Leonxlnx · Audio for agents

earshot

earshot is an open-source tool by Leonxlnx that gives coding agents the ability to listen to audio and video and to make sounds in code. It turns sound into evidence an agent can reason about: descriptions in words, how well a file fits a text description, the best moments in a long recording, whether a clip is a real recording or code-synthesised, a second-by-second listening transcript, an annotated image, and a scorecard against a profile of 7 professional launch films. Version 0.4 adds 51 code-made sound effects (the bank now holds 252) and a `launch` kit preset that mixes effects against the music bed and the cut. `earshot forge` has an agent write a sound as a physical-synthesis recipe, score renders with a critic built from earshot, and refine the recipe; `earshot fit` translates a recording into such a recipe. It runs locally on CPU as a CLI, a Python API and an MCP server, and each CLI call loads the models in about 20-40 s. Released under the MIT licence with 4,198 stars; v0.4.1 was published on 6 November 2026.