Intron unveils Sahara v2 to expands African voice AI to 57 Languages

Intron has launched Sahara v2, the second generation of its flagship speech recognition model, expanding support to 57 languages and more than 500 African English accents in what it describes as a major leap for voice technology built specifically for Africa.

Sahara v2 is trained on over 14 million audio clips spanning 50,000 hours of speech from more than 40,000 speakers across 30 African countries.

The company says the model was developed using recordings captured in real-world environments such as clinics, courts, call centres and busy streets to better reflect how Africans actually speak.

“Sahara v2 proves that when technology is built with deep cultural and linguistic understanding, amazing things can happen,” said Tobi Olatunji, CEO of Intron.

Ayo Oluleye, head of Data & Insights at ARM Investments, said Intron’s models have shown stronger transcription accuracy and contextual understanding compared to alternatives previously tested.

Voice assistants such as Siri and Alexa have long struggled with African names, tonal languages and code-switching patterns.

Common phrases and names are often mistranslated, creating friction for users and limiting access to voice-powered services. Intron argues that global models were not designed for the tonal richness and accent diversity across the continent.

Sahara v2 introduces 24 additional African languages, including Hausa, Swahili, Zulu, Yoruba, Igbo, Twi, Kinyarwanda and Xhosa, among others.

It also debuts what the company calls the world’s first bilingual Swahili-English automatic speech recognition (ASR) model, developed in collaboration with Penda Health in Kenya. The bilingual model is designed to handle rapid switching between English and Swahili, a common speech pattern in East Africa.

Beyond transcription, Intron has rolled out its first Hausa text-to-speech (TTS) model, enabling local language voice bots for round-the-clock conversations.

New offline deployment options also allow organisations to run models securely in environments where data sovereignty and privacy are critical.

Intron claims Sahara v2 outperforms leading global AI systems, including Gemini, GPT-4, Whisper, ElevenLabs, Amazon Web Services and Microsoft Azure on several Africa-focused benchmarks.

According to company data, Sahara v2 delivers 68.6 percent better performance on African names, organisations and locations; 55.6 percent stronger accuracy with numbers; 36.5 percent improved robustness against background noise and overlapping speakers; and 46.7 percent better performance across sectors such as health, legal, finance and telecommunications.

The model is accessible through APIs designed for both real-time and asynchronous deployment, with integration reportedly possible in as little as five minutes.

Intron stated that organisations across six African countries are already using Sahara v2 to power voice banking, automated form filling, KYC processing, medical ambient listening, courtroom transcription and call centre operations.

The company reports that some use cases have reduced processing times by up to 4.4 times.

Sarah Morris, Chief Product Officer at Audere, noted high accuracy across Southern African accents and API reliability exceeding 99 percent in internal tests.