Synthesize against the user's own host-native audiocpp_server (outside GPU host / GuideAnts containers) — including families the GPU host image binary lacks. TTS only. Use when the user runs their own audio.cpp build.
Installs into .claude/skills of the current project.
Are you the author of Audiocpp Host Tts?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/elumenotion-audiocpp-host-tts-guideants)
---
name: audiocpp-host-tts
description: "Synthesize against the user's own host-native audiocpp_server (outside GPU host / GuideAnts containers) — including families the GPU host image binary lacks. TTS only. Use when the user runs their own audio.cpp build."
metadata:
guideants:
enabled: true
display_order: 36
requires_toolsets: [sandbox]
---
# audio.cpp host-native engine (experimental)
Paths — fixed layout, do not probe or re-derive. The sandbox CWD is the
notebook's **output directory**. This skill's scripts live under
`Skills/audiocpp-host-tts/scripts/` relative to it. Write deliverables with
**bare filenames**; never prefix with `Output/`.
This skill does **not** use the GPU host raw audiocpp gateway. It talks to an
`audiocpp_server` the user runs on their own machine (or LAN), for loaders the GPU host
does not ship (e.g. Kokoro forks).
For the GPU host-hosted ASR/TTS/diarize/deferred catalog gaps, use the other audiocpp
skills with `AUDIOCPP_SKILL_BASE_URL` instead.
## Preflight
```bash
python3 Skills/audiocpp-host-tts/scripts/preflight.py --for host-tts
```
Resolution: `AUDIOCPP_ENGINE_URL`, else `http://host.docker.internal:8080`.
## Synthesize
`--model` is required:
```bash
python3 Skills/audiocpp-host-tts/scripts/engine_tool.py speech "Hello" \
--engine-url http://host.docker.internal:8080 --model <engine-model-id> \
-o hello.wav [--voice <id>] [--seed 42] [--language en] [--voice-ref ...]
```
Consent rule applies to `--voice-ref`.
## Hard limit: TTS only
Host `/v1/audio/transcriptions` needs host-local paths — sandbox files are not
visible. Use **audiocpp-asr** (GPU host gateway) for transcription.
## Troubleshooting
| Symptom | Likely cause |
|---|---|
| Unreachable | Not bound to `0.0.0.0`, or firewall |
| `/health` 200 but no `/v1/audio/*` | Wrong process (e.g. llama-server) on that port |
| Model errors | Family/option mismatch — surface engine text |
## Reporting
End by telling the user what worked and what was blocked, quoting preflight evidence.
