GPT-Live Speak

Speak aloud through a Linux/ALSA audio device using OpenAI GPT-Live. Designed for embodied agents, robots, and other Linux systems with a speaker.

Speak

Use this skill when the user explicitly asks the agent to speak aloud, say something through its physical body, or when another approved workflow requires a spoken announcement.

Do not automatically speak every normal chat response.

Run the bundled speech tool using the exec tool.

Set:

text

SPEAK_TEXT = the message to communicate aloud

Then execute:

text

python3 {baseDir}/scripts/run.py "$SPEAK_TEXT"

The tool uses OpenAI GPT-Live and streams PCM audio to ALSA.

The default GPT-Live voice is Cedar. The voice, ALSA device, and playback buffer can be configured through environment variables.

GPT-Live may naturally paraphrase the supplied commentary slightly.

Keep spoken announcements reasonably concise.

Never speak passwords, API keys, authentication tokens, or other secrets.

If speech fails, report the failure. Do not retry repeatedly without understanding the cause.