Nigerian AI startup Intron has launched Sahara v2, a significant upgrade to its speech recognition platform designed for African languages. The new version expands support to 57 languages and introduces advanced features to address the continent's unique linguistic diversity. This release marks a major step in building inclusive voice technology tailored for African users and enterprises.
Addressing a Continental Gap in Voice AI
Global voice assistants have historically struggled to comprehend the tonal richness and accent diversity prevalent across Africa. Common names and phrases are often misinterpreted, limiting the accessibility of voice-powered services for millions of users. Intron aims to bridge this gap by developing AI that understands how Africans actually speak.
Enhanced Capabilities and Performance
Sahara v2 now supports 23 African languages within its total of 57, adding 24 new ones including Hausa, Swahili, and Yoruba. The platform is also trained to recognize over 500 distinct African accents, with language selection guided by enterprise demand. This expansion significantly broadens the technology's reach across key markets.
A key innovation is the world's first bilingual Swahili-English model, designed to handle the common practice of code-switching. Developed with Penda Health, it facilitates seamless transcription in settings like clinics where multiple languages are used. The update also includes a new Hausa text-to-speech model for creating local-language voice bots.
According to company benchmarks, Sahara v2 significantly outperforms leading global AI systems on Africa-focused datasets. It demonstrates up to 68% better performance on African names and locations and 36% greater robustness in noisy environments. These results highlight the advantage of using locally sourced training data for superior accuracy.
Data-Driven and Enterprise-Ready
The model's accuracy is built on a massive dataset of over 50,000 hours of audio from 40,000 speakers across 30 African nations. Intron curated this data from real-world settings like courts and call centers to ensure the AI understands contextual nuances. This meticulous data collection process was essential for building a robust and reliable system.
Intron is targeting enterprise clients in sectors like healthcare, finance, and legal services across at least six African countries. The platform now offers offline deployment options via a partnership with Nvidia, addressing data sovereignty and low-connectivity challenges. This makes the technology accessible to organizations operating in diverse environments.
Future Outlook and Strategic Growth
Looking ahead, Intron plans to raise $3 million to continue its expansion and develop more domain-specific models. The company's roadmap includes creating additional bilingual models for languages like Yoruba, Hausa, and Kinyarwanda to meet growing demand. Intron is also publishing its first Africa Voice AI Report to share insights on the market.
The launch of Sahara v2 represents a pivotal moment for voice technology in Africa, moving beyond inherited systems to create tailored solutions. By prioritizing local languages and speech patterns, Intron is building a foundational AI infrastructure that enhances accessibility and drives innovation. This development empowers businesses and individuals by making technology more intuitive and inclusive across the continent.