Earshot
Leonxlnx/earshotReviewed by Radar
What it does
earshot turns audio into text and numbers an AI coding assistant can reason about: what a sound is, whether it fits a description, the best moments in a long recording, and a scorecard for a video's soundtrack. It can also create sound effects in code and ships a bank of 252 of them. It is for developers who use agents such as Claude Code, through a command line, a Python library or an MCP server. You need Python 3.10+, ffmpeg and a terminal. It runs on the processor (no graphics card) and downloads about 1.1 GB of models on first use.
How easy it is to install
C · install 1 or 2 tools · Windows
There is no installer: you install Python and ffmpeg, clone the repository, create a virtual environment and run several pip commands in a terminal.
Numbers on GitHub
- Stars
- 3,384on Oct 31, 2026
- Downloads
- 4,861900 this week
- Growth
- +470stars in 5 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Python, FFmpeg
- To build from source
- Python, Graphics card, FFmpeg
- Disk space
- 950 MB–4.3 GB · estimate
- Accounts or keys
- Google Gemini (GEMINI_API_KEY, only for the optional llm critique)
- Cost
- Free · The optional Gemini critique uses your own Google key, billed by Google.
- License
- MIT (Permissive)
- Can replace
- ElevenLabs
- Last update
- Oct 22, 2026
Risk signals (3)
- Infovery new alpha project
Created 2026-10-15 (11 days before this check); pyproject.toml classifier 'Development Status :: 3 - Alpha'.
AI review
- Infomodels downloaded on first use
README Install: 'The first model use downloads two open models (~1.1 GB)', CLAP laion/larger_clap_general and MIT/ast-finetuned-audioset, each 'at a pinned revision with safetensors weights'.
AI review
- Infooptional audio upload to gemini
README: '`llm FILE --yes-upload` | optional audio-LLM critique (Gemini), only with your own key and explicit consent' — this sends the audio file to Google.
AI review
Evidence is shown in its original language.
What Radar could not check
- Whether the install and the MCP server work on Windows; the README commands mostly use Linux/macOS paths (.venv/bin/python).
- Real speed on a typical laptop; the README says each CLI call takes about 20-40 seconds to load the models.
- Real disk use after installing torch and the models; the checks estimate 950 MB to 4.5 GB.
- How accurate its sound descriptions and soundtrack verdicts are outside the author's own tests.