Sauti Labs
Sauti Labs is an African-language AI company. It develops the research, technologies, infrastructure, and products that enable AI to understand, generate, and communicate across African languages — from speech recognition and speech synthesis to multilingual language modelling and voice AI.
I founded the company and lead it as CEO and Chief Research Officer. I set its research agenda and long-term direction, working alongside a leadership team that covers operations, technology, machine learning, and legal and responsible-AI governance.
- African Speech & Language AI
- Speech Intelligence · Language Intelligence · Language Resources · Model Intelligence · Responsible AI
- Speech IntelligenceRecognition, synthesis, voice
- Language IntelligenceUnderstanding, generation, evaluation
- Language ResourcesDatasets, corpora, provenance
- Model IntelligenceTraining, adaptation, efficiency
- Responsible AISafety, privacy, misuse prevention
Why Sauti Labs exists
AI is changing how people communicate with technology, but the quality of that experience still depends heavily on the language a person speaks. Across many African languages, the technology needed to make speech and language AI feel natural, work reliably, and stay accessible is not where it needs to be.
Closing that gap takes more than a research breakthrough or a new model. It takes representative data, rigorous research, reliable engineering, linguistic knowledge, and products designed around the people who use them. Sauti Labs was founded to do that work end to end.
How the company works
The company deliberately spans the boundary between research and product. The quality of a speech system depends on the data behind it, the models that power it, the engineering that makes it reliable, and the experience it delivers. Progress in one of those means little if it never reaches the others.
Research is organised into five teams — Speech Intelligence, Language Intelligence, Language Resources, Model Intelligence, and Responsible AI — whose work feeds the technology and infrastructure behind Sauti’s products. The first of those products, Sauti Cloud, brings speech recognition, speech generation, and voice AI together for developers and organisations.
Research teams and product
Speech Intelligence
Speech recognition, speech synthesis, multilingual and cross-lingual speech modelling, and robustness across speakers, accents, and real recording conditions.
Language Intelligence
Multilingual language understanding, text generation and adaptation, and evaluation and benchmarking for African languages.
Language Resources
Speech, text, and translation datasets; corpus engineering, validation, annotation, and documentation of provenance, licensing, and versions.
Model Intelligence
Training and adapting models that learn from limited data, evaluating where they fail, and making them efficient enough to deploy under real constraints.
Responsible AI
Safety, fairness, privacy, robustness, and misuse prevention — including the specific risks of speech synthesis and voice cloning.
Sauti Cloud
African speech and language AI for developers and organisations: speech recognition, speech generation, and voice AI.
My role
As founder, CEO, and Chief Research Officer, I direct Sauti Labs’ research agenda, strategy, and long-term development, and I am responsible for how research priorities translate into technology and product.
My own technical work centres on machine learning research and AI systems engineering for speech and language: the models, datasets, and systems needed to develop, optimise, and deploy language technology across Africa’s linguistic landscape. I work closely with the technology and machine learning leads on how research results become reliable, deployable systems.
Principles we work by
Research and product, together
We pursue research to overcome technical limits, develop the technology to address them, and engineer that into products designed around people’s needs. Keeping these in one company keeps each honest about the others.
Language resources are infrastructure
Speech and text data carry voices, personal information, and cultural knowledge. Every resource stays connected to its origin, licence, version, and permitted use, so the models built on it remain traceable.
Say what has been demonstrated
We distinguish demonstrated capabilities from research objectives and unverified assumptions, and avoid presenting a model or result as more capable than the evidence supports.
Plan for misuse, not just use
Speech synthesis and voice cloning enable impersonation and deception as easily as accessibility. Foreseeable misuse is part of how we assess a technology before it ships, not after.
Work that can be reviewed and reproduced
Documentation, appropriate testing, and reviewable contributions are expected across research and engineering, so that results can be checked by someone other than the person who produced them.
Where the company is now
Sauti Labs has an established leadership team and five research teams, and its first product, Sauti Cloud, is in development with an early-access waitlist open to developers, researchers, and organisations.
- Sauti Cloud is not yet generally available; early-access sign-ups are taken through the Sauti Labs website.
- The company is open to contributors across machine learning research, AI systems engineering, African language resources, product and design, and responsible AI.
Questions we are working on
- What does an honest evaluation set look like for African-language speech when recordings are scarce and unevenly spread across speakers, dialects, and domains?
- How far can cross-lingual transfer carry speech models into languages with very little training data, and where does it break down?
- What is the right trade-off between model capability, latency, and compute cost for speech systems that must run under practical resource constraints?
- Which safeguards make expressive speech synthesis useful without making impersonation easy?