Use the bundled scripts to create one stable local text-to-voice path.

Goal

Produce voice from any text file with:

• one canonical input text path,
• one canonical output mp3 path,
• one high-level wrapper for routine use,
• low-level scripts for debugging only.

Canonical paths

Default input text:

• tmp/text-to-voice-input.txt

Canonical output:

• tmp/voice-mode-latest.mp3

State directory:

• skills/text-to-voice-local/state/

State pointer:

• skills/text-to-voice-local/state/last-output.txt

Main command

For normal use, run:

• scripts/text_to_voice.sh voice [voice] [max_direct_chars]

Useful helpers:

• scripts/text_to_voice.sh status
• scripts/text_to_voice.sh voices

status now also checks runtime dependencies and prints install hints when something is missing.

Examples:

scripts/text_to_voice.sh text
scripts/text_to_voice.sh voice ./tmp/text-to-voice-input.txt
scripts/text_to_voice.sh voice ./tmp/text.txt ru-RU-SvetlanaNeural 280

What the scripts do

• scripts/text_to_voice.sh

- high-level entrypoint for normal use

• scripts/tts_from_file.sh

- one text file to one mp3

• scripts/tts_from_file_chunked.sh

- long text to multiple chunks and final merged mp3

• scripts/voice_reply.sh

- safe wrapper that updates canonical output and pointer

• scripts/voice_reply_latest.sh

- always refresh canonical latest mp3

• state/text-to-voice.json

- stores default voice, max chars, and canonical paths

• scripts/edge_tts.js

- low-level TTS helper used by the file wrappers

Install notes

Ensure these dependencies exist on the target machine:

• node
• ffmpeg
• node-edge-tts

The skill checks these at runtime and, if something is missing, prints suggested install commands instead of failing silently.

Verify:

node -v
ffmpeg -version
node -e "require('node-edge-tts'); console.log('node-edge-tts ok')"

If node-edge-tts is missing:

npm i -g node-edge-tts

Setup steps on another OpenClaw

• Copy this skill folder into the target workspace skills directory.
• Make scripts executable.
• Ensure tmp/ exists.
• Put text into a txt file.
• Run the main command.

Minimal setup:

chmod +x skills/text-to-voice-local/scripts/*.sh
mkdir -p tmp
skills/text-to-voice-local/scripts/text_to_voice.sh voice ./tmp/text-to-voice-input.txt

Delivery rule

If the result is sent as Telegram voice, send only the canonical file:

• ./tmp/voice-mode-latest.mp3

Prefer sending text and voice as separate messages.

Important constraint

Progress printed by shell scripts is useful in terminal diagnostics, but chat-side live progress editing depends on OpenClaw preview streaming, not shell stdout alone.

When to use low-level scripts

Use low-level scripts only for debugging or careful manual control. Default to the high-level wrapper unless there is a reason not to.

scripts/text_to_voice.sh text
scripts/text_to_voice.sh voice ./tmp/text-to-voice-input.txt
scripts/text_to_voice.sh voice ./tmp/text.txt ru-RU-SvetlanaNeural 280

node -v
ffmpeg -version
node -e "require('node-edge-tts'); console.log('node-edge-tts ok')"

npm i -g node-edge-tts

chmod +x skills/text-to-voice-local/scripts/*.sh
mkdir -p tmp
skills/text-to-voice-local/scripts/text_to_voice.sh voice ./tmp/text-to-voice-input.txt