# How to synthesize a new sound in code and refine it by ear with earshot

Repo: [Leonxlnx/earshot](https://github.com/Leonxlnx/earshot) | Difficulty: Advanced | Time: 1 hour | Basis: Seen in the wild

Let the agent write a sound recipe, then have a critic score and improve it against real references.

**Who it is for:** Sound-curious developer who wants a custom click, whoosh or key press made entirely in code

**The problem:** Stock effects never quite fit, and an agent cannot judge a synthesized sound without hearing it.

**The outcome:** A folder of rendered WAVs, a report in the critic's words, spectrogram comparisons and an audition file.

## Steps

1. Run `earshot forge --list` to see recipes like click, impact, keypress and whoosh.
2. Run `earshot forge keypress --refs path/to/real/keypresses/ --budget 144 --variants 4 --dir made/keypress` with your own reference folder.
3. Open report.md and compare.png in the output folder and listen to audition.wav.
4. Re-render a pick with `earshot forge --render keypress --params made/keypress/report.json --seeds 1 2 3 --dir made/again`.

## Prompt to paste into your agent

```text
Use earshot forge to make four variants of a <sound type> against my reference folder <refs dir>, then read report.md and tell me which pick is closest to the references. Verify by listening to audition.wav, not just the scores.
```

## Caveat

The search optimizes earshot's own models and can fool them (the README notes this), so check spectrograms and listen; the v1 blind test was a coin flip on one listener and 15 trials.

## Seen in the wild

- X (Oct 2026): https://x.com/i/web/status/2112336745945788688
- X (Oct 2026): https://x.com/i/web/status/2111927730933575857

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/leonxlnx-earshot/synthesize-new-sound-with-forge/