Skip to content

earshot

by Leon LinConnectorMake video or audio

Lets your coding agent hear audio and judge sound effects, music and film soundtracks

View on GitHub

The problem it solves

Your coding agent builds a film or demo but cannot hear it, so it cannot tell that a hit lands late or the music swells in the wrong place.

For Film and demo makersDevelopersSound designers

Before
You listen to every render yourself and tell the agent what is wrong with the sound.
After
The agent reads a listening transcript and a scorecard, and proposes timing and level fixes with timestamps.
Stars
★ 4.1k
License
MIT
Cost
Free
Level
Advanced
Runs
Local

Screenshots

From the project's own repository.

An annotated sheet of a launch film made by earshot, with video frames, sound events, loudness and sync markers
Image: Leon Lin (MIT). Original file

Overview

earshot gives a coding agent measured readings of sound: what a file sounds like in words, how well it fits a description, where the cleanest moment is, and how a film's sound compares with professional launch films. It runs on your CPU as a command line tool, a Python library and an MCP server, and it can also render a bank of 252 sound effects made in code. It suits makers who build films or demos with a coding agent.

It turns sound into words and numbers an agent can act on, and audio stays on your computer by default.

Try these first

Start with the first one: it is a small task you can check easily.

  1. Describe what [audio file path] sounds like and give its loudness.First try · read-only

  2. Find the cleanest single key press in [recording path] and give me the trim start and end times.read-only

  3. Critique [film path] and list the fixes to make first, with timestamps.reads video

  4. Render a short whoosh from the sound bank for a cut at 2 seconds and save it to [output path].writes files

Example prompts written by Agent Flows Hub.

Capabilities

From the creator's documentation.

ToolsetWhat your agent can doAccess
DescribeSays what a file sounds like in words, with loudness and other measurementsRead
Match and findScores how well a description fits and finds the cleanest moments in a long recordingRead
Timeline and sheetMakes a second-by-second listening transcript and an annotated picture of a filmRead + write
CritiqueScores a launch film against a profile of seven professional launch filmsRead
Sound bankRenders sound effects stored as code, aimed at a moment in the pictureRead + write
Optional Gemini critiqueSends audio to Gemini only with your own key and explicit consentRead

Good to know

You build a film or product demo with a coding agent, which cannot hear the sound it makes. earshot lets it describe a recording, find the cleanest moment in a long file and score a film's soundtrack, so it can fix timing and level problems by reading numbers.

Limits

  • Setup needs uv and ffmpeg, and the first run downloads two models of about 1.1 GB in total.
  • Each command line call loads the models again (about 20 to 40 seconds); use the MCP server for more than a few calls.

Notes by (original project name: earshot), written from the creator's documentation, not from running it ourselves. earshot: README · earshot: what it can and cannot judge

Stuck setting it up?

Tell us what happened. We take a look first; if it needs paid work, you get an estimate before anything starts.

Get help

Other connectors

  • Exa MCP Server Lets your AI agent search the web and read pages as clean text, without an account
  • Figma MCP Server Lets your AI agent turn Figma designs into code and create designs in your Figma files
  • Firecrawl MCP Server Lets your AI agent search the web and turn pages and documents into clean, readable text
  • Fireflies MCP Server Lets your AI assistant search your recorded meetings, summaries and action items
  • Fli Lets your AI assistant search Google Flights and find the cheapest dates to fly
  • Framelink MCP for Figma Lets your coding agent read Figma layout and style data to build the design you point at

See all 57 connectors · An MCP server is missing or disconnected

GitHub