Buying AI training data from Nigeria? Email [email protected]Request a free sample

Yorùbá (yo) · Lagos, recording to order

Yorùbá speech data, with the tone marks your model needs.

Read and conversational Yorùbá from ID-checked speakers who consent on record, with transcripts that keep the tone marks and under-dots, and metadata on dialect and region.

Email [email protected] Request a dataset

Free 10-hour sample first. A person replies within one working day.

A Nigerian woman speaking into her phone at home

Where it is spoken

Yorùbá at a glance.

Spoken mainly in south-west Nigeria (Lagos, Oyo, Ogun, Osun, Ondo, Ekiti and parts of Kwara and Kogi) and in Benin and Togo, with many regional dialects alongside Standard Yorùbá.

Code: yo (ISO 639-1)
Script: Latin alphabet with under-dots (ẹ, ọ, ṣ) and tone marks (high, mid unmarked, low)
Tone marks: Full diacritics on request; many online texts drop them, which hurts ASR and TTS
Dialects: Standard Yorùbá by default; Ẹ̀gbá, Ìjẹ̀bú, Ọ̀yọ́, Èkìtì and others tagged when you need them

Why it is hard for models

What speech models get wrong.

Tone changes meaning

Yorùbá is tonal: the same letters with different tones are different words, so a model needs correctly tone-marked text.

Diacritics are often missing

Much Yorùbá text online has no tone marks, so clean, tone-marked transcripts are the scarce part.

Yorùbá–English mixing

Urban speakers switch to English mid-sentence, especially in Lagos; this must be transcribed, not filtered out.

ASR and TTS

Read prompts or conversation? Usually both.

For speech recognition (ASR)

For ASR we record many speakers on read prompts and conversation, with tone-marked transcripts so you can train with or without diacritics.

For text-to-speech (TTS)

For TTS we record a few speakers for longer, from a phonetically and tonally balanced prompt list, in quiet rooms, with voice-synthesis consent named explicitly.

Speaker diversity

A fair mix of real speakers.

Speakers across south-west states and dialect areas, to a quota you set
Age band and gender balance agreed up front
Every speaker ID-checked so one voice is one person
Topics from markets, family, transport, health and services

Recording spec

How each clip is recorded and delivered.

Audio: WAV 48 kHz, 16-bit, mono per speaker (one track each in two-person sessions)
Tasks: Read prompts (your list or ours), natural two-person conversation, scripted service dialogues
Transcripts: UTF-8, one per clip, speaker turns and non-speech events marked; second-pass checks on every batch
Metadata: Language, age band, gender, region, device, consent ID and consent version per clip
Consent: Recorded in the speaker’s own language; voice synthesis (TTS) named explicitly when it applies
Delivery: Weekly batches over signed links, with a manifest and a QA report

Open data first

Honest advice: what to try before buying.

NaijaVoices (open, non-commercial)

NaijaVoices is an open speech dataset of Igbo, Hausa and Yorùbá released under CC BY-NC-SA 4.0, which does not allow commercial use. Good for research and benchmarks; licensed data is for products.

Source: NaijaVoices dataset card, Hugging Face

Open sets exist; check their licences

Yorùbá is in open collections such as Mozilla Common Voice and several academic corpora. Use them to start; come to us for licensed, consented data to your spec.

Source: Mozilla Common Voice

Questions

Buyers usually ask.

Do your transcripts include tone marks?

Yes, on request. Transcribers trained in Yorùbá orthography add tone marks and under-dots, and a second person checks a sample of every batch.

Can you deliver dialect-specific Yorùbá?

Yes. Tell us the dialects you need and we recruit and tag speakers by dialect area. Lagos is the easiest place to start because people from across the south-west live there.

How do I order Yorùbá speech data?

Email [email protected] or use the request form with the hours, read/conversation split, speaker mix and transcript rules you need. We record a free sample of about 10 hours to your spec first. Pricing is quoted per approved hour after scoping.

How do speakers consent?

Each speaker passes an ID check and records consent in their own language before any paid task. If the data will be used for voice synthesis (TTS), that use is named explicitly. Speakers can withdraw from future deliveries.

Keep exploring

Tell us what you need. We send a free sample first.

Email [email protected] with the data type, hours, languages or trades, labels and format. Pricing is quoted per approved hour after scoping.

Email us Request form