Jarvis Voice Persona

A metallic AI voice with visual transcript styling for OpenClaw assistants.

Features

• TTS Output: Local speech synthesis via sherpa-onnx (no cloud API)
• Metallic Voice: ffmpeg audio processing for robotic resonance
• Purple Transcripts: Visual distinction between spoken and written content
• Fast Playback: 2x speed for efficient communication

Requirements

• sherpa-onnx with VITS piper model (en_GB-alan-medium recommended)
• ffmpeg for audio processing
• aplay (ALSA) for audio playback

Installation

1. Install sherpa-onnx TTS

# Download and extract sherpa-onnx
mkdir -p ~/.openclaw/tools/sherpa-onnx-tts
cd ~/.openclaw/tools/sherpa-onnx-tts
# Follow sherpa-onnx installation guide

2. Install the jarvis script

cp {baseDir}/scripts/jarvis ~/.local/bin/jarvis
chmod +x ~/.local/bin/jarvis

3. Configure audio device

Edit ~/.local/bin/jarvis and set your audio output device in the aplay -D line.

Usage

Speak text

jarvis "Hello, I am your AI assistant."

In agent responses

Add to your SOUL.md:

## Communication Protocol

- **Hybrid Output:** Every response includes text + spoken audio via `jarvis` command
- **Transcript Format:** **Jarvis:** \x3Cspan class="jarvis-voice">spoken text\x3C/span>
- **No gibberish:** Never spell out IDs or hashes when speaking

Transcript styling (requires UI support)

Add to your webchat CSS:

.jarvis-voice {
  color: #9B59B6;
  font-style: italic;
}

And allow span in markdown sanitization.

Voice Customization

Edit ~/.local/bin/jarvis to adjust:

Presets

More robotic:

aecho=0.7:0.7:5|10|15:0.4|0.35|0.3

More human:

aecho=0.4:0.4:20:0.2

Deeper:

highpass=f=200,lowpass=f=3000

Troubleshooting

No audio output

• Check aplay -l for available devices
• Update the -D plughw:X,Y parameter

Voice too fast/slow

• Adjust --vits-length-scale (0.3=very fast, 1.0=normal)

Metallic effect too strong

• Reduce echo delays and chorus depth

Files

• scripts/jarvis — TTS script with metallic processing
• SKILL.md — This documentation

A voice persona for assistants who prefer to be heard as well as read.