“Cheerful and upbeat, like a friendly station announcer.”
4.9 s · 2 credits
Real Gemini 3.8 Flash TTS output · 24.5 seconds · 6 credits
Gemini TTS voices both sides of the conversation
One request, two people. This podcast cold open was written as six lines of Name: text, and Gemini TTS returned a single file that switches between Maya and Theo on every turn, with a laugh where the script asked for one.
Maya voice KoreTheo voice Puck
- 0:00
MayaOkay, we're recording. Welcome back to Small Kitchen, Big Ideas.
- 0:04
TheoToday's topic is the humble pancake, and I have strong opinions.
- 0:10
MayaYou always have strong opinions.
<laugh>Go on. - 0:13
TheoRest the batter for ten minutes. That's it. That's the whole secret.
- 0:19
MayaTen minutes? That's the big reveal?
- 0:21
TheoTrust me. Lighter, fluffier, every single time.
Style instruction for the whole take: “Relaxed, warm podcast banter between two old friends.”
Same voice, same words, three directions
Steer Gemini TTS with one sentence of direction
Every take below reads the same line in the Sulafat voice. Only the style instruction changed. Pacing moved with it: the tired reading is two seconds longer than the cheerful one.
“The last train leaves at midnight. If you miss it, you're walking home.”
“Whispered and conspiratorial, as if sharing a secret.”
5.5 s · 2 credits
“Exhausted and deadpan, at the end of a very long night shift.”
6.9 s · 2 credits
30 prebuilt voices
Pick a Gemini TTS voice by character, not by name
Google describes each prebuilt voice in one word. Use that word to shortlist two or three, then let the style instruction do the rest. Highlighted voices are the ones you can hear on this page.
- AchernarSoft
- AchirdFriendly
- AlgenibGravelly
- AlgiebaSmooth
- AlnilamFirm
- AoedeBreezy
- AutonoeBright
- CallirrhoeEasy-going
- CharonInformative
- DespinaSmooth
- EnceladusBreathy
- ErinomeClear
- FenrirExcitable
- GacruxMature
- IapetusClear
- KoreFirm
- LaomedeiaUpbeat
- LedaYouthful
- OrusFirm
- PulcherrimaForward
- PuckUpbeat
- RasalgethiInformative
- SadachbiaLively
- SadaltagerKnowledgeable
- SchedarEven
- SulafatWarm
- UmbrielEasy-going
- VindemiatrixGentle
- ZephyrBright
- ZubenelgenubiCasual
No language setting needed
Gemini TTS in French and Japanese
Gemini TTS reads the language it is given. Both announcements use the same prebuilt voices as the English takes; the translation under each script is for you, not for the model.
FrenchCharon13.7 s · 3 credits
Bienvenue au musée. Dans cette salle, vous découvrirez des cartes marines dessinées à la main il y a plus de trois siècles. Prenez votre temps : chaque détail raconte un voyage.
Welcome to the museum. In this room you will find sea charts drawn by hand more than three centuries ago. Take your time: every detail tells of a voyage.
Style: Calm, knowledgeable museum guide, unhurried.
JapaneseAoede10.7 s · 2 credits
本日はご来店いただき、ありがとうございます。ただいま、焼きたてのクロワッサンをご用意しております。どうぞごゆっくりお選びください。
Thank you for visiting us today. Freshly baked croissants are ready now. Please take your time choosing.
Style: Bright, polite in-store announcement.
Flash or Flash Lite
Two Gemini TTS tiers, one cooking lesson
We sent the same lesson, voice Kore and instruction “Clear, encouraging cooking-class instructor” to both tiers. Both read every step in order, and the lengths differ by one second. Flash Lite costs about a third less, which makes it the sensible default for drafts and high-volume notification audio.
Step one: heat the oven to two hundred degrees. Step two: while it warms, cut the vegetables into even pieces, so everything roasts at the same speed. Step three: a little oil, a pinch of salt, and into the oven they go.
Gemini 3.8 Flash TTS
Gemini 3.8 Flash Lite TTS
| Measure | Flash | Flash Lite |
|---|---|---|
| This 220-character lesson | 4 credits | 3 credits |
| Length of the take | 16.4 s | 17.4 s |
| 100 characters | 2 credits | 2 credits |
| 1,000 characters | ≈ 16 credits | ≈ 11 credits |
| 5,000 characters (maximum) | ≈ 79 credits | ≈ 53 credits |
Credits count the spoken text only; style instructions are free. New accounts get 50 free welcome credits, about two dozen short Gemini TTS lines.
The form behind the dialogue
How to write a Gemini TTS script on Voor
These are the five fields in the generator above, filled in exactly as they were for the podcast sample.
- Start every new speaker line with the name and a colon. A line without a name continues the previous turn.
- Use the same two spellings throughout; "Theo" and "Theo (laughing)" count as different names.
- Put stage directions in the style instruction, not in the text, or they will be read aloud.
- Write numbers the way they should be heard. Our lesson spelled out "two hundred degrees".
- Check one short take before generating a long script, then keep the voices and instruction fixed.
- Text
Maya: Okay, we're recording…The exact words. Up to 5,000 characters; <laugh> and <sigh> work inline.- Style instructions
Relaxed, warm podcast banter…How to speak, never spoken. One sentence about mood, pace and listener.- Voice
KoreThe single voice, or the first name that appears in a dialogue.- Two-speaker dialogue
onReads Name: text lines as a conversation. Exactly two names.- Second speaker voice
PuckUsed for the second name. Ignored when dialogue is off.
What Gemini TTS will not do
- No voice cloning. Only the 30 prebuilt voices. For speech from a reference recording, compare cloning models such as Qwen3 TTS.
- Two speakers at most. A third name is rejected before anything is generated; split larger casts into scenes.
- No inline emotion tags. Direction lives in one style instruction per take. For a mood change on every sentence, ElevenLabs v4 follows bracketed audio tags instead.
- Audio only. To put the voice on a face, take the file to ; to compare every speech model, start at text to speech.