MAI-Voice-2.1 Mandarin: six zh-CN voices, from 6 to 19 styles

Microsoft lists six zh-CN voices for MAI-Voice-2.1: Bo, Grant and Wei are male, and Harper, Lan and Mei are female. Their style lists range from 6 items (Grant, Harper) to 19 (Bo, Mei). For an expressive Mandarin ad, Bo and Mei are the two to audition first (read 2026-10-07).

The six voices

I counted the style column for zh-CN in the voice table on Microsoft Learn. Three lists are long, one is mid-length and two are the role set.

Which voice for which job

Four voices list excited: Bo, Lan, Mei and Wei. Only Bo and Mei list whispering and shouting, which matters for ASMR-style product videos and for sale-countdown reads. Grant and Harper are the role voices for a service line, an explainer or an audiobook, and they have no emotion list.

The split is even, three men and three women, but the expressive pair is one of each, so a campaign that wants both a male and a female lead can hold the style set constant across the two.

• Expressive ads: Bo (male) and Mei (female), 19 styles each.
• Calm explainers and courses: Grant and Harper, with educational and narrator.
• A mid-range option: Wei with 14 or Lan with 11.

Chinese on Sume

Sume's TTS has no style list, but it takes a language field. Send zh for Chinese. It also accepts an emotion string of up to 64 characters and a speed between 0.6 and 1.5. The price is $0.0475 per 1,000 characters, with a minimum of one cent a job.

Chinese scripts count characters, not words, so a Mandarin line is short in characters. A 30-second read of about 150 Chinese characters (a rough planning figure, not a measured one) is 0.15 x $0.0475 = $0.007, which bills 1 cent. The same ad on Flash is 150 x $15 / 1,000,000 = $0.00225. Both are rounding noise, so choose by sound.

Audition without wasting time

Use one real line, with numbers and a brand name in it, for all six voices. On Sume that is six jobs and six cents. Keep the same speed on each so you compare the voices, not the pacing.

If Mandarin is not the only language, remember that Microsoft lists zh-CN only. Cantonese and Traditional Chinese variants do not appear in the locale list I read, so do not assume them. Ask the vendor before you plan a Hong Kong or Taiwan release on this model.

Numbers and names in Mandarin

Ads are full of prices, dates and Latin-letter brand names, and these are the places Mandarin voices differ most. Put three such items in your audition line: an amount with a decimal, a date, and your brand name in English letters. Listen for how each voice says them, and write the accepted reading into your script (for example, spelling a number out in Chinese characters) so the next render does not change it.

Sume's TTS accepts a pronunciation dictionary id for exactly this case, so a fix that works once can be applied to every later job.