A structured speech and linguistic archive documenting how African languages are actually spoken: pronunciation, intonation, prosody, speaker variation and, over time, dialectal variation.

Explore a language →Need data like this? →

African languages need speech infrastructure, not just text.

A language can have dictionaries, translated text and even a large language model, and still be poorly represented in speech technology. ALPIA is designed to close that gap.

What ALPIA is

Three things at once

An archive

A growing collection of authentic recordings from speakers of African languages.

A dataset

Structured, machine-readable speech and linguistic data for research and AI development.

An infrastructure layer

A standardised way of representing pronunciation and intonation across African languages.

The data model

Three layers, one entry

Speech

Words, phrases, read speech, spontaneous speech, conversation and storytelling. Controlled and naturalistic recordings, side by side.

Pronunciation

Orthographic form linked to audio, phonetic representation, speaker, region and variant, so recordings become usable for technology, not only listening.

Intonation & prosody

Pitch, rhythm, phrase boundaries and emphasis: how meaning moves through a sentence, not only which sounds a language contains.

Explore a language

See what an ALPIA entry looks like

🇿🇦 isiZulu

1,284

Speakers

24,600

Recordings

KwaZulu-Natal

Lead region

v0.5

Archive version

▶

Sawubona

Hello. Recorded by three speakers, KwaZulu-Natal and Gauteng.

Coverage by phase

Phase 1: South Africa

isiZuluisiXhosaSesothoSetswanaSepediAfrikaans

Phase 2

KiswahiliYorubaHausa

Coverage is not identical across languages. Depth follows where the data gaps are greatest.

Built with speakers, not only about them

Contributor roles

Native speakers

Record their language.

Language specialists

Verify language quality.

Transcribers

Transcribe recordings.

Validators

Check linguistic accuracy.

Linguists

Guide annotation and method.

Join & Contribute →

Access

ALPIA is layered, not all-or-nothing

Public

Archive catalogue, methodology, language profiles and sample recordings.

Research

Larger datasets and richer annotations, available under controlled access.

Commercial

Specialised and large-scale collections, custom datasets, API and model-training licences.

Every dataset carries an explicit licence and access status. No part of ALPIA should be assumed freely reusable without checking.

ALPIA