What Google AI Studio text to speech can do
Google AI Studio is Google’s browser workspace for trying Gemini models. Its speech generation mode turns text into audio with a choice of prebuilt voices, in single-speaker or multi-speaker mode. You can steer delivery with plain instructions, such as asking for a line in a tired whisper or with rising excitement, and the model follows them well.
Multi-speaker mode is what makes it interesting for fiction: you label lines with speaker names, assign a voice to each and the model performs the exchange as a scene. At the time of writing that mode handles two speakers per request. Gemini TTS also speaks many languages, and the same models are available to developers through the Gemini API.
Voicing one chapter in Google AI Studio
- 1
Open speech generation
In AI Studio, choose the speech generation mode and pick a TTS model.
- 2
Choose speakers and voices
Use single-speaker mode for narration alone; for a dialogue scene, switch to multi-speaker and map two speaker names to voices.
- 3
Paste a section and direct it
Paste part of the chapter, label who says what and add a short style instruction at the top.
- 4
Run, listen, download
Generate the clip, listen through and download it. Re-run any section with a wrong reading.
- 5
Repeat and join
Work through the chapter section by section, then join the clips in order in an audio editor.
Where AI Studio runs out on a full-length book
None of these are flaws in the voices. They come from using a model playground as a production tool. Exact numbers change, so check Google’s current documentation.
- Two speakers per request. A scene with a narrator and three characters has to be split across passes, or voices doubled up.
- Input and output caps. Each request takes a limited amount of text and returns a limited length of audio, so one chapter usually needs several runs.
- Free-tier rate limits. Free use is capped and the caps can change. Paid API use raises them and is billed by usage.
- Data terms. Google’s terms for unpaid use may allow inputs and outputs to be used to improve its products. Read them before you paste an unpublished manuscript.
- No memory of your cast. Nothing ties a character to a voice across sessions; you keep that list yourself and re-enter it every time.
- Clips, not a book. You end up with many separate files. Joining, levelling, chapter markers and the cover are separate jobs.
Working around the two-speaker limit
The usual workaround is to split each scene into pairs: run the narrator with one character, then the narrator with the next, and stitch the clips in order. Another is to voice the narration in single-speaker mode and keep multi-speaker passes for dialogue. Both work, but every split is a place where timing, volume or tone can jump.
Across a 20 chapter novel that means hundreds of clips and a spreadsheet of who sounds like whom. It is a fine way to make a short story or a sample scene, and a slow way to make a book.
AI Studio by hand vs a full audiobook workflow
| Task | Google AI Studio | AI Audiobooks studio |
|---|---|---|
| Voices | Prebuilt Gemini voices | 70 standard voices from the Gemini Flash TTS family plus 67 premium ElevenLabs voices |
| Speakers in a scene | Two per request | A narrator plus up to 10 named characters and an “Other voices” role |
| Chapters | You split the text | Found from the EPUB contents, Word headings or chapter lines |
| Casting | You assign voices each session | Characters detected from dialogue tags and cast automatically |
| Directions | Plain-language style prompts | Bracketed notes such as [whisper] plus inline sound tags |
| Fixing a line | Re-run and re-splice the clip | Edit and regenerate that line only |
| Output | Separate clips | Chapter MP3s, a ZIP, or an M4B with chapters and cover |
| Cost | Free within limits, then API usage pricing | About $9 to $12 per finished hour on Standard voices |
Google’s features, limits and prices change. Check the current AI Studio and Gemini API documentation before you compare.
Publishing a Gemini-narrated audiobook
Before you sell anything, check two things. First, Google’s current terms for the Gemini API and AI Studio on how generated audio may be used, including commercially. Second, each store’s policy on AI narration: at the time of writing, ACX does not accept AI narration uploaded by authors, so tutorials that suggest putting Gemini audio on Audible lead you into a rule breach.
Stores such as Google Play Books and Kobo accept disclosed AI narration under their own rules, and your own website is always open. Label the narration honestly wherever you publish, and see where to sell AI audiobooks for the current picture.
An audiobook with Gemini voices, without the clip juggling
The Standard and Lite voice models here come from the Gemini Flash TTS family, so if you like how Gemini sounds you can keep that sound for a whole book. The difference is the workflow around it: you upload the manuscript, chapters and speakers are found for you, each character keeps one voice from the first chapter to the last, and you export a finished M4B.
Directions work the way you would expect: a bracketed note such as [whisper] or [like a late-night radio host] at the start of a line, and inline tags such as [sigh] and [laugh] on the Standard and Lite models. The multi-voice text to speech page lets you paste a scene and hear it cast, and AI voice emotion tags lists the tags.
