We're a small team at Deep Origin, and one of the tools we build is AIVoiceBooth, a text to speech and voice cloning tool. This post is a short, practical walkthrough of how we think about getting good voice-over out of it, for videos, audiobooks and ads.

You can try it at https://aivoicebooth.com/

What it does, briefly

  • Text to speech in 32 languages, studio-grade, ready for videos, audiobooks and ads.
  • Character voices. The voice library is a cast of characters, from a folk storyteller to a cold CEO. You tap a portrait to audition a voice before using it.
  • Voice cloning from a short clip. Upload a short recording of your own voice; the clone keeps your tone and rhythm, then speaks any of the 32 languages.

A simple voice-over workflow

  1. Write for the ear, not the eye. Short sentences read better aloud. Spell out abbreviations the way you want them spoken.
  2. Audition before you commit. Try a few character voices on the same paragraph. A narrator that suits a product demo is rarely the one that suits a bedtime story.
  3. Generate in sections. For long scripts (an audiobook chapter, a 10-minute video), work paragraph by paragraph so a single fix doesn't mean regenerating everything.
  4. Line it up in your editor. Drop the audio into your video or audio editor and trim the gaps to match your cuts.

Getting a clean voice clone

The clone can only be as good as the clip you give it:

  • Record in a quiet room with soft furnishings; avoid echo and background music.
  • Keep a steady distance from the microphone.
  • Speak at your natural pace. The clone keeps your tone and rhythm, so a rushed sample gives a rushed clone.

One voice, several languages

Because a cloned voice can speak any of the 32 supported languages, you can keep the same voice across localized versions of an ad or a tutorial instead of casting a different speaker per market.

Questions or feedback are welcome in the comments.