1How It Works & Mechanism

Measures and quantifies sound characteristics based on local CPU-powered audio acoustic and embedding models (CLAP + AST), rather than making unfounded guesses

skills/earshot/SKILL.md

earshot runs locally on CPU (audio never leaves the machine, except with the optional `llm` command) and caches every embedding and result on disk.

Executes locally offline; audio features and embedding vectors are cached in-place

Provides a complete closed-loop workflow for audio asset selection: candidate ranking, clean point trimming, semantic acoustic descriptions, and synthetic vs. real recording discrimination

skills/earshot/SKILL.md

1. `earshot rank "a soft felt mallet hit on a wooden block" downloads/sfx/ --top 5` - shortlist from a folder. 2. `earshot find winner.ogg "a single clean mallet hit" --top 3` - the best moments as sample trims (start / end in seconds).

Quantified acoustic calibration and editing assistance

Video soundtrack review and alignment dashboard: extracts audio-visual sync timelines and generates composite sheet images with spectrograms and alignment scores

skills/earshot/SKILL.md

2. `earshot sheet film.mp4 -o sheet.png --cues cues.json` (`--from 10 --to 25` zooms, `--detail low` is faster) - an annotated PNG: frames, sections, labelled events, spectrogram, loudness, beats, cue markers coloured by sync

Outputs high-fidelity inspection sheets with storyboard frames and loudness alignment

2Core Capabilities

01Audio acoustic loudness measurement and real vs. synthetic discrimination

02Asset semantic retrieval and precise transient point trimming

03Audio-visual sync timeline extraction and alignment sheet rendering

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Graded/per-use authorization, sandbox recommended

For Administrators

Approval workflow, rollback support, full audit

Evidence & Triggers (File Paths & Excerpts)

scripts/: Executes local earshot CLI commands or Python API (earshot rank/find/describe/sheet)
本地文件系统: Reads media files and writes spectrogram analysis image sheet.png alongside embedding feature caches

Audit Notes: Local execution scripts and analysis tools: Relies on the locally installed earshot CLI tool for CPU-level audio feature extraction, reads local media, and writes visualization results to disk, with no unexpected external network requests or credential access.

5Second-Pass Verification

[ok] 网络外发检测 — 默认命令音频数据绝不离开本机，除可选 llm 命令外无外部网络请求
[ok] 模型与性能声明 — CLAP 与 AST 纯 CPU 运行，首跑加载约 20-40s，内嵌多线程并发保护

6Verdict & Scenarios

Completely eliminates the industry-wide flaw where LLMs 'cannot hear sound' and hallucinate audio performance

Provides professional-grade LUFS loudness measurement and audio-visual alignment spectrogram visual artifacts

Best for: Agent developers and audio/video creators performing audio-visual sync verification for video editing, podcast scoring, and motion graphics

Not for: Pure voice interactive dialogue scenarios requiring cloud-based, real-time millisecond-level low-latency streaming

Install Agent-Ready

① Verified Mirror 2026-10-25

Option A · Direct DownloadDownload earshot.tar.gz

sha256: 3f830b68f7ebc045…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 591fa3e028.