Which Korean voices does MAI-Voice-2.1 have, and with what styles?

Microsoft's MAI voices page lists four Korean voices: ko-KR-Grant (male) and ko-KR-Harper (female), each with six role styles, ko-KR-Haena (female) with 14 emotion styles, and ko-KR-Junho (male) with 12. All styles are set through SSML with mstts:express-as, as listed on the page read on 2026-10-04.

The four voices

Picking by job

For narration, a call-centre prompt or an educational read, the role-style voices (Grant, Harper) name the job directly. For an emotional ad read, Haena has the widest emotion list, and Junho is the male option with emotions, though it has no regretful or surprised style.

The page is in public preview with no SLA, so test the style you choose rather than assuming every style sounds equally natural.

After the voice: Korean captions in Sume

If the audio ends up on a video, Sume's Video captions endpoint burns captions for $0.20 per video up to 60 seconds. For Hangul text the default style is black-outline, and a korean-ad style is available. Sending a Hangul script with a Latin-only style returns caption_hangul_text_latin_style.

• cues skip speech-to-text when you already have timings.
• script_text aligns your own script to the audio.
• source_caption_id restyles an earlier caption job.

Where Sume's own TTS fits

Sume's TTS endpoint takes a plain transcript, no SSML, so mstts:express-as styles are not a Sume feature. Its catalog is Sonic models. Choose it when you want one HTTP call and a flat $0.0475 per 1,000 characters; choose MAI when you need those named Korean styles.