namu
namu-ai.orgHear Aïr

In development

Build voice products for the Sahel on Namu’s platform

One API for the Lab’s speech models: hear, speak, translate and hold a conversation in the languages of the Sahel.

Talk to usRequest early access
  • Speech recognition
  • Speech synthesis
  • Speech translation
  • Live conversation
  • Languages of the Sahel

Powered by the Lab’s speech models

Each one comes from the research on this site, and each will be released with a model card.

Tatsuniya

Task
Speech recognition
Works with
Speech in, text out
Languages
Languages of the Sahel
Status
In development
Learn more

Tafiya

Task
Speech synthesis
Works with
Text in, speech out
Languages
Languages of the Sahel
Status
In development
Learn more

Kora

Task
Speech translation
Works with
Speech in, translation out
Languages
Languages of the Sahel, French
Status
In development
Learn more

Aïr

Task
Live conversation
Works with
A spoken conversation
Languages
Languages of the Sahel
Status
In research
Learn more

Pricing, rate limits and data terms will be published with each model’s release.

Get ready to build with speech from the Sahel

How we score speech

Every result is reported twice: strict, and folded for spelling variants. Learn to read both.

Read the method

Know your test set

Who is, and is not, in the public test set that most results are measured on.

Read the analysis

Hear a conversation

Listen to Aïr, the Lab’s research system for holding a spoken conversation.

Hear Aïr

One platform for voice in the Sahel

Listen, speak and converse through one API, with the same measurement behind every model.

Read the researchRequest early access
transcribe.sh
curl api.namu-ai.org/v1/transcribe \
  -H "Authorization: Bearer $NAMU_API_KEY" \
  -F model="tatsuniya" \
  -F file="@kira.wav"
{
  "text": "Sannu, ina kwana?",
  "audio_seconds": 2.4
}

Listen

Turn speech into text you can search, caption and act on.

Illustrative. The API is not public yet, and names may change before release.

Speak

Turn writing into natural speech, for apps, phone lines and devices.

Converse

Hold a live spoken conversation, or translate speech between the languages of the Sahel and French.

What you can build with it

For teams that need to reach people in the language they speak.

Customer service

Voice agents that answer customers in their own language, by voice, at any hour.

Talk to us

Captions and transcripts

Turn local radio, video and meetings into searchable text.

Talk to us

Education

Lessons, reading support and tutoring that students can speak to.

Talk to us

Information lines

Health and farming information that people can ask for out loud, without reading or typing.

Talk to us

Translation

Carry content between the languages of the Sahel and French, both ways.

Talk to us

Accessibility

Products that work for people who would rather speak than read and type.

Talk to us

Built the way the Lab works

The commitments behind every model the API will serve.

Measured honestly

  • A model card with every release
  • Every score reported strict and folded
  • Test sets frozen before training, with no speaker in both
  • Results published with the conditions they were measured in

Built on consented data

  • Speakers told, in their language, how their voice will be used
  • Speakers choose the licence for their recordings
  • Speakers paid for their time
  • A way to withdraw from every release

Made for the Sahel

  • Written in the orthography each country uses
  • Speakers from Niger, Nigeria and across the region
  • Test sets split by dialect and by gender
  • New languages added as each programme starts

Get started

Partner with us

  • Evaluate the models in your own setting
  • Contribute speech through the Rumbu programme
  • Help decide what we release first
Talk to us

Build on your own

  • Early access as each model is released
  • A model card and evaluation for every release
  • Code from the Lab on GitHub
Request early access