Our next-generation intelligent voice model built from the ground up for Africa
Unlike Global systems retrofitted for the continent Sahara v-2 understands how Africa actually sounds
Built for African Speech
From Lagos Call Centers, to Hospitals in Nairobi, Boardrooms in Cape Town, and Courtrooms in Yobe
Sahara v-2 is trained natively on African speech patterns that are not adapted as an afterthought. It captures the tone, code-switching, accent shifts, and context, reduces noise the way they happen in real life
New Code-Switching Capabilities
The world’s first bilingual Swahili-English ASR model
Introducing the world’s first bilingual Swahili-English ASR model, developed in collaboration with Penda Health, Kenya to support rapid switching between English and Swahili, better reflecting how people naturally speak
Features
Discover the tools built for the way your enterprise speaks, operates, and scales across Africa.
Create natural, multilingual interactions for customer support, service delivery, and real-time responses across African languages and accents. Your users get conversations that feel local, not translated. Learn more
Turn speech into structured data instantly for forms, KYC, and applications. African names, numbers, and details captured with industry-leading accuracy, so your teams spend less time correcting and more time moving forward. Learn more
Enable secure, multilingual banking interactions with high numeric precision, even in noisy, real-world settings. Because in financial services, every digit counts.
Finish notes 7x faster with Voice. Cut patient wait time, increase clinical note quantity and quality, and reduce errors. Accelerate your digital transformation. Learn more
Reliable, private, and accurate transcription and workflow automation, with or without internet connectivity. Sahara-v2 works wherever your enterprise does. Learn more
Transform African Call Centers with Voice AI. Agent Assist, Automated QA, Inbound and Outbound autonomous agents. Supercharge your customer experience. Cut costs. Grow revenue. Learn more
Redefining Justice in Africa with AI. Save time, restore speed and integrity to the justice system. Learn more
Performance and Benchmarks
Sahara-v2 is redefining Speech AI for African Languages and Accents
Sahara-v2 consistently outperforms leading global speech models, including Gemini-3, Meta Omni-language ASR, Azure, Whisper, GPT-4 Audio, Deepgram, and ElevenLabs.
better performance on African names, locations, and organizations
stronger performance with numbers, IDs, decimals, and currency
greater hallucination robustness with long pauses, silence, background noise, and overlapping speakers
better performance across verticals such as health, legal, finance, telco, and call-centers
Word Error Rate (WER). Lower is better.
| Dataset / Language | Sahara v2 | Gemini 3 Flash | Meta Omni ASR 7B | ElevenLabs Scribe v2 | Azure | GPT-4o | Whisper Large v3 | Deepgram Nova 3 |
|---|---|---|---|---|---|---|---|---|
| African Accented English Datasets | ||||||||
| African Names | 11.66 | 29.9 | 52.39 | 25.41 | 35.69 | 52.49 | 43.23 | 36.08 |
| Afrispeech Numbers | 4.27 | 15.28 | 23.93 | 54.46 | 22.43 | 17.45 | 18.11 | 59.18 |
| Hallucination Robustness | 7.855 | 91.92 | 70.41 | 237.08 | — | 33.59 | 115.91 | — |
| Med-Convo-Nig | 18.31 | 28.35 | 49.19 | 28.9 | 29.17 | 30.8 | 31.76 | 29.46 |
| Afrispeech-Clinical | 15.24 | 23.24 | 42.73 | 27.08 | 32.9 | 28.54 | 32.59 | 40.65 |
| Health | 14.5 | 19.63 | 38.44 | 21.14 | 25.74 | 26.49 | 28.66 | 28.46 |
| Call Center | 15.36 | 20.35 | 57.63 | 23.97 | 23.51 | 23.2 | 24.69 | 22.28 |
| Afro-Finance | 6.1 | 59.71 | 37.35 | 41.68 | 23.91 | 32.06 | 24.85 | 49.9 |
| Afrispeech-Parliamentary | 13.01 | 29.05 | 26.51 | 20.2 | 18.75 | 64.39 | 19.99 | — |
| Legal | 15.03 | 17.41 | 67.04 | 22.32 | 31.335 | 32.54 | 30.68 | 23.41 |
| African Languages – Afrivox Transcribe-v1 | ||||||||
| Swahili | 17.66 | 15.83 | 17.35 | — | — | — | — | — |
| Hausa | 19.5 | 31.37 | 41.87 | — | — | — | — | — |
| Zulu | 21.86 | 26.85 | 31.05 | — | — | — | — | — |
| Yoruba | 21.95 | 27.35 | 31.96 | — | — | — | — | — |
| Kinyarwanda | 15.44 | 40.44 | 26.6 | — | — | — | — | — |
| Igbo | 23.17 | 50.02 | 57.42 | — | — | — | — | — |
| Luganda | 19.42 | 35.14 | 22.41 | — | — | — | — | — |
| Xhosa | 28.36 | 33.82 | 32.34 | — | — | — | — | — |
| Shona | 29.26 | 70.56 | 20.29 | — | — | — | — | — |
| Tswana | 22.6 | 78.09 | 48.6 | — | — | — | — | — |
| African French | 10.02 | 6.63 | 33.41 | — | — | — | — | — |
| Akan | 24.69 | 46.74 | 52.53 | — | — | — | — | — |
| Twi | 10.63 | 44.35 | 52.81 | — | — | — | — | — |
| Sesotho | 19.8 | 161.41 | 64.26 | — | — | — | — | — |
| Fulani | 38.83 | 61.62 | 53.41 | — | — | — | — | — |
| Arabic | 15.34 | 12.13 | 22.92 | — | — | — | — | — |
| Ga | 10.91 | 74.55 | 101.82 | — | — | — | — | — |
| Amharic | 27.5 | 57.54 | 23.64 | — | — | — | — | — |
| Afrikaans | 24.79 | 16.47 | 21.8 | — | — | — | — | — |
| Sepedi | 23.7 | 39.78 | 42.82 | — | — | — | — | — |
Capabilities
Sahara-v2 is purpose-built for the linguistic realities of Africa. It captures local dialects and major African languages with high-fidelity transcription designed for production environments.
Regional dialects, diverse English accents, seamless code-switching. Sahara-v2 processes conversations naturally and understands speakers the first time.
Accurately identifies African names, locations, and organizations, giving finance, healthcare, legal, and customer teams data they can act on with confidence.
Captures decimals, currency, IDs, dates, and fractions with precision, ensuring numerical data is reliable from input to output.
Deploy natural, localized voice experiences across seven languages. Automate at scale without losing authenticity.
Turn live speech into clean, structured data in real time. Reduce manual entry, cut errors, and accelerate workflows.
Run fully offline for uninterrupted performance in remote, low-connectivity, or high-security environments.
Handles background noise, overlapping speakers, and long pauses without compromising accuracy.
Industry Impact
Sahara-v2 is proven across critical sectors, delivering powerful, high-precision voice AI that enterprises can trust.
The Evolution of Sahara-v2
Africa speaks differently, and traditional voice AI could not keep up. Sahara v1 was built to understand African languages, dialects, and the subtle nuances of real conversations.
Sahara v1.5 improved on the foundation, becoming faster, smarter, and more reliable while recognizing every accent, tonal pattern, and informal speech style across the continent.
Sahara-v2 marks a new era in voice AI. It does more than transcribe words- it understands context, captures meaning, and transforms every conversation into actionable insights for enterprises.
Most voice AI systems are built elsewhere then adapted for Africa.
Sahara v2 is built in Africa , for Africans . It understands:
Because Africa isn’t one voice. It’s many. And Sahara-v2 listens to all of them.
Trust and Responsibility by Design
As we build new technologies, we take our responsibility seriously, making safety and security a priority. Privacy and reliability are not just features – they are the foundation of everything we do.
For Developers
Build next-generation voice applications with Sahara-v2’s APIs, offline intelligence, and bilingual support.
For Enterprises
Deploy Sahara-v2 at scale across any organization. Turn every conversation into accurate, actionable data that powers smarter decisions and streamlined workflows.
Gain deep insights into the evolving technological landscape with our upcoming 2026 Africa Voice AI Report. The report explores the trends, challenges, and opportunities shaping the future of speech interfaces across the continent.