SpeakScribe has two Japanese voices, the Global voices Nanami (female) and Keita (male). The five Studio voices (Alpha, Gongitsune, Nezumi, Tebukuro and Kumo) were retired in October 2026. Japanese mixes three scripts, so the spelling choices below matter as much as the voice you pick.
How Japanese is written, and what to type so the voice reads it right
Japanese mixes kanji, hiragana and katakana in one sentence, and many kanji have more than one reading. Type the text as it would appear in print, using 。 and 、 for pauses. The voice reads the characters you write, so the reading you want must be the one in the text.
Furigana (ruby text) is not supported. Text copied from web pages often arrives as 漢字(かんじ), and both the kanji and the reading are spoken, so delete the bracketed readings before you generate.
Write the reading you want in hiragana for words with two common readings: 今日 (きょう or こんにち), 明日 (あした or あす), 上手 (じょうず or うわて) and 一日 (ついたち or いちにち).
Names have many possible readings. 中田 can be なかた or なかだ, so write personal and place names in kana where the exact reading matters.
Write foreign words in katakana with the long-vowel mark, such as コンピューター, and avoid romaji in the script.
Keep one speech style for the whole script. です/ます sounds polite and friendly, while the plain だ/である style is common in written explanations. The voice reads the endings you write.
Numbers, dates and loanwords in Japanese text
Arabic digits are common in Japanese text, but kanji numerals are clearer when the reading matters. Counters attach to the number and change with the object, as in 三本 (three long objects) and 三人 (three people), so write the counter together with the number.
Katakana carries most loanwords and brand names. Write them in the katakana form Japanese speakers use, such as テレビ and インターネット, rather than in English spelling.
Write dates as 2026年10月22日 and times as 三時十五分 when the reading must be exact.
Write amounts with 円 after the number, as in 千円.
Write 二〇二六年 when you prefer kanji numerals for a year.
Accents and regional varieties
Japanese has regional dialects, such as the Kansai and Kyushu varieties, that differ in vocabulary, sentence endings and accent. The accent field of both voices is empty, so the library does not label any of them as a dialect speaker.
Write standard Japanese for these voices. If a script includes dialect words, preview that line before you publish.
Choosing between the Japanese voices
The two Global voices are Nanami (female, 264.4 Hz, 9.9 semitones) and Keita (male, 140.2 Hz, 9.2 semitones). Nanami sits about 124 Hz above Keita in median pitch, so the two make a clear female and male pair. Test one sentence with a difficult kanji in two voices before you commit.
The five Studio voices, Alpha, Gongitsune, Nezumi, Tebukuro and Kumo, were retired in October 2026. In a speech-recognition check, 41–49% of the characters they read came back wrong, against 2% for Nanami and Keita. Their old pages now lead to Nanami (female voices) or Keita (Kumo).
Common mistakes when turning Japanese text into speech
Japanese errors usually come from copied text, mixed character widths and dropped kana. Check these before generating.
Dropping okurigana, the hiragana that follow a kanji stem. 食べる must stay 食べる; 食る is not a word.
Writing small kana at full size, as in きや for きゃ. Small ゃ, ゅ, ょ and っ change the sound of the syllable.
Mixing full-width and half-width punctuation. Use 。 and 、 in Japanese sentences instead of . and ,.
Mixing full-width and half-width digits in one sentence, such as 123 and 123. Pick one width for the whole script.
Typing product names in romaji where Japanese uses katakana, such as ユーチューブ for YouTube. Katakana is how the name appears in Japanese text.
Japanese voices compared — measured
Each voice read the same Japanese text at normal speed. Lower pitch sounds deeper; a wider intonation range sounds livelier, a narrower one steadier.
Measured by SpeakScribe in October 2026 from each voice's audio. Pitch is the median fundamental frequency; intonation range is the spread between the 10th and 90th percentile of pitch. Speaking rate is characters per second, spaces included, on the ~15-second sample passage that every Japanese voice reads. st = semitones. Methodology · How SpeakScribe voices work