- Who it helps
- Creative developers building sound effects or checking audio alongside generated video.
- Availability
- A public repository documents CPU execution and local tools. Optional Gemini critique uploads audio with explicit consent and a user-provided key. Some reference recordings require your own licensed files.
Measurements and judgments
The measure command reports signal properties without models. Description and critique tools use learned models or reference profiles. Keeping those outputs separate helps you distinguish a clipping measurement from an opinion about how a sound fits a scene.
Where the experiment is useful
A practical evaluation could compare three versions of your own button-click sound. Keep volume and playback conditions fixed, listen blind, then compare the tool’s ranking with your preference. Save the original recordings and settings so another person can repeat the comparison.
What you can do
- Start by reading the command table, model setup and reference-data requirements. Choose one audio file you have permission to analyze and define a listening check before trying automated revisions. Open reference ↗
Where the evidence stops
No audio model or generated sound was executed here. Model scores are imperfect proxies for listener judgment; the repository says its newer synthesis version still lacks a published blind-test score.
Sources read. Primary material was read on October 18, 2026. The external software, account workflow and reported results were not independently executed.
Sources & further reading
GitHub · implementation
Earshot: local audio analysis and synthesis for agents ↗
Selected for its direct evidence about this subject, its access conditions or a specific practical limitation.
Leonxlnx · page checked 2026-10-18
Coverage note: Primary evidence is available. Independent demonstrations and current cross-platform discussion are incomplete in this bounded sample.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
Earshot exposes audio descriptions, measurements and comparison tools to coding agents through a command line, Python and MCP. It also supports generating and revising sounds written as code.