Speech
Transcription and voice tools that understand African languages.
Speech
Planned
API
Process
~/nzinga
$
audio in, accurate text out
→ dialects and code-switching reviewed
→ in-region or on-premise planned
● transcripts with timestamps
Nzinga · speech
languages
First languages planned
Swahili, Hausa, Yoruba
quality
Native-speaker review
Ongoing
hosting
Where models can run
In-region
Overview
Speech technology still struggles with African languages, dialects and mixed-language speech. Nzinga Speech will transcribe and understand them, trained only on recordings members chose to share and corrected by native-speaker reviewers.
What it includes
transcribe
api
Audio to text with timestamps and speakers
assistants
toolkit
Speech recognition and approved voices for assistants
analytics
calls
Transcripts and summaries of customer calls
on-device
offline
Small models that run offline on low-end phones
provenance
proof
Training data you can trace to consent
Process
01
Tell us your use
Share the languages, audio and accuracy your product needs to reach.
02
We tune the model
Open models, fine-tuned on consented data and reviewed by speakers.
03
Deploy where needed
Run it in-region, on your own servers or on the phone.
Build with speech in your language.
Tell us the languages and the audio you work with, and we will scope a pilot with you.
→ choose languages and domain
→ tune on consented data
→ deploy in-region or on device
✓ consented data · ✓ native review · ✓ in-region
● planned · pilots open