Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/ZJU-REAL/Easelnpx agentmods add skills/zju-real/easel/audio-denoiseWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/zju-real/easel/audio-denoise)<a href="https://agentmods.dev/skills/zju-real/easel/audio-denoise"><img src="https://agentmods.dev/badge/skills/zju-real/easel/audio-denoise.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00143 | $0.01552 |
| Opus 5 | $0.00072 | $0.00776 |
| Sonnet 5 | $0.00029 | $0.00310 |
| Haiku 4.5 | $0.00014 | $0.00155 |
Grade A, and why
audio-denoise scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -L https://github.com/GregorR/rnnoise-models/raw/master/somnolent-hogwash-2018-09-01/sh.rnnn \ How it starts
The opening of the file, as written. The whole thing — 117 lines — stays where its author put it; the contents beside it link to each section on GitHub.
音频降噪
清理录音中的背景噪声。降噪专项 SKILL——通过共享脚本 skills/shared/scripts/audio_ops.py denoise 封装 ffmpeg 降噪滤镜,提供三级方案(从基础滤波到 RNN 神经网络),参数确定、可复现,不现场手拼命令。
通用音频操作(裁剪/转码/音量/提取/拼接/淡入淡出/变速)见 audio-editing。本 SKILL 专注降噪。
输入
| 字段 | 必填 | 说明 |
|---|---|---|
| input_file | 是 | 音频或视频文件路径 |
| tier | 否 | 1 / 2 / 3(默认 2) |
| output_file | 否 | 默认 outputs/主题名/{filename}-clean.{ext} |
| mix | 否 | 降噪强度 0.0-1.0(默认 0.8,仅 tier3 RNNoise) |
| preserve_original | 否 | 保留原始文件(默认 true) |
支持格式:wav, mp3, flac, aac, m4a, mp4, mkv, mov。
输出
- 降噪后的音频文件(放入
outputs/主题名/) - 处理报告:原始文件信息、所用 tier 与滤镜链、输出文件信息、大小对比
三级降噪方案(对应脚本 --tier)
脚本按 tier 自动选滤镜链并打印实际执行的 ffmpeg 命令。
Tier 1 — 基础降噪(ffmpeg 内置滤波)
切除低频隆隆声、高频嘶嘶声 + FFT 降噪,纯 ffmpeg 无外部依赖。
滤镜:highpass=f=80,lowpass=f=8000,afftdn=nr=12:nf=-40:tn=1
适用:轻度噪声、无需模型的快速处理。
Tier 2 — 加强降噪(更强 FFT + 非局部均值)
更激进的 FFT 降噪叠加 anlmdn,仍纯 ffmpeg。
滤镜:highpass=f=70,afftdn=nr=24:nf=-30:tn=1,anlmdn=s=0.0005
适用:中度噪声、稳态背景噪声(空调/风扇/底噪)。默认档。
Tier 3 — RNN 神经网络降噪(arnndn + 后处理)
RNNoise 针对人声优化 + 高通预处理 + 动态压缩 + 响度归一化。
滤镜:highpass=f=60,arnndn=m=<model>:mix=<mix>,acompressor=...,loudnorm=...
适用:人声录音、播客、访谈、复杂噪声环境。
需要 RNNoise 模型 sh.rnnn;缺失时脚本自动降级 Tier2 并打印下载提示。
执行步骤
脚本路径(相对项目根):skills/shared/scripts/audio_ops.py。
1. 探测输入文件
python skills/shared/scripts/audio_ops.py info input_file
脚本自身检查 ffmpeg/ffprobe,缺失时给安装提示。向用户展示时长/码率/声道,判断音频还是视频。
2. 准备 RNN 模型(仅 Tier3)
检查脚本旁 skills/shared/scripts/models/sh.rnnn 是否存在。不存在则下载:
mkdir -p skills/shared/scripts/models
curl -L https://github.com/GregorR/rnnoise-models/raw/master/somnolent-hogwash-2018-09-01/sh.rnnn \
-o skills/shared/scripts/models/sh.rnnn
不下载也可——脚本会自动降级 Tier2。也可用 --model <path> 指定其它模型。
3. 执行降噪
# 默认 Tier2
python skills/shared/scripts/audio_ops.py denoise input.wav -o outputs/主题名/input-clean.wav --tier 2
# Tier3(RNNoise,人声)
python skills/shared/scripts/audio_ops.py denoise input.wav -o outputs/主题名/input-clean.wav --tier 3 --mix 0.8
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 117 lines · 143 tokens per session scan A 97b79956fc00
audio-denoise is a skill published in the GitHub repository ZJU-REAL/Easel (411 stars, last pushed today), licensed Apache-2.0. It adds 143 tokens to every session and 1,552 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
ppt-generation
Generate PPTX presentations from slide plan + content.
chart-visualization
Generate charts: select type, extract data, render image.
jacky-motion2-0-srt
A workflow for turning a Chinese spoken script and matching SRT subtitle file into a single 16:9 HTML information animation. SRT is a subtitle file that stores text with start and end times; the animation follows those times and adds recorded-screen placeholders when needed.
video-podcast-maker
Use when the user gives a topic and wants an automated topic-driven narrated explainer, podcast, or knowledge-summary video (Bilibili / YouTube / Xiaohongshu / Douyin / WeChat Channels), or asks to learn visual design patterns from a reference video/image. Trigger when the user mentions creating a knowledge video…
video-podcast-maker-lite
Minimal personal narrated-video pipeline — a topic becomes a talking-head-free explainer MP4 (1080p or 4K) via script → Azure TTS (SSML) → Remotion. Use when the user wants a quick narrated video from a topic without the full video-podcast-maker machinery (no extra skills, no thumbnails/shorts/publish matrix). Do NOT…
video-podcast-maker-nano
Smallest personal narrated-explainer-video pipeline (spoken narration over visuals, not an audio podcast), fully tool-agnostic and autonomous by default — topic → research ∥ asset collection → script → TTS → video → 4K render ∥ publish info + cover. The skill defines the pipeline logic and self-verified checkpoints…