← 返回 Skills 市场
1443
总下载
2
收藏
5
当前安装
2
版本数
在 OpenClaw 中安装
/install audio
功能描述
Process, enhance, and convert audio files with noise removal, normalization, format conversion, transcription, and podcast workflows.
使用说明 (SKILL.md)
Requirements
Required:
ffmpeg/ffprobe— core audio processing
Optional (for advanced features):
sox— additional noise reductionwhisper— local transcription (or use API)demucs— stem separation
Quick Reference
| Situation | Load |
|---|---|
| FFmpeg commands by task | commands.md |
| Loudness standards by platform | loudness.md |
| Podcast production workflow | podcast.md |
| Transcription workflow | transcription.md |
Core Capabilities
| Task | Method |
|---|---|
| Convert formats | FFmpeg (-acodec) |
| Remove noise | FFmpeg filters or SoX |
| Normalize loudness | ffmpeg-normalize or -af loudnorm |
| Transcribe | Whisper → text, SRT, VTT |
| Separate stems | Demucs (vocals, drums, bass, other) |
Execution Pattern
- Clarify goal — What format? What loudness? What platform?
- Analyze source —
ffprobefor codec, sample rate, channels, duration - Process — FFmpeg/SoX for transformation
- Verify — Check output plays, meets specs, sounds correct
- Deliver — Provide file to user
Common Requests → Actions
| User says | Agent does |
|---|---|
| "Convert to MP3" | -acodec libmp3lame -q:a 2 |
| "Remove background noise" | Apply highpass/lowpass or dedicated denoiser |
| "Normalize for podcast" | -af loudnorm=I=-16:TP=-1.5:LRA=11 |
| "Transcribe this" | Whisper → output SRT/VTT/TXT |
| "Extract audio from video" | -vn -acodec copy or re-encode |
| "Make it smaller" | Lower bitrate: -b:a 128k or -b:a 96k |
| "Speed up 1.5x" | -af atempo=1.5 |
Format Quick Reference
| Format | Use Case | Quality |
|---|---|---|
| WAV | Master, editing | Lossless |
| FLAC | Archive, audiophile | Lossless compressed |
| MP3 | Universal sharing | Lossy, 128-320 kbps |
| AAC/M4A | Apple, podcasts | Lossy, efficient |
| OGG/Opus | WhatsApp, Discord | Lossy, very efficient |
Quality Defaults
- Podcast: -16 LUFS (Spotify), -19 LUFS (Apple)
- Music: -14 LUFS (Spotify), -16 LUFS (Apple Music)
- MP3 quality: VBR
-q:a 2(~190 kbps) or CBR-b:a 192k - Sample rate: 44.1kHz for music, 48kHz for video sync
Scope
This skill:
- Processes audio files user explicitly provides
- Runs FFmpeg commands on user request
- Does NOT access cloud services without user knowing
- Does NOT store files persistently (user manages their files)
安全使用建议
This skill is a documentation-only helper for running FFmpeg/Whisper-style audio workflows and appears coherent with its purpose. Before using: (1) ensure ffmpeg and ffprobe are installed from trusted packages, (2) review any shell snippets (especially batch scripts) before running them to avoid accidental mass changes to files in the current directory, (3) be aware optional transcription/diarization examples call external APIs and require API keys — only provide those keys if you trust the service, and (4) confirm how your agent is allowed to access files/network (if the agent has broad file or network privileges, treat any automated commands cautiously). If you want a lower-risk test, run commands on a disposable/sample file or in a sandboxed environment first.
功能分析
Type: OpenClaw Skill
Name: audio
Version: 1.0.1
The skill bundle is classified as suspicious due to significant shell injection vulnerabilities. Specifically, the use of `ffmpeg -f concat -safe 0` in `commands.md` and `podcast.md` allows arbitrary file paths in the concatenation list, posing a risk if the agent constructs this list from unsanitized user input. Additionally, other shell commands involving command substitution and arithmetic (`$()`, `bc`) in `commands.md` and `podcast.md` could lead to arbitrary code execution if user-provided input is not properly sanitized by the OpenClaw agent before being passed to the shell.
能力评估
Purpose & Capability
Name/description (audio processing, noise removal, normalization, format conversion, transcription, podcast workflows) match the declared requirements (ffmpeg, ffprobe) and the provided command examples. Optional tools (sox, whisper, demucs) are appropriate for the advanced features the skill describes.
Instruction Scope
SKILL.md and the included docs instruct the agent to run FFmpeg/SoX/Whisper commands on user-provided files, inspect files with ffprobe, and run batch scripts. This is within scope. The docs also include examples of using external APIs (AssemblyAI) and diarization tools that require tokens—these are explicit, optional, and consistent with transcription use cases, but they do introduce network interactions if the user chooses them.
Install Mechanism
Instruction-only skill with no install spec or downloaded code. That minimizes disk-write and supply-chain risk; the only runtime requirement is that the environment has the declared binaries installed.
Credentials
The skill declares no required environment variables or credentials. Optional examples mention API tokens (AssemblyAI, Hugging Face) only where those services are used. There are no unrelated secrets requested.
Persistence & Privilege
always is false and the skill contains no installation steps or actions that modify other skills or system-wide agent settings. The included batch scripts operate on files in the working directory (expected for a file-processing skill).
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install audio - 安装完成后，直接呼叫该 Skill 的名称或使用
/audio触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.1
Declare required binaries (ffmpeg, ffprobe), add requirements section with optional deps, add explicit scope
v1.0.0
Initial release
元数据
常见问题
Audio 是什么？
Process, enhance, and convert audio files with noise removal, normalization, format conversion, transcription, and podcast workflows. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 1443 次。
如何安装 Audio？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install audio」即可一键安装，无需额外配置。
Audio 是免费的吗？
是的，Audio 完全免费（开源免费），可自由下载、安装和使用。
Audio 支持哪些平台？
Audio 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（linux, darwin, win32）。
谁开发了 Audio？
由 Iván（@ivangdavila）开发并维护，当前版本 v1.0.1。
推荐 Skills