Jarvis Vocal

Authentic J.A.R.V.I.S. voice synthesis using Piper TTS with HuggingFace-trained model. Generates movie-accurate voice locally and can push to connected Andro...

Security Scan

The skill's files and runtime instructions are consistent with a local Piper TTS + HuggingFace model workflow and the Android push/ADB integration it advertises; it does not request unrelated credentials or perform unexplained actions.

Purpose & Capability

Name/description (J.A.R.V.I.S. voice via Piper TTS and HuggingFace model) match the instructions and package metadata: the SKILL.md and README show how to install Piper, download the model, generate WAVs, and push them to Android devices via ADB/Tailscale. No unrelated binaries, env vars, or config paths are required by the skill itself.

Instruction Scope

Instructions are narrowly scoped to: installing piper-tts/ffmpeg, using the HuggingFace CLI to download model files into ~/.local/share/piper/voices, generating audio, and optionally streaming/pushing via adb. They do not instruct reading unrelated system files or exporting arbitrary data. Note: using ADB/Tailscale gives device-level access to paired Android devices (expected for the advertised capability); README contains a sample IP which is just illustrative but could be misleading if copied without understanding.

Install Mechanism

This is an instruction-only skill (no automated install). It tells the user to pipx install piper-tts and to use the HuggingFace CLI and ffmpeg. Those are reasonable for TTS but do require installing third-party software that will run locally—verify the upstream packages (piper-tts, hf CLI) before installing. Because install is manual, the skill itself does not download or execute code automatically.

Credentials

The skill declares no required environment variables or credentials. Operationally, the workflow may prompt for HuggingFace auth if model access requires it and requires an ADB-paired Android device (device pairing grants access to push/play files). There are no unrelated secret requests in the manifest or instructions.

Persistence & Privilege

The skill is not force-included (always: false) and has no install step that persists code automatically. It does not request elevated agent privileges or modify other skills. Note that the platform default allows autonomous invocation; that is expected and not by itself a concern here.

Assessment

This skill appears coherent, but before installing or running anything: 1) Inspect and verify the upstream packages it asks you to install (piper-tts, HuggingFace CLI, ffmpeg) and install them manually via trusted channels. 2) Confirm the HuggingFace model license and provenance — README claims movie-line training (legal/ethical risk, even if model metadata says MIT). 3) Understand ADB/Tailscale implications — pairing a phone gives the host permission to push files and trigger playback; only pair with devices you trust. 4) Review any wrapper scripts (jarvis-speak/jarvis-tts) before running to ensure they do only generation/push/cleanup and contain no unexpected commands. 5) Because installation is manual, nothing in the skill will automatically exfiltrate credentials, but installed third-party tools will run on your machine—treat them as you would any pip-installed software.

Uses the authentic J.A.R.V.I.S. voice model from HuggingFace (trained on actual movie lines) via Piper TTS. No audio effects needed — the voice is naturally cinematic and British.

Credit: Voice model by jgkawell — see the discussion for details on training and samples.

Usage

Generate a WAV file:

{baseDir}/bin/jarvis-tts "Text to speak" ./output.wav

Stream directly to an Android device (if ADB connected):

{baseDir}/bin/jarvis-tts "Text to speak" - | adb push - /sdcard/Download/temp.wav

Installation

Prerequisites

pipx install piper-tts
sudo apt install ffmpeg  # or equivalent

Install Voice Model

# Create voice directory
mkdir -p ~/.local/share/piper/voices/en_GB

# Download models via HuggingFace CLI
cd ~/.local/share/piper/voices/en_GB
hf download jgkawell/jarvis en/en_GB/jarvis/high/jarvis-high.onnx --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/high/jarvis-high.onnx.json --local-dir .
# Optional: medium quality model
hf download jgkawell/jarvis en/en_GB/jarvis/medium/jarvis-medium.onnx --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/medium/jarvis-medium.onnx.json --local-dir .

Integration

Works with OpenClaw Android nodes via ADB over Tailscale. Use jarvis-speak wrapper for one-command push+play:

jarvis-speak "Systems at your service, Sir."

Or use streaming mode (faster, ephemeral):

jarvis-speak "Message" --stream

Configuration

Setting	Default	Description
Model	jarvis-high	Voice quality: high (114MB) or medium (63MB)
Speed	1.0 (native)	Piper length-scale — adjust for faster/slower speech
Volume	1.0	Post-processing volume boost

Edit jarvis-speak script to change defaults.

Troubleshooting

"Model not found" → Download models to ~/.local/share/piper/voices/en_GB/jarvis-*

ADB connection refused → Ensure phone's ADB over WiFi is enabled and paired with laptop (port 5555)

Audio doesn't play → Check Android receives the file at /sdcard/Download/jarvis-current.wav and has a WAV-capable media player

License

MIT — The voice model is MIT licensed by jgkawell.

Credits

• Voice model: jgkawell/jarvis on HuggingFace — trained on Marvel movie lines
• TTS engine: Piper by Rhasspy
• Integration: OpenClaw by Aidan Park