audio-production

audio-production-plugin

Claude Code plugin for audio engineering & production — voice profiling, EQ preset suggestion and application, compression, de-essing, normalisation, VAD segmentation, mastering, tagging, and podcast assembly. ffmpeg-first primitives plus a personal voice-profile workflow that persists to a versioned user-data directory.

Part of the danielrosehill Claude Code marketplace.

For transcription, diarisation, or transcript export, install the companion Claude-Transcription-Plugin.

What you get

Voice profiling & EQ workflow

The plugin captures a reference voice sample for each microphone the user records with, analyses its spectral characteristics, and generates tailored EQ + dynamics presets that are bound to that mic. Profiles, presets, and A/B auditions all persist in a versioned user-data directory.

• /audio-production:onboard — first-run setup. Creates the user-data directory and walks through registering the user's primary microphone.
• /audio-production:add-mic — register a new mic (id, make/model, interface, environment notes), extract a 3-min sample from a source recording, profile it, and seed presets bound to it.
• /audio-production:list-mics — show all registered mics and the presets bound to each.
• /audio-production:extract-sample <input> — auto-pick the loudest 3-min window from a longer recording.
• /audio-production:profile-voice [--mic=<id>] — analyse a mic's reference sample with librosa. Writes F0, spectral centroid, sibilance/mud band energy, resonant peaks, and (optionally) formants.
• /audio-production:suggest-eq --use-case=<podcast|vocals|spoken-word|broadcast> [--mic=<id>] — translate the analysis into an EQ + dynamics preset and emit a 1-min A/B audition.
• /audio-production:audition-preset <preset> — emit a fresh 1-min before/after WAV pair for any saved preset.
• /audio-production:tune-preset — interactively narrow in on a preset by listening to 15s A/B variants (with side-by-side spectrograms and a single-file compare.wav that announces "Sample 1" / "Sample 2" via TTS so you don't have to track which file is which). Iterate based on your feedback ("more presence", "less mud", "softer compression") until you're happy, then save the winner.
• /audio-production:generate-cues — pre-render the TTS announcement clips (default: edge-tts neural voices) once to <data-dir>/tts/. Reused by tune-preset and audition-preset on every session.
• /audio-production:list-presets — list saved presets with a one-line summary of each chain.
• /audio-production:apply-preset <name> <input> — run a saved preset against an audio file via ffmpeg.

One-shot finisher

• polish <input> [--mode=clean|noisy] — orchestrates the full chain. clean (default): truncate-silence → EQ preset chain → loudnorm. noisy: denoise → truncate-silence → EQ preset chain → loudnorm. Writes <stem>.polished.wav plus a .log.txt audit trail.

Audio engineering primitives

• normalize — two-pass EBU R128 loudnorm (default target -16 LUFS, configurable)
• check-loudness — measure integrated LUFS, true peak, LRA without modifying the file
• denoise — local-first noise reduction (DeepFilterNet ML, validated; ffmpeg afftdn fallback)
• compress — single-band ffmpeg acompressor with use-case shortcuts
• de-ess — band-limited dynamic cut for sibilance reduction (ffmpeg-only proxy)
• apply-chain — full chain (HPF → EQ → de-ess → compressor → loudnorm) in one invocation, from a preset or use-case shortcut
• trim-silence — strip leading/trailing silence via silenceremove
• truncate-silence — collapse internal silences throughout a recording (validated ffmpeg silenceremove tuning, optional silero-vad)
• silence-cut — tighten a recording with real cuts (auto-editor); threshold + margin driven, more aggressive than truncate-silence
• silence-cut-edl — same detection but emits an editable timeline (Kdenlive / Final Cut / Premiere / Shotcut) for review before render
• time-stretch — speed-up/slow-down preserving pitch, or pitch-shift preserving duration (rubberband, ffmpeg atempo fallback)
• detect-cues — acoustic cue/chapter detection (aubio onset / beat / pitch) — emits sidecar JSON for assembly or chapter authoring
• concat-audio — concat or crossfade intro + body + outro into a single master
• convert-format — convert between WAV / FLAC / MP3 / Opus / AAC with explicit bitrate/sample-rate
• tag-audio — show or set ID3/Vorbis/FLAC tags and embed cover art
• vad-segment — voice-activity-detect an audio file and emit per-segment outputs or a timing sidecar

Podcast primitives