Korean Text to Speech: Speech Levels and Natural Flow

Korean text to speech reads Hangul aloud, but the speech level you write in changes how the audio sounds. Korean has seven speech levels, and the verb ending sets the level, so the same message can sound formal or intimate. Korean has about 81 million native speakers, according to the Wikipedia article on the language.

This guide explains speech levels, how numbers are read, why Hangul is not read letter by letter, which Korean voices the main providers list as of October 2026, and how to make Korean audio with OndaLabs.

Speech levels and why they matter for audio

Speech levels show respect to the listener. Honorifics are a separate system that shows respect to the person being discussed. The table lists the seven levels using the verb hada, "to do," as the sources do.

| Speech level | Ending for "to do" | Group | Usage note | |---|---|---|---| | hasoseo-che | hanaida | Formal | Very high formality; uncommon, mainly in historical dramas and religious speech | | hasipsio-che | hamnida | Formal | Common | | haeyo-che | haeyo | Informal | Common, polite and informal | | hao-che | hao | Formal | Less common | | hage-che | hane | Formal | Less common | | haera-che | handa | Formal | Common | | hae-che | hae | Informal | Common, intimate |

The sources group the seven levels as five formal and two informal (Korean speech levels). The level comes from the endings you write. No source says a voice switches register by itself, so check every ending in the script.

Pick one level for a single narrator and keep it. Sources show which levels are common, not which one suits every purpose. Write the same sentence in two levels, generate both and listen to the difference.

Numbers: native Korean or Sino-Korean?

Korean has two number systems, and each is used for different things (Korean numerals).

| Use | System | Example | |---|---|---| | Counting objects, people, hours and ages | Native Korean | 세 시 is 3:00; 스물다섯 살 is 25 years old | | Minutes | Sino-Korean | 삼십오 분 is 35 minutes | | Addresses, phone numbers and numbers above 99 | Sino-Korean | Phone numbers are normally Sino-Korean | | Age with 세 | Sino-Korean | 이십오 세 |

Phone numbers are normally read in Sino-Korean, but native 1 and 2 are sometimes used to avoid confusion. The number 555-1212 is read as o-o-o hana-dul-hana-dul.

Azure supports say-as types for Korean locales, and its math reading covers ko-KR. No source here says how a provider chooses between the two systems for a given digit string. Spell the number out in Hangul when the system changes the meaning, then check the result in a test.

Sound changes: why Hangul is not read letter by letter

Hangul is built from 24 basic letters and 27 complex letters. King Sejong created it in the 15th century to promote literacy. Spelling does not match sound in many places, because consonants change in running speech (Korean language). The main changes are:

Plosives are voiced between voiced sounds.
Word-final obstruents are unreleased.
Plosives become nasals before nasal sounds.
The l sound becomes a flap between vowels.
The m and n sounds often lose their nasal quality at the start of a word.

This research found no benchmark of how any engine applies these changes. Test sentences that contain many plosives, nasals and l sounds before you publish. Azure's Korean phoneme control uses IPA, with symbols such as kʰ and tʰ for aspirated consonants (Azure phonetic sets).

Korean voices across providers

| Provider | Korean voices listed | Notes | |---|---|---| | Azure | 16 ko-KR entries, including HD voices SunHi and Hyunsu, nine standard voices, HyunsuMultilingual, and MAI-Voice-2 voices Haena and Junho | Each MAI-Voice-2 voice has a Flash version | | Amazon Polly | Seoyeon (generative, neural and standard) and Jihye (neural) | Both voices are female | | Google Cloud | About 41 ko-KR voices (approximate), including 30 Chirp3-HD | Chirp 3 HD supports ko-KR with IPA or X-SAMPA custom pronunciation | | ElevenLabs | Korean on Multilingual v2, Flash v2.5, v3 and v4 | Also supported for dubbing |

Counts are as of October 2026.

ElevenLabs lists character limits that differ by model: 40,000 characters on Flash v2.5, 10,000 on Multilingual v2 and v4, and 5,000 on v3. Check the limit for your model before you paste a long script. On YouTube, English videos can be dubbed into Korean, and Korean videos can be dubbed into English. Auto-dubbed videos cannot be edited directly, so correct the original video instead.

Script tips for natural Korean flow

Choose one speech level for each narrator and check every verb ending.
Write numbers in the system you want spoken, and spell them out in Hangul when the system matters.
Decide how an English word or abbreviation should sound, and write that reading in Hangul.
Break long passages into paragraphs, generate them one at a time, and listen to the joins.
For a two-speaker script, use the dialogue tool so each line has its own voice.

Making Korean audio with OndaLabs

Open the text to speech studio and paste your Korean script.
Pick one of the 30 prebuilt voices, which work in every supported language. Korean has its own page: Korean text to speech on OndaLabs.
Choose Onda Lite at 1 credit per character, or Onda Flash at 1.5 credits per character on paid plans.
Describe the delivery in plain words and generate a short test. Compare two speech levels side by side.

For a conversation between two speakers, the dialogue tool handles two voices in one script.

A 1,000-character Korean script costs 1,000 credits on Onda Lite and 1,500 on Onda Flash. The Free plan is $0 and includes 2,000 credits a month, about 2 minutes. Starter is $9 a month and includes 170,000 credits, about 189 minutes, Onda Flash and commercial use. See the pricing page for all plans.

For Mandarin, read Chinese text to speech. To dub a full video, see AI dubbing explained.

Frequently asked questions

Which Korean speech level should I write my script in?

Choose the level that fits your audience and use it throughout. Haeyo-che and hasipsio-che are both common, but sources describe which levels are common, not which one suits every purpose. Generate a sample in each level and listen before you decide.

Does Korean text to speech handle sound changes?

The sound changes described here are standard Korean phonology, but this research found no benchmark of how any engine applies them. Test sentences with many plosives, nasals and l sounds before you publish. Azure offers IPA phoneme control for Korean.

How do Korean AI voices read numbers?

Native Korean numbers count objects, people, hours and ages. Sino-Korean numbers cover minutes, addresses, phone numbers and numbers above 99. Spell the number out in Hangul when the system changes the meaning, and check the result in a test.

Is there a free Korean text to speech tool?

OndaLabs has a Free plan at $0 with 2,000 credits a month, about 2 minutes of audio on Onda Lite. That is enough for a short test. Check the current terms of other tools before you rely on their free plans.

Can I use Korean AI voices commercially?

OndaLabs includes commercial use on Starter and above, from $9 a month. Check each provider's terms before you publish. OpenAI's usage policy, for example, requires clear disclosure to end users that a TTS voice is AI-generated.