Kid voice for the Dex language creative

One child voice says Earth · Terre · 地球 · Tierra — exactly one second apart.

5 finalists out of 15 candidates tried, across two engines. Listen on headphones; the differences are in the consonants.

What to listen for

Japanese is where these break. Of 15 candidates, 5 said 地球 wrong — including one that said だいすき ("I love you"). Four of the five purpose-built child voices failed it. Every voice below says it correctly, which is most of why they're here.

The 1-second spacing is exact, not model-guessed — the words are cut and spaced in ffmpeg, verified by arithmetic on every file. Stems are published separately so the editor can re-time to picture.

Finalists

Top pick

Boy ~7 · curious, warm

MiniMax · falpurpose-built child voice
EarthEnglish
TerreFrench · la Terre
地球Japanese · chikyū
TierraSpanish · la Tierra
#2

Sweet_Girl_2 · stock

MiniMax · falstock voice
EarthEnglish
TerreFrench · la Terre
地球Japanese · chikyū
TierraSpanish · la Tierra
#3

Decent_Boy · stock

MiniMax · falstock voice
EarthEnglish
TerreFrench · la Terre
地球Japanese · chikyū
TierraSpanish · la Tierra
#4

Aoede · kid ~7

Gemini TTSadult voice prompted young
EarthEnglish
TerreFrench · la Terre
地球Japanese · chikyū
TierraSpanish · la Tierra
#5

Achernar · girl ~8

Gemini TTSadult voice prompted young
EarthEnglish
TerreFrench · la Terre
地球Japanese · chikyū
TierraSpanish · la Tierra

Why voice choice isn't cosmetic here

A stock voice saying だいすき instead of 地球

Same prompt, same pipeline — a voice we didn't vet. For a product whose promise is "hear its name in a new language," a wrong word is the one defect the audience is guaranteed to catch. Every finalist above was transcribed back and checked per language before it made this page.

How they scored

VoiceEngineTypeScoreClean wordsLength
Boy ~7 · curious, warm MiniMax · fal purpose-built child voice 36/36 4/4 5.72s
Sweet_Girl_2 · stock MiniMax · fal stock voice 36/36 4/4 4.88s
Decent_Boy · stock MiniMax · fal stock voice 36/36 4/4 4.77s
Aoede · kid ~7 Gemini TTS adult voice prompted young 36/36 4/4 7.13s
Achernar · girl ~8 Gemini TTS adult voice prompted young 36/36 4/4 5.56s
Lovely_Girl · stock MiniMax · fal stock voice 30/36 3/4 6.08s
Leda · girl ~4 Gemini TTS adult voice prompted young 30/36 3/4 7.01s
Zephyr · girl ~5 Gemini TTS adult voice prompted young 30/36 3/4 6.22s
Puck · boy ~6 Gemini TTS adult voice prompted young 30/36 3/4 5.41s
Girl ~5 · bright, delighted MiniMax · fal purpose-built child voice 24/36 3/4 6.08s
Girl ~8 · confident big-kid MiniMax · fal purpose-built child voice 24/36 3/4 5.48s
Boy ~6 · excited, giggly MiniMax · fal purpose-built child voice 24/36 3/4 5.58s
Exuberant_Girl · stock MiniMax · fal stock voice 24/36 3/4 4.98s
Autonoe · girl ~6 Gemini TTS adult voice prompted young 24/36 3/4 6.08s
Girl ~4 · soft, wondering MiniMax · fal purpose-built child voice 18/36 2/4 6.15s

Every clip was transcribed back by an independent model and scored on three axes: is it the right word, is the pronunciation native, does the speaker read as a child aged 4–8. Machine scoring is a filter, not a verdict — it disagreed with itself once on Aoede's English, so trust your ears over the table.