Earshot

Earshot is an open-source tool by GitHub user Leon Lin (Leonxlnx) that comes as a Skill, a command-line tool, a Python API and an MCP server. In the README's words, it turns sound into evidence a coding agent can reason about. Give it a local audio or video file and it can describe the sounds, match and score existing sound effects, line them up on a timeline, or synthesise new sounds in code, which suits sound for animation, product videos and sound design. A separate real-sound sub-Skill looks for material on sound-effect sites such as Freesound and Mixkit. Analysis runs on your own computer by default, using two audio models, CLAP and AST. The README also says it doesn't do speech recognition and can't tell you a lyric's language, whether a clip is copyrighted, or whether it sounds good.

Who it's for:People making animation or video with tools like Claude Code or Codex who want AI help choosing and timing sound effects. Readers who don't write code won't get much from it.

Before you start: You need Python 3.10 or later and ffmpeg, and you install it with pip. The first run downloads two models from Hugging Face, about 1.1 GB in total. You set up MCP yourself. By default it only reads files on your computer and has no microphone recording. The optional earshot llm feature sends audio to Google only after you add your own Gemini key and explicitly agree to the upload; we suggest turning it on only when you need it. The real-sound sub-Skill downloads effects from the web, so check the licence of each clip. The code is MIT-licensed; the two models are Apache-2.0 and BSD-3-Clause.

Based on the project's public documentation. Not tested by AI Inspiration World.