Low-resource language speech data from Africa and the Andes
Speech in the languages most AI still cannot hear: Swahili, Dholuo and Kikuyu from Kenya, Kinyarwanda from Rwanda, Yoruba and Hausa from Nigeria, Amharic from Ethiopia and Quechua from Peru. Read sentences and everyday conversation from adult speakers of every age group, transcribed by native speakers in each language’s own spelling, with switches to English, French or Spanish marked.
Specification
- Languages
- Swahili, Dholuo, Kikuyu, Kinyarwanda, Yoruba, Hausa, Amharic and Quechua
- Speakers
- Adult native speakers across age groups, from young adults to elders, rural and urban, balanced across gender
- Material
- Read sentences and prompts, and spontaneous conversation on everyday topics
- Audio
- 16 kHz, 16-bit WAV, recorded on phones in homes, markets and outdoors
- Annotation
- Verbatim transcript in each language’s standard spelling (Ge’ez script for Amharic, tone marks for Yoruba), switches to English, French or Spanish, timestamps and speaker turns
- Quality check
- A second native speaker checks every transcript
Example of the data
What you see, and the file you get.
| Swahili | Habari za asubuhi, unaenda sokoni? | conversation · adult |
|---|---|---|
| Amharic | እንደምን አደርክ? | conversation · elder |
| Quechua | Imaynallam kachkanki? | read · young adult |
| audio_url | language | speaker_age | style | transcript | switches |
|---|---|---|---|---|---|
| …/audio/rec_0021.wav | sw | 30-44 | conversation | Habari za asubuhi, unaenda sokoni? | |
| …/audio/rec_0022.wav | sw | 18-29 | conversation | Nitakupigia simu after lunch. | sw→en |
| …/audio/rec_0102.wav | am | 60+ | conversation | እንደምን አደርክ? | |
| …/audio/rec_0187.wav | yo | 45-59 | read | Ẹ káàárọ̀, ṣé dáadáa ni? | |
| …/audio/rec_0233.wav | rw | 30-44 | read | Mwaramutse, amakuru? | |
| …/audio/rec_0310.wav | qu | 18-29 | read | Imaynallam kachkanki? |
Sample rows. Also as JSON or Excel; audio as WAV or MP3.
Contact us to get the dataset
Tell us what you need. We walk you through the data, its formats and how it is checked on a short call.
What it is used for
Speech recognition for underserved languages
Train ASR for languages that general models barely recognise.
Voice services in local languages
Let farmers, patients and customers use voice assistants and phone lines in the language they speak at home.
Multilingual model evaluation
Measure how your speech model really performs in each language, not only in English.
Language identification
Tell languages, and the switches between them, apart in the same recording.
Questions
Who transcribes the recordings?
Native speakers of each language, in its standard spelling. A second native speaker checks every transcript.
Which languages are covered?
Swahili, Dholuo, Kikuyu, Kinyarwanda, Yoruba, Hausa, Amharic and Quechua. Ask us about other low-resource languages you need.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples