Developers

Hmong text-to-speech (two voices) and speech-to-text for your apps — API keys

Why our Hmong model

Hmong is spoken by roughly 4 million people worldwide and supported by essentially no commercial speech platform. We built Hmong speech recognition and voice synthesis from scratch, trained on original studio recordings from native speakers, transcribed and verified by hand in RPA orthography.

What we provide

Hmong Text-to-SpeechFive distinct voices — male and female

Hmong Speech-to-TextPurpose-trained; general models don't support Hmong

Conversational Voice AgentsFull speak-listen-respond loop in Hmong

English ⇄ Hmong Translation + SpeechTranslate and hear it spoken

REST APIJSON in, audio out; API keys and usage metering

Why a purpose-built model matters

Hmong is a tonal language with eight tone marks written as trailing consonants. Generic multilingual models — where they attempt Hmong at all — treat these as spelling and produce the wrong word entirely. Pob (ball), Poj (female), and Pov (a name) differ only by that final letter.

Capability	Generic multilingual AI	Our Hmong model
Hmong speech recognition	Not offered	Purpose-trained
Hmong voice synthesis	Not offered	Five native voices
Tone-mark handling	Misreads as spelling	Trained on RPA orthography
Foreign-word handling	Mispronounced	Phonetically adapted to Hmong

Where it's used

Built for community organizations, clinics, schools, and public agencies serving Hmong-speaking families — voice announcements, translated instructions, spoken-language accessibility, and conversational assistance for elders more comfortable listening than reading.

How to evaluate it

Don't take numbers on a page. Send us a sample of your own Hmong audio or text and we'll run it through the live API and return the results, usually the same day. API keys are free.

Your API Keys

Create an account to get your API keys and start using Hmong Voice Agent.

Create an account

Text to Speech

Send Hmong text and get back the audio as base64-encoded WAV in the audio_base64 field. Maximum 2,000 characters per request. Choose a voice with the optional voice field: maivyeeb (female, default), paj (female), nkauj (female), tub (male) or pov (male).

Speech to Text

Send recorded Hmong speech as base64-encoded 16-bit mono PCM audio (with its sample_rate) and get back the transcript in the text field. Transcribed characters count against your monthly quota.

Hmong, Xieng Khouang (Laos) traditional clothing — Hmong,
Xieng Khouang (Laos)

Hmong, Luang Prabang (Laos) traditional clothing — Hmong,
Luang Prabang (Laos)

Hmong, Guizhou (China) traditional clothing — Hmong,
Guizhou (China)

Hmong, Sichuan (China) traditional clothing — Hmong,
Sichuan (China)