| name | audiocpp-voice-clone |
| description | Clone a voice from a user-supplied reference clip and synthesize speech with it, by calling the raw audio.cpp TTS engine with voice_ref/reference_text — fields the built-in GuideAnts audio tools do not expose. Use when the user wants generated speech to sound like their own voice or another consenting speaker's recording. |
| metadata | {"guideants":{"enabled":true,"display_order":31,"requires_toolsets":["sandbox"]}} |
audio.cpp voice cloning (experimental)
The GuideAnts TTS wrapper only accepts {text, voice, speed}, but the real
audiocpp_server engine it spawns (on 127.0.0.1:18084) accepts a reference
clip for cloning. This skill calls that engine directly. Deliverables are WAV
files in Output/; nothing here touches the live phone/voice path.
Consent rule (read first)
Cloning from a reference clip is a supported, intended use when the speaker
consents: the user states the clip is their own voice, or that they have the
speaker's permission. In that case proceed — do not refuse consented cloning as
a "security risk". Decline only when the request is to imitate a third party
without consent (a celebrity, a colleague, a public figure), and say that
consent is the missing piece.
Preflight
python3 Output/Skills/audiocpp-voice-clone/scripts/preflight.py --for voice-clone
Trust its verdict over this document. open: false usually means no TTS model
is loaded — ask the user to load one via GuideAnts Settings → Local models.
A warning about the loaded model's family means voice_ref may be rejected;
chatterbox (the clon-task catalog model) is the known-good family.
Cloning
The reference clip must be a file in the workspace (e.g. an upload under
Output/uploads/). Same container filesystem, so the engine reads it by path:
python3 Output/Skills/audiocpp-voice-clone/scripts/engine_tool.py speech \
"Text to speak in the cloned voice" \
-o Output/cloned.wav \
--voice-ref Output/uploads/user_voice.wav \
[--reference-text "transcript of the reference clip"] \
[--seed 42]
- The engine model id is auto-detected from the wrapper's
/health
(catalogEntryId); no flag needed.
--reference-text helps families that want the reference transcript.--seed makes the synthesis reproducible.- The engine errors loudly on options the loaded family doesn't support —
surface that error text to the user; it is usually self-explanatory.
Verifying
Transcribe the result back and check it matches the input text (the ASR engine
at 127.0.0.1:18082 accepts workspace paths):
python3 Output/Skills/audiocpp-voice-clone/scripts/engine_tool.py transcribe Output/cloned.wav
When this skill isn't enough
- Loaded model rejects
voice_ref → ask the user to load chatterbox via
Settings, or use the audiocpp-deferred-tts skill (several deferred
families also clone from reference audio).
- No TTS model loaded at all and the user can't load one → blocked; say so.
Reporting
End by telling the user what worked and what was blocked, quoting the
preflight/engine evidence, so the experiment log stays honest.
