audio-master

name	audio-master
description	Master audio to streaming-spec LUFS (two-pass ffmpeg loudnorm) and concatenate multiple mastered files into one with silence gaps. Use when a directive needs podcast/album/narration audio prepared for Spotify, YouTube, Apple Podcasts, etc. One skill, many callers — pass --codec and --target-lufs to tune per platform.
metadata	{"clawdbot":{"emoji":"🔊","requires":["node","ffmpeg"],"credentials":"none — reads/writes local files only"}}

Audio Master

General-purpose audio mastering. Every directive that finishes raw TTS or recorded audio and needs a clean, loudness-normalised output calls this skill.

Install

cd ~/clawd/skills/audio-master
pnpm install
pnpm run build

Requires ffmpeg on PATH.

Usage

Master a single file to Spotify-spec (-14 LUFS integrated)

audio-master master \
  --input ./raw/segment-03.mp3 \
  --output ./out/segment-03.mp3 \
  --target-lufs -14 \
  --true-peak-db -1 \
  --codec mp3

Two-pass loudnorm: pass 1 measures integrated loudness/true-peak/LRA/threshold, pass 2 applies normalisation with the measured values for accurate targeting.

Concatenate multiple mastered files with silence between them

audio-master concat \
  --inputs "./segments/seg-01.mp3,./segments/seg-02.mp3,./segments/seg-03.mp3" \
  --output ./full.mp3 \
  --gap-ms 500 \
  --codec mp3

Re-encodes on concat (not -c copy) and inserts a silence file matching the output codec/sample-rate/channels to avoid the "stereo-silence + mono-segments = malformed MP3" bug.

Emits JSON on stdout on success:

{ "ok": true, "op": "master", "path": "./out/segment-03.mp3",
  "measured": { "input_i": "-23.1", "input_tp": "-5.4", "input_lra": "7.3", "input_thresh": "-33.1", "target_offset": "0.15" } }

{ "ok": true, "op": "concat", "path": "./full.mp3", "segments": 3, "gap_ms": 500 }

Flags

master

Flag	Default	Notes
--input <path>	required	Source audio file
--output <path>	required	Destination (parent dirs auto-created)
--target-lufs <n>	-14	Spotify/YouTube/Apple default. Broadcast uses -23.
--true-peak-db <n>	-1	dBTP ceiling
--lra <n>	11	Loudness range target (LU)
--sample-rate <hz>	44100	Output sample rate
--codec <mp3|wav>	mp3	mp3 = libmp3lame 128k mono. wav = PCM 16-bit.

concat

Flag	Default	Notes
--inputs <a,b,c>	required	Comma-separated paths, in order
--output <path>	required	Destination
--gap-ms <n>	500	Silence duration between segments
--sample-rate <hz>	44100
--codec <mp3|wav>	mp3	Must match the format of the input segments for best results

Error classification

exit JSON error	Meaning
config	Bad arguments, missing input files, or ffmpeg not found on PATH
upstream	ffmpeg ran but exited non-zero — content error (unsupported codec, corrupt input, etc.)
unknown	Unexpected — check stderr

Callers

• My Podcast — my-podcast-audio-producer (concat ElevenLabs segments → episode MP3)
• My Album — suno-track-master (master each generated track to album LUFS)
• My Show — cartoon-voice-director (optional: master polished voice lines)

See also

• elevenlabs-tts — upstream TTS that produces the raw segments
• Vault Overview: 08 - System/Agent Skills/audio-master - Skill Overview.md