A narrator from a sentence

Design a narrator from a written description and keep it. It is made on your Mac, with no recording and no provider.

Design a narrator voice in GroundSlate named Marlowe: a woman, adult, mid in her range, with a warm, unhurried documentary read. If designing a voice from a description is not available, say so rather than substituting. Attach it to avatar av_4d5e6f7a and read the first minute of ~/Documents/script.txt with it so I can hear it on real words.

Paste it into any MCP client

Describe the body first: gender, age, and where the voice sits in its range. The model responds to those; adjectives only add finish. A description that is all mood gets the same default voice every time, and naming an accent changes nothing, so leave it out.
Making a voice from a description needs the larger speech tier, a one time download on the app's AI Voice page. The light tier only makes voices from stock speakers, which is why the ask says what to do if the larger tier is missing.
One description makes one voice, and a voice's short preview is always the same fixed line. To hear several takes of a description and pick by ear, use the app. To hear it on your own words, the ask ends with a script read.
The voice is invented from the description, so there is no real person behind it, nothing to license, and nothing leaves your Mac.
Each read is identical on every render, so a re-render never shifts an edit timed to the words.
The name is fixed when the voice is made. Rename or remove a voice on the app's AI Voice page, which also has a library of ready made voices to audition and add.