# Send Her Voice to your agent
Hand the extracted package to your coding agent with a concrete install brief instead of figuring it out manually.
## Fast path
- Download the package from Yavira.
- Extract it into a folder your agent can access.
- Paste one of the prompts below and point your agent at the extracted folder.
## Suggested prompts
### New install

```text
I downloaded a skill package from Yavira. Read SKILL.md from the extracted folder and install it by following the included instructions. Tell me what you changed and call out any manual steps you could not complete.
```
### Upgrade existing

```text
I downloaded an updated skill package from Yavira. Read SKILL.md from the extracted folder, compare it with my current installation, and upgrade it while preserving any custom configuration unless the package docs explicitly say otherwise. Summarize what changed and any follow-up checks I should run.
```
## Machine-readable fields
```json
{
  "schemaVersion": "1.0",
  "item": {
    "slug": "her-voice",
    "name": "Her Voice",
    "source": "tencent",
    "type": "skill",
    "category": "AI 智能",
    "sourceUrl": "https://clawhub.ai/matusvojtek/her-voice",
    "canonicalUrl": "https://clawhub.ai/matusvojtek/her-voice",
    "targetPlatform": "OpenClaw"
  },
  "install": {
    "downloadUrl": "/downloads/her-voice",
    "sourceDownloadUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=her-voice",
    "sourcePlatform": "tencent",
    "targetPlatform": "OpenClaw",
    "packageFormat": "ZIP package",
    "primaryDoc": "SKILL.md",
    "includedAssets": [
      "CHANGELOG.md",
      "SKILL.md",
      "assets/HerVoice.swift",
      "scripts/config.py",
      "scripts/daemon.py",
      "scripts/setup.py"
    ],
    "downloadMode": "redirect",
    "sourceHealth": {
      "source": "tencent",
      "slug": "her-voice",
      "status": "healthy",
      "reason": "direct_download_ok",
      "recommendedAction": "download",
      "checkedAt": "2026-05-01T20:30:11.280Z",
      "expiresAt": "2026-05-08T20:30:11.280Z",
      "httpStatus": 200,
      "finalUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=her-voice",
      "contentType": "application/zip",
      "probeMethod": "head",
      "details": {
        "probeUrl": "https://wry-manatee-359.convex.site/api/v1/download?slug=her-voice",
        "contentDisposition": "attachment; filename=\"her-voice-1.0.2.zip\"",
        "redirectLocation": null,
        "bodySnippet": null,
        "slug": "her-voice"
      },
      "scope": "item",
      "summary": "Item download looks usable.",
      "detail": "Yavira can redirect you to the upstream package for this item.",
      "primaryActionLabel": "Download for OpenClaw",
      "primaryActionHref": "/downloads/her-voice"
    },
    "validation": {
      "installChecklist": [
        "Use the Yavira download entry.",
        "Review SKILL.md after the package is downloaded.",
        "Confirm the extracted package contains the expected setup assets."
      ],
      "postInstallChecks": [
        "Confirm the extracted package includes the expected docs or setup files.",
        "Validate the skill or prompts are available in your target agent workspace.",
        "Capture any manual follow-up steps the agent could not complete."
      ]
    }
  },
  "links": {
    "detailUrl": "https://openagent3.xyz/skills/her-voice",
    "downloadUrl": "https://openagent3.xyz/downloads/her-voice",
    "agentUrl": "https://openagent3.xyz/skills/her-voice/agent",
    "manifestUrl": "https://openagent3.xyz/skills/her-voice/agent.json",
    "briefUrl": "https://openagent3.xyz/skills/her-voice/agent.md"
  }
}
```
## Documentation

### Her Voice 🎙️

Give your agent a voice. Audio responses powered by Kokoro TTS — a compact, naturally expressive model running entirely on-device.

### ✨ Features

Highly optimized response time thanks to on-the-fly audio streaming technology. 100% free, no API keys required. Inspired by Samantha and Sky.

⚡ On-the-fly Streaming — Audio plays as it generates, very low latency
👄 The Voice of an angel — Cutting-edge local text-to-speech model Kokoro TTS
🧠 TTS Daemon — Keep the model warm in RAM for instant responses (can be disabled to save RAM)
🖥️ Persist Mode — Drag & drop audio, paste text, use as a voice station
🔧 Fully Configurable — Voice, speed, visualizer, notification sounds
🍎 MLX + PyTorch — Native Metal acceleration on Apple Silicon, PyTorch fallback everywhere else
🎨 Real-time Visualizer — Floating 60fps LED bars that react to speech (macOS only)

### First-Run Setup

python3 SKILL_DIR/scripts/setup.py

Note: SKILL_DIR is the root directory of this skill — the agent resolves it automatically when running commands.

The setup wizard will:

Detect platform and select TTS engine (MLX on Apple Silicon, PyTorch elsewhere)
Find or install the appropriate TTS backend (mlx-audio or kokoro)
Install espeak-ng (Homebrew on macOS, apt on Linux)
Patch espeak loader if needed (macOS compatibility)
Compile the native visualizer binary (macOS only)
Download the Kokoro model
Create config at ~/.her-voice/config.json

Check status anytime:

python3 SKILL_DIR/scripts/setup.py status

### Post-Setup: Names & Pronunciation

After setup, configure the agent and user names:

python3 SKILL_DIR/scripts/config.py set agent_name "Jackie"
python3 SKILL_DIR/scripts/config.py set user_name "Matúš"
python3 SKILL_DIR/scripts/config.py set user_name_tts "Mah-toosh"

TTS pronunciation tip: If the user's name is non-English, figure out a phonetic English spelling that Kokoro will pronounce correctly. Store it in user_name_tts and use that spelling whenever speaking the name aloud. The real name stays in user_name for display purposes.

### Speaking Text

# Basic usage
python3 SKILL_DIR/scripts/speak.py "Hello, world!"

# Skip visualizer for this call
python3 SKILL_DIR/scripts/speak.py --no-viz "Quick note"

# Save to file instead of playing
python3 SKILL_DIR/scripts/speak.py --save /tmp/output.wav "Save this"

# Override voice or speed
python3 SKILL_DIR/scripts/speak.py --voice af_bella --speed 1.2 "Faster!"

# Pipe text from stdin
echo "Piped text" | python3 SKILL_DIR/scripts/speak.py

### Options

FlagDescription--no-vizSkip the visualizer for this call--persistKeep visualizer open after playback ends--save PATHSave audio to WAV file instead of playing--voice NAMEOverride the configured voice--speed NOverride the configured speed multiplier--mode MODEOverride visualizer mode (v2 or classic)

### Agent Workflow

When the user wants voice responses:

Check voice mode — is voice enabled or did the user ask for it?
Play notification sound (instant feedback while TTS generates):
afplay /System/Library/Sounds/Blow.aiff &


Speak the response:
python3 SKILL_DIR/scripts/speak.py "Response text here"


Always provide text alongside voice — accessibility matters.

### Notification Sound

The notification sound plays instantly (~0.1s) while TTS generates (~0.3-3s). This gives the user immediate feedback that the agent is responding.

Configure in ~/.her-voice/config.json:

{
  "notification_sound": {
    "enabled": true,
    "sound": "Blow"
  }
}

Available macOS sounds: Blow, Bottle, Frog, Funk, Glass, Hero, Morse, Ping, Pop, Purr, Sosumi, Submarine, Tink. Located in /System/Library/Sounds/.

### TTS Daemon

The daemon keeps the Kokoro model warm in RAM, eliminating ~1.1s of startup overhead per call.

The daemon auto-resolves the mlx-audio venv — no need to find the venv Python manually.

# Start (persists in background)
nohup python3 SKILL_DIR/scripts/daemon.py start > /tmp/her-voice-daemon.log 2>&1 & disown

# Status
python3 SKILL_DIR/scripts/daemon.py status

# Stop
python3 SKILL_DIR/scripts/daemon.py stop

# Restart
python3 SKILL_DIR/scripts/daemon.py restart

speak.py auto-detects the daemon: uses it if available, falls back to direct model loading.

The daemon is optional. Without it, speech still works — just ~1s slower per call as the model loads each time. Skip the daemon to save ~2.3GB RAM.

Note: The daemon writes its PID file and socket after the model is fully loaded and ready to accept connections. They live in ~/.her-voice/ with restricted permissions (owner-only access). The daemon won't survive a reboot — start it again after restart if needed.

### Visualizer

A floating overlay with three animated LED bars that react to speech in real-time. 60fps, native macOS (Cocoa + AVFoundation). macOS only — on other platforms, audio plays without the visualizer.

### Modes

v2 (default) — Three-tier pure red, center raw amplitude, sides with lag
classic — Original smooth gradient look

### Controls

KeyActionESCQuitSpacePause/Resume (file mode)← →Seek ±5s (file mode)⌘VPaste text to speak (persist mode)

### Persist Mode

Keep the visualizer on screen between playbacks. Use as a standalone voice station:

# Launch in persist mode (stays open, idle breathing animation)
~/.her-voice/bin/her-voice-viz --persist

# Stream mode + persist (stays open after speech ends)
python3 SKILL_DIR/scripts/speak.py --persist "Hello!"

In persist mode:

Drag & drop audio files (.wav, .mp3, .aiff, .m4a) onto the visualizer to play them
⌘V pastes clipboard text → streams directly from TTS daemon with full visualizer animation
Idle breathing — subtle center bar pulse when waiting for input

### Standalone Usage

# Play a file with visualizer
~/.her-voice/bin/her-voice-viz --audio /path/to/file.wav

# Demo mode (simulated audio)
~/.her-voice/bin/her-voice-viz --demo

# Stream raw PCM
cat audio.raw | ~/.her-voice/bin/her-voice-viz --stream --sample-rate 24000

### Disable Visualizer

python3 SKILL_DIR/scripts/config.py set visualizer.enabled false

### Configuration

Config file: ~/.her-voice/config.json

# View all settings
python3 SKILL_DIR/scripts/config.py status

# Get a value
python3 SKILL_DIR/scripts/config.py get voice

# Set a value (dot notation for nested keys)
python3 SKILL_DIR/scripts/config.py set speed 1.1
python3 SKILL_DIR/scripts/config.py set visualizer.mode classic

### Key Settings

KeyDefaultDescriptionagent_name""Agent's name (e.g. "Jackie")user_name""User's real nameuser_name_tts""Phonetic spelling for TTS (e.g. "Mah-toosh" for Matúš)voiceaf_heartBase voice namevoice_blend{af_heart: 0.6, af_sky: 0.4}Voice blend weightsspeed1.05Speech speed multiplierlanguageenLanguage codetts_engineautoTTS engine: auto, mlx, or pytorchmodelmlx-community/Kokoro-82M-bf16Model identifier (MLX)visualizer.enabledtrueShow visualizer overlayvisualizer.modev2Animation mode (v2/classic)visualizer.remember_positiontrueSave window position between sessionsnotification_sound.enabledtruePlay sound before speakingnotification_sound.soundBlowmacOS system sound namedaemon.auto_starttrueAdvisory flag only — the daemon never self-starts. When true, the agent should start it on first voice use (saves ~1s/call, costs ~2.3GB RAM)daemon.socket_path~/.her-voice/tts.sockUnix socket path

### Voice Blending

Mix multiple voices for a unique sound. Configure voice_blend in config:

{
  "voice_blend": {"af_heart": 0.6, "af_sky": 0.4}
}

The blended voice is stored as a .safetensors file in the model's voices directory (e.g., af_heart_60_af_sky_40.safetensors). Create it by running TTS once — speak.py looks for the pre-blended file automatically.

### Error Handling

ErrorCauseFix"mlx-audio not found"Venv missing or brokenRun setup.py"espeak-ng not found"Phonemizer missingbrew install espeak-ngCompilation failedXcode tools missingxcode-select --install"Model not found"First run, no downloadRun setup.py or speak onceDaemon "not running"Crashed or rebootedStart daemon againNo sound outputmacOS audio permissionsCheck System Settings → Sound → OutputVisualizer not showingBinary not compiledRun setup.py"kokoro not found"PyTorch venv missingRun setup.pyPyTorch CUDA errorGPU driver mismatchpip install torch --force-reinstall in kokoro venv"soundfile not found"Missing dependencypip install soundfile in kokoro venv

### Requirements

macOS + Apple Silicon recommended for best experience (MLX engine + visualizer + notification sounds)
Linux/Intel Mac supported via PyTorch Kokoro engine (no visualizer)
Windows is not supported
Xcode Command Line Tools for visualizer on macOS (xcode-select --install)
espeak-ng for phonemization (brew install espeak-ng on macOS, apt install espeak-ng on Linux)
~500MB disk (model + venv)
~2.3GB RAM when daemon is running

### Uninstall

Remove all Her Voice data (config, venvs, compiled binary, daemon state):

python3 SKILL_DIR/scripts/daemon.py stop
rm -rf ~/.her-voice

### How It Works

Kokoro 82M — A compact neural TTS model with two backends: MLX (Apple's framework for native Metal GPU acceleration on Apple Silicon) and PyTorch (works everywhere). The engine is auto-detected based on platform, or can be forced via the tts_engine config option (auto, mlx, or pytorch)
Streaming — Audio generates and plays simultaneously. First sound in ~0.3s (with daemon) vs ~3s batch
Visualizer — Native macOS app (Swift/Cocoa) reads raw PCM from stdin, plays via AVAudioEngine with real-time amplitude metering
Daemon — Unix socket server holding the model in RAM. Eliminates Python import + model load overhead on every call
## Trust
- Source: tencent
- Verification: Indexed source record
- Publisher: matusvojtek
- Version: 1.0.2
## Source health
- Status: healthy
- Item download looks usable.
- Yavira can redirect you to the upstream package for this item.
- Health scope: item
- Reason: direct_download_ok
- Checked at: 2026-05-01T20:30:11.280Z
- Expires at: 2026-05-08T20:30:11.280Z
- Recommended action: Download for OpenClaw
## Links
- [Detail page](https://openagent3.xyz/skills/her-voice)
- [Send to Agent page](https://openagent3.xyz/skills/her-voice/agent)
- [JSON manifest](https://openagent3.xyz/skills/her-voice/agent.json)
- [Markdown brief](https://openagent3.xyz/skills/her-voice/agent.md)
- [Download page](https://openagent3.xyz/downloads/her-voice)