South African speech AI company Saigen today announced the successful close of a new investment from its shareholder, specialised South African technology group Alphawave.
The funding follows Alphawave’s announcement in August 2026 that it had merged its Saigen and Callbi businesses into a single company trading as Saigen, under CEO Sam Clarke.
Since it was founded in 2017, the company has built speech recognition (ASR) for under-resourced languages in specialised domains. Saigen has ASR models in production in 11 African countries, for 22 different African languages.
In South Africa, its multilingual ASR model can recognize any of the eleven official spoken languages, seamlessly handling code-switching, even within a single utterance. Callbi’s AI call intelligence product for contact centres, built on the Saigen engine, processes hundreds of thousands of calls every day, giving the team a deep base of real-world South African conversations to build on.
While the investment sum was not disclosed, Clarke says that Saigen will use the funding to develop capabilities in three main areas. These are:
- Speech-to-text: Saigen will improve and expand the range of its existing speech recognition models. These improvements will be available to all existing and future Callbi customers.
- Text-to-speech: Clarke says that Saigen will extend its text-to-speech models beyond the current English, Afrikaans and isiZulu to include more languages, with more voices in each language.
- Agentic AI: Saigen will accelerate and intensify investment into software agents that can hold a conversation with a customer, understand what they need, and then act on it.
“The agentic applications target a gap that existing offerings leave open,” says Clarke. “South Africans routinely mix languages within a conversation, switching back-and-forth between different South African languages, and use local vernacular that global models rarely see. Most voice and chat bots are built on models trained largely on standard English. When a customer code-switches, these bots misunderstand, ask the customer to repeat themselves, or hand the call to a human.”
He explains that Saigen’s new applications are designed to work the other way round. They will understand vernacular and code-switched responses, and they will also reply in kind, in the customer’s language or the same mix of languages. “This is fundamentally different from the status quo, where technology forces the customer to adapt to the bot. The speech model investment is what makes this possible because broader speech-to-text coverage lets the agents understand more customers, and new text-to-speech languages and voices allow them to answer in a natural, local voice,” he says.