But now, as voice becomes how people write critical messages and work documents, errors can cause them to lose their train of thought.
“So accuracy isn’t one feature among several for us,” Kothari added. “It’s what decides whether any of this becomes how people live and work, and it’s where the first and largest part of this round is going.”
Beyond the funding, the company is previewing its first proprietary speech model, Canto, which is designed to be used in places where recordings might not be as clean, like in a car, a public space or an open office.
“In the hardest conditions, with background noise, wind, heavy accents or music, error rates fall from more than 30% of words to somewhere between 5 and 10%,” said Kothari. “Across everyday use, we expect it to reduce the number of dictations you need to edit by 30 to 35%.”
PYMNTS wrote last month about the evolution of AI voice assistants, from competing on whether they could understand what people said to whether AI can do something useful once it can grasp its instructions.
“Many business processes already function like conversations, even when they occur through email threads, spreadsheets and enterprise platforms,” that report said. “A warehouse supervisor can ask about inventory shortages while walking the floor. A field technician can diagnose equipment without stopping to type. A procurement manager can negotiate, retrieve supplier history and update an ERP system during the same conversation.”
For B2B companies, the implication is workflow compression, PYMNTS added. That means voice AI agents collapsing “multiple corporate interactions into a single conversation.”
