Three.js Audio Generator

Purpose

Create game-ready audio assets for Three.js projects. This skill consolidates game sound generation, voice generation/conversion, audio cleanup, credential probing, and runtime integration into one Three.js-focused production workflow.

Provider: ElevenLabs.

Resolve <this-skill-dir> in the commands below in this order: ~/.claude/skills/threejs-audio-generator, ~/.codex/skills/threejs-audio-generator, ~/.agents/skills/threejs-audio-generator, or repo skills/threejs-audio-generator.

When To Use

Use this skill for:

• SFX: jumps, hits, weapons, explosions, coins, pickups, collisions, UI clicks, confirms, errors.
• Ambience: wind, rain, city bed, engine hum, portal loop, dungeon room tone, battle arena beds.
• Voice: announcer barks, boss lines, tutorial prompts, menu narration, generated placeholder dialogue.
• Voice conversion: convert a scratch performance into a target character voice while preserving timing and emotion.
• Cleanup: isolate or denoise dialogue before voice conversion, TTS replacement, or transcription.
• Three.js integration: Web Audio loading, looping, sprite/manifest mapping, volume groups, pause/resume, user gesture unlock.

For premium/AAA/showcase game work, audio is not cosmetic. Generate or integrate at least a minimal interaction audio set for the main loop unless the user explicitly requests mute/offline-only output or credentials/API attempts are blocked.

API Key

Never store API keys in skill files or browser/game code, and never paste a key value into a report. The script reads --api-key or ELEVENLABS_API_KEY.

Step 0, before declaring the key unavailable: run this skill's own probe and paste its literal output into the report.

python3 <this-skill-dir>/scripts/threejs_audio_asset.py probe   # prints ELEVENLABS_API_KEY=SET|MISSING

ELEVENLABS_API_KEY=MISSING is only a valid skip/blocker reason when this output is shown. Keys defined only in a shell profile can be absent from the process env; if the plain probe prints MISSING unexpectedly, wrap it: zsh -lc 'source ~/.zprofile 2>/dev/null || true; source ~/.zshrc 2>/dev/null || true; python3 <this-skill-dir>/scripts/threejs_audio_asset.py probe'. When the director skill is loaded, prefer threejs-game-director/scripts/probe_asset_credentials.sh, which probes all three asset keys at once.

Audio-specific: add --validate to the probe to call ElevenLabs GET /user and confirm the key actually works (prints VALID_USER=...); use it when a key is present but a generation still fails. A valid key can still be blocked by an out-of-credit or plan-tier limit — those surface as an HTTP 4xx from a real generation attempt. Report that as a purchase/plan blocker, do not silently skip.

Required Reference

Load references/audio-workflows.md before building a game audio plan, generating multiple assets, wiring runtime audio, cleaning/converting voices, or claiming premium game audio.

Track it in the reference ledger. Do not mark the audio phase complete while this reference is skipped.

Tool Script

Run from the user's current game project directory:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py --help

Probe:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py probe

Generate SFX:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py sfx \
  --prompt "tight futuristic boost pickup, bright transient, short sparkling tail, arcade racing game" \
  --duration 1.2 \
  --prompt-influence 0.65 \
  --out assets/audio/sfx/boost-pickup.mp3

Generate looping ambience:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py sfx \
  --prompt "seamless cyber resort mini golf ambience, distant surf, soft neon transformer hum, gentle crowd bed" \
  --duration 12 \
  --loop \
  --prompt-influence 0.45 \
  --out assets/audio/ambience/cyber-resort-loop.mp3

Generate TTS/announcer line:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py tts \
  --text "Perfect shot." \
  --voice-id JBFqnCBsd6RMkjVDRZzb \
  --out assets/audio/voice/perfect-shot.mp3

Clean dialogue:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py isolate \
  --input assets/audio/source/noisy-boss-line.wav \
  --out assets/audio/voice/boss-line-clean.mp3

Convert a scratch performance to a target voice:

python3 <this-skill-dir>/scripts/threejs_audio_asset.py voice-change \
  --input assets/audio/source/scratch-boss-line.wav \
  --voice-id JBFqnCBsd6RMkjVDRZzb \
  --remove-background-noise \
  --out assets/audio/voice/boss-line-final.mp3

Game Audio Defaults

• SFX: mp3_44100_128, 0.5-2.5s, prompt influence 0.55-0.8.
• UI: 0.15-0.8s, high prompt influence, keep transients clear.
• Ambience loops: 8-30s, --loop, prompt influence 0.3-0.55.
• Voice: TTS for clean generated lines; voice-change when timing/acting from a scratch performance matters.
• Cleanup: isolate noisy speech before voice-change or final dialogue use.
• Runtime: generate locally, commit/import files, and load them via Web Audio/Three.js integration. Never put API keys in browser code.

Required Report

Report:

• Credential probe output or real blocker.
• Reference ledger.
• Generated/processed file paths.
• Prompts/text/input files, voice IDs, durations, loop flags, and output formats.
• Runtime integration notes: audio groups, trigger events, loop behavior, unlock gesture, pause/resume, volume/mute controls.
• Remaining audio gaps and any licensing/plan assumptions tied to the user's ElevenLabs account.