# earshot

> Lets coding agents hear, judge and synthesize sound, locally on CPU

- Repository: https://github.com/Leonxlnx/earshot
- Category: Image, video & audio
- Stars: 2,298 · Forks: 142 · Language: Python · License: MIT
- Experience: Advanced
- Page: https://ir.adyog.com/repos/leonxlnx/earshot/

Coding agents can write code that renders a film or a game, but they can't hear the result. They can't tell a cheap key click from a good one, or notice a hit that lands 80 ms late. Earshot turns audio into text, numbers and images an agent can reason about.

It uses local models on CPU plus model-free measurements such as loudness and spectrum. It can describe a file, check how well it fits a description, find the best moment in a long recording, and tell a real recording from code-made sound. It can also produce a second-by-second listening transcript of a film, an annotated image, and a scorecard against seven professional launch films.

It also lets the agent make sound. forge writes sounds as synthesis recipes and improves them using a critic's feedback. fit turns a recording into editable code. A bank of 252 code-made effects renders by id and can be timed to the picture. The README is open about limits: the search can overfit earshot's own models, and its blind tests used one listener.

## What it does

- **Describe, match and find**: describe says what a file sounds like, match scores it against a text description, and find returns the best moments inside a long recording as trims.
- **Listening transcript and scorecard**: timeline lists events, sections, loudness and sync to picture cuts. sheet makes an annotated PNG, and critique scores a launch video with timestamped fixes.
- **Forge: sounds made in code**: Recipes use physical synthesis with no recorded audio inside. A critic scores candidates, a search mutates the best, and the output includes WAVs, a report and spectrogram comparisons.
- **Fit: audio to code**: fit analyses a recording and writes a recipe module that re-renders it, with knobs for pitch, decay and brightness, and morphs between two recordings.
- **Sound bank and film kits**: sfx renders effects from a bank of 252 by id, aimed at a moment in the picture. kit renders a whole film's effects with measured sync points.

## What you could build

- Ask a coding agent to check that sound effects in a generated product demo line up with the cuts.
- Score a launch video against a profile of professional launch films and get timestamped fixes.
- Find the best key press or whoosh inside a long recording and trim it.
- Generate UI sounds like clicks, pops and whooshes in code, with no recorded audio.
- Convert a recording into an editable recipe and make variations or morphs of it.

## Who it's for

Developers building agent-made videos, demos or games who use an MCP-capable agent such as Claude Code and are comfortable with Python and basic audio terms.

## Get started

The excerpt of the README doesn't show install commands, so check the full README for setup. After installing, run earshot doctor to verify models and data, then try earshot describe on a sound file. For many calls, use the MCP server so models stay loaded.



Tags: audio, mcp, sound-design, claude-code, synthesis