Japanese AI voiceover

The Japanese voices, with samples

Each sample reads the same line so you can compare them fairly.

The script the director writes

Japanese in the Language field commits the video to ja-JP voices and to a natively written Japanese script: narration in a natural spoken register and on-screen text in the compact headline style Japanese design uses. Politeness is steerable, desu-masu for most business videos, plain form when your brief describes something casual, and honorific keigo when you ask for it. Briefs in English are standard; the director takes facts, not sentences. Any Japanese you quote in the brief, a product name in katakana especially, is carried through exactly.

Eight voices and the pitch accent question

The full premium cast exists in ja-JP: Puck and Laomedeia for energy, Charon and Erinome for instruction, tutorials pinned to that pair by rule, Umbriel and Sulafat for warmth, Iapetus and Aoede for neutral reads. Pitch accent is handled natively by the voices, and where a homograph could read two ways, the fix is writing the intended word in kana so the reading is explicit. Scene-level re-casting works through the voice dialog, with the whole Google catalogue searchable, including ja-JP voices in other tiers.

Katakana is your pronunciation panel

There is no phonetic input anywhere in the product, but Japanese has one built into the writing system: respell any foreign name in katakana inside the narration and the voice reads exactly those syllables. It is the most reliable respelling mechanism of any language here, about half a credit per regenerated clip. The same trick pins readings of ambiguous kanji, swap the kanji for kana and ambiguity disappears. Captions display what you wrote, so a scene whose narration went full kana may want its caption toggled off in the voice dialog.

Density, timing and the sound bed

Japanese packs meaning densely, so spoken lines often run shorter than their English drafts, and a 15 second promo fits comfortably. When a line does run long, the scene stretches to the clip and dynamic pace floats the overrun across the next cut. Per scene: delay in seconds, speed steps 0.85x to 1.25x, and the narration box. Music ducks to 22 percent under the voice automatically. A typical clip uses about half a credit, billed by length, and an export bills its render time, usually a small amount.

Captions in this script

Japanese writes without spaces, and the captions handle it with a dictionary segmenter: each sentence, cut on the 。！？ punctuation, is split into real words joined without gaps in the pill, and the highlight walks them in time with the voice. Long unspaced sentences wrap on those word boundaries at roughly 22 characters a line. The glyphs come from the render host's fallback font, since no catalogue family ships kana or kanji, and that fallback can style kana and kanji slightly differently from each other; short sentences and one test render are the practical safeguards.

Questions people ask

Related pages