Haitian Creole speech data.

Recorded by Kreyòl speakers in Haiti and abroad. Transcribed and checked by people. Built for AI companies that want their products to work in Kreyòl.

Contact us See how it works

First collection underway. Now taking requests for evaluation sets and custom collections.

M pral nan mache a talè.

“I’m going to the market soon.”

One sentence. One speaker. Checked by two reviewers.

What we provide.

More than twelve million people speak Haitian Creole. Most voice AI still can’t understand it, because the data that exists is old, narrow or unconsented. We collect the Kreyòl people speak today.

Evaluation sets

A few carefully checked hours, balanced across regions, to measure how well your model handles Kreyòl.

Training data

Natural speech and read speech, with human transcripts in the official Kreyòl spelling.

Custom collections

Your topics, regions or scenarios, such as medical conversations or customer calls.

Voice data

For text-to-speech, only from speakers who said yes to voice generation.

How it works.

Every recording passes four steps before it counts. Nobody checks their own work.

  1. Record

    Uncompressed audio from the speaker’s phone, with noise filters off.

  2. Check

    Automatic checks for length, silence, clipping and duplicates.

  3. Transcribe

    A reviewer checks the audio is clear and on topic, then writes exactly what was said.

  4. Validate

    A second reviewer compares the transcript with the audio.

We re-check a random 5% of finished recordings ourselves.

Coming soon

Do you speak Kreyòl?

Piplis pays Haitian Creole speakers to record their voice with their phone. We also need people who read and write Kreyòl well to review recordings.

In Haiti, payment by MonCash or NatCash. Payment for people abroad is coming soon.

Where do you live?

Let’s talk.

Datasets, evaluation sets, custom collections or partnerships.

Jehud Neri Sabbat, founder