Following the launch of its real-time speech recognition engine that can run offline anywhere, Vancouver AI startup Picovoice has announced a new platform that lets you easily create and train your own voice models.
Picovoice’s new web console supports training universal wake words and domain-specific natural language understanding (NLU) skills that run entirely offline—even on tiny microcontrollers.
To create speech models using Picovoice, users do not need to provide audio training data; they simply type in phrases. By using proprietary transfer learning technology, the platform removes the data collection burden and shortens training time from weeks to hours. This reduces training costs by an order of magnitude while surpassing the accuracy of alternative solutions.
“The platform creates new opportunities for companies to offer private and personalized voice experiences while retaining control of their brand,” said Mehrdad Majzoobi, VP of Strategic Partnerships at Picovoice. “Our AI partnership with ARM accelerates our vision of embedding Picovoice technology in billions of edge devices.”
The web platform introduces an NLU designer to build voice interfaces that infer user’s intent directly from speech. Customers can use the web console to rapidly prototype, iterate, and validate—without having to write a single line of code.
This makes building offline voice interfaces with Picovoice as approachable as cloud-based offerings but without a cloud dependency.
“The Picovoice Console lets our customers build sophisticated speech models without domain expertise,” says David Bartle, VP of Software at Picovoice. “You can focus on designing a voice interface without thinking about signal analysis or machine learning.”