Google Chirp 3 HD TTS Skill#

Overview#

Generate ultra-realistic, human-like speech using Google's latest Chirp 3 HD generative models. This skill handles its own dependencies locally to remain portable.

Security Note: On first execution, this skill will run npm install locally within its own folder to fetch the official @google-cloud/text-to-speech library from the public npm registry.

---

Trigger Detection#

Recognize keywords like "tts", "speak", "voice", or "read this out loud" as TTS requests.

• Action: Extract the target text, strip the trigger keyword, apply "Natural Speech" formatting (see below), then call the tool using the auto-run logic below.

---

Auto-Run Logic (Agent Instructions)#

Follow these steps every time TTS is needed. Never assume gtts is in PATH — always run via node directly.

Step 1 — Find the skill folder#

gtts.js lives in the same folder as this SKILL.md. Resolve that path:

bash

SKILL_DIR="$(cd "$(dirname "$0")" && pwd)"

Step 2 — Check Node.js (Version 18+ Required)#

bash

if ! command -v node &>/dev/null; then
    echo "ERROR: Node.js is not installed."
    exit 1
fi

NODE_MAJOR=$(node -v | cut -d'v' -f2 | cut -d'.' -f1)
if [ "$NODE_MAJOR" -lt 18 ]; then
    echo "ERROR: Node.js 18 or higher is required for Google Chirp 3 HD."
    exit 1
fi

Step 3 — Auto-install dependency if missing#

This installs the package locally inside the skill folder only.

bash

if [ ! -d "$SKILL_DIR/node_modules/@google-cloud/text-to-speech" ]; then
    npm install @google-cloud/text-to-speech --prefix "$SKILL_DIR" --silent
fi

Step 4 — Run the script#

bash

node "$SKILL_DIR/gtts.js" --text "$TEXT" --voice "$VOICE" --out "$OUTFILE"

---

Command Arguments#

ArgumentDescriptionDefault--textText to synthesize. Supports [pause] tags.(required)--voiceVoice short-name (e.g. Aoede, Charon, Puck)Aoede``--outOutput filename (saved to $OPENCLAW_WORKSPACE)output.mp3

Returns SUCCESS:/absolute/path/to/file.mp3 on success, or ERROR: ... on failure.

---

Voice Selection#

GenderRecommended HD VoicesFemaleAchernar (Default), Aoede, Leda, Kore, Zephyr, Despina, Gacrux, VindemiatrixMaleCharon, Puck, Fenrir, Orus, Achird, Algenib, Enceladus

---

Prompt Engineering for Natural Speech#

1. Pause Tags#

Converted automatically into SSML <break> tags:

TagDuration[pause short]300ms[pause]600ms[pause long]900ms

2. Human-Like Formatting#

• Contractions: Use "I'm", "don't", "can't" for a conversational tone.
• Ellipses: Use ... for trailing hesitation.
• Fillers: Use "Well,", "Um,", or "So," to mimic natural thought.

---

Authentication#

Uses Google Application Default Credentials (ADC). One-time setup:

bash

gcloud auth application-default login

---

Requirements#

RequirementStatusNode.js 18+❌ Must be installed on systemOPENCLAW_WORKSPACEℹ️ Optional (Defaults to current dir)@google-cloud/text-to-speech✅ Auto-installed locally in skill folderGoogle Cloud SDK + ADC login❌ One-time manual step required

---

Workflow#

1. Detect Intent — Identify a TTS trigger keyword.
2. Format Text — Apply contractions, ellipses, and [pause] tags.
3. Check Environment — Confirm node (18+) is available and OPENCLAW_WORKSPACE is known.
4. Auto-install deps — Run Step 3 if node_modules is missing.
5. Execute — node "$SKILL_DIR/gtts.js" --text "..." --voice "..." --out "..."
6. Confirm Output — Reference the SUCCESS: path and confirm to the user.

text