo
master and mono fold-down, and a 16-bit stereo preview. Leave -6 dBFS
stereo sample-peak headroom. Preserve code, seed, complete settings,
commands, versions, measurements, and hashes as v001.

Generate the files if you have execution tools. Otherwise provide full
local steps and say that generation remains pending. Do not claim the
knocks sound like wood or work in the scene until someone listens.

Your assistant may produce a different implementation. Use our kit for the reference sound.

Download the complete recipe kit

What we actually did

Our local coding assistant wrote and ran Python on a Mac. It used tones and repeatable noise, without an audio-generation service.

Make a resonant body and a separate contact tick.
Place impacts at 0.250, 0.630 and 1.090 seconds.
Keep the rhythm fixed while changing body decay.

Reproduce our sound: files and local steps

A fresh prompt can produce a different implementation. To reproduce these takes, give your assistant the complete kit and ask it to use the supplied generator and presets. If it cannot open ZIP files, extract the kit first and provide its files.

Includes both takes, separate layers, source, settings and verification tools. You do not need access to our repository.

Or run it yourself: extract the ZIP, open Terminal in the folder containing README.txt, effects/, scripts/ and library/. Use Python 3.10 or later. No Python packages or API keys are needed.

python3 effects/011-wooden-knocks/src/render.py --preset effects/011-wooden-knocks/presets/v001.json --output tmp/011-wooden-knocks-v001

python3 effects/011-wooden-knocks/src/render.py --preset effects/011-wooden-knocks/presets/v002.json --output tmp/011-wooden-knocks-v002

Open each new preview.wav in an audio player. Keep a new output directory for every attempt; the generator refuses to overwrite previous takes.

python3 scripts/verify_library.py --reproduce

The verification command checks the retained files and regenerates both takes. Exact reproduction was tested on Python 3.14.3/macOS; another runtime needs its own check.

Download the written walkthrough

3 Change one thing

Does faster damping suggest the right door, or remove too much weight?

We tested a specific change: shorter body decay. The first take stays available so you can compare.

A — first take

Download WAV

B — Shorter body decay

Download WAV

What changed in the files

Damping rises from 80 to 120 per second. The later body contains less energy; all three start times stay fixed.

This is a measured result. No listening preference has been recorded for these takes. Equal peak levels do not guarantee equal perceived loudness.

Read or copy the revision prompt

Prompt text — select and copy, or use the buttons below

Keep v001. Create v002 by changing only damping_per_second from 80 to
120 and the take label. Keep the same source, seed, body frequencies,
contact noise, hit times and strengths, panning, duration, and peak target.
Render a separate take. Compare full mixes and isolated body stems.
Measure late body energy as well as overall RMS, and explain the shared
gain adjustment. Leave the material and preference judgments pending.

Your assistant may produce a different implementation. Use our kit for the reference sound.

This prompt names controls in our supplied source. If your assistant wrote a different generator, ask it to identify the matching control first.

Inspect settings, masters and provenance

damping_per_second: 80 → 120. The take label also changes; all other sound settings remain fixed.

v001

1.65 seconds · stereo sample peak -6 dBFS · Listening decision pending

Stereo master · Mono master · Source and audio checksums

{
  "effect_id": "011-wooden-knocks",
  "take": "v001",
  "sample_rate": 48000,
  "peak_dbfs": -6,
  "character": "three-dry-wooden-knocks",
  "duration_seconds": 1.65,
  "seed": 11011,
  "first_hit_seconds": 0.25,
  "interval_one_seconds": 0.38,
  "interval_two_seconds": 0.46,
  "body_seconds": 0.22,
  "damping_per_second": 80,
  "body_gain": 0.82,
  "contact_gain": 0.07,
  "pitch_hz": 290,
  "stereo_position": -0.12
}

v002

1.65 seconds · stereo sample peak -6 dBFS · Listening decision pending

Stereo master · Mono master · Source and audio checksums

{
  "effect_id": "011-wooden-knocks",
  "take": "v002",
  "sample_rate": 48000,
  "peak_dbfs": -6,
  "character": "three-dry-wooden-knocks",
  "duration_seconds": 1.65,
  "seed": 11011,
  "first_hit_seconds": 0.25,
  "interval_one_seconds": 0.38,
  "interval_two_seconds": 0.46,
  "body_seconds": 0.22,
  "damping_per_second": 120,
  "body_gain": 0.82,
  "contact_gain": 0.07,
  "pitch_hz": 290,
  "stereo_position": -0.12
}

The 48 kHz/24-bit masters and layers are retained unchanged; previews use 16-bit PCM. Source checkpoint: 8dc3dd143f8eb56223e62d40c694689330a94853.

4 Decide what works

Put the knocks before a character reacts. Decide what kind of door each take suggests.

Compare both takes at comfortable playback levels, first alone and then beside narration.
Try the passage without the effect. Keep the version that serves the moment; “neither” is a useful answer.
Describe what you heard before asking for one next change. Preserve your earlier takes.

Keep listening notes

Optional notes saved in this browser. Copy them into your AI assistant when you want to discuss the next take.

What did you hear?Your choice and whyOne next change

The fields become editable after saved notes load.

Take your saved audio into Story Studio. Opening the page does not transfer your files or notes.