Automate Voice-Overs With AI: Turn a Folder of Scripts Into Audio

You have a folder of scripts. Maybe it is twenty lesson texts, a stack of product descriptions or a season of short video scripts. You want an audio file for each one. Doing that by hand means opening each script, pasting it, picking the voice, pressing Generate and saving the file. Twenty times.

You can hand that loop to an AI agent instead. The agent reads the folder, sends each script to Clonylab through MCP, waits for each job and saves each MP3 under a clear name. You write one request. The agent repeats the steps.

This guide shows how to set that up, what the agent does at each step, and the limits that decide how big a batch can be. It builds on the text to speech MCP server guide, so connect your AI app first. Then come back here.

Keep the layout simple. One script per file, named so the MP3 can take the same name. Plain text or Markdown files both work, because the agent reads them as text.

lessons/
  01-welcome.txt
  02-setup.txt
  03-first-project.txt
  04-troubleshooting.txt
  05-next-steps.txt
audio/            <- the agent saves the MP3 files here

If you want different voices for different files, add a small list. A text file with one line per script is enough:

01-welcome        US female, calm
02-setup          US female, calm
03-first-project  UK male, warm
04-troubleshooting UK male, warm
05-next-steps     US female, calm

Script tags go inside the files. [pause 1s] after a heading, [slow] for a key sentence and [VoiceName] to change voice partway through all work the same way as in the Studio.

A clear request names the folder, the voice rule, the output folder and what to report. Here are three examples, from simple to careful.

Your Clonylab tab keeps a queue of jobs. The agent sends a script, the tab starts a job and the job moves through its steps. When it is done, the agent fetches the MP3. The job statuses are the real ones: queued, downloading, generating, done, error and cancelled.

We do not give a speed in minutes per script, because it depends on your device and the length of the script. A short test run of two files tells you how your own computer behaves.

A batch has no fixed maximum, but several limits shape how you run it.

When the agent says it is finished, check three things. Count the MP3 files against your scripts. Open two or three and listen, one from the start, one from the middle and one near the end. Then ask the agent for a short report: file name, voice, length and any failures.

clonylab.com/connect

You can also ask for a usage summary. usage_summary shows characters, number of voice-overs and audio minutes over the last 7, 30 or 90 days, so you see what the batch used. On Pro and Team the character count is unlimited.

A short report is the easiest way to see what happened. Ask for it in the same request, so the agent keeps notes as it goes.

If part of a batch fails, for example because the tab was closed, ask the agent to run only the files that have no MP3 yet. The request that skips existing files makes this safe. Your History in Clonylab also lists every voice-over that finished, so nothing is lost.

The batch needs the connection from the first guide. If you have not done it yet, switch on the browser permission and keep the tab open.

clonylab.com/connect

clonylab.com/connect

Step-by-step guides: Claude Code, Codex CLI, Cursor and Gemini CLI.

For one clip you want to hear and adjust, the Studio is better. For a folder, let the agent do the loop. The text to speech MP3 page explains the files you get.