China's Baidu partnered with Conexant, an American-based company that provides products for voice and audio processing technology, to release a development kit and reference design that other companies can use to build far-field voice-enabled AI devices running on Baidu’s DuerOS platform.
Voice-Enabled AI Devices
Some of the first voice-enabled AI devices were iPhones and Android smartphones, but in the past few years we’ve started seeing other types of devices that can be controlled through voice and have AI that can recognize and respond to our commands. These include devices such as Amazon’s Echo and Google Home, but also smart TVs, robots, smart light bulbs, and all sorts of other devices.
Baidu and Conexant are taking part in this rapidly growing market by releasing development kits with their own technologies built in that other companies can then use to build their own retail products.
Baidu's speaker reference design will include Conexant’s CX20924 four-microphone and CX20921 two-microphone voice input processing solutions, as well as Baidu’s DuerOS, which is an entire operating system that’s enhanced by AI capabilities. The DuerOS AI can reply to questions, find local restaurants, play music, make video calls, and control smart home devices.
Improved Speech Recognition
One of the main hardware components that will go in the development kit is Conexant’s AudioSmart(r) voice input processor (CX20924 for four-mic applications, and CX20921 for two-mic applications), which run Conexant’s far-field voice pre-processing software technology. Conexant’s far-field voice input processors focus on the user’s voice and removes the noise and echoes in the background before providing that data to the DuerOS for a more accurate speech recognition.
“The voice revolution is a global phenomenon. By working with Baidu, we help more third-party manufacturers bring to market innovative voice-enabled AI devices with an exceptional conversational AI experience,” said Saleel Awsare, President of Conexant. “The launch of DuerOS development kits and reference designs will drastically reduce development time and cost, allowing manufacturers to quickly bring their innovative ideas to market.”
Baidu has also been working on its own DeepSpeaker software technology improve the accuracy of speech recognition. Combined, the two technologies should be able to raise the accuracy of the system’s speech recognition to a level that should be adequate for natural conversations between people and their AI-enabled devices.
“Conexant brings a valuable asset to Baidu and third-party product developers looking to create innovative applications for the DuerOS AI platform,” said Kun Jing, General Manager of Baidu Duer Business Unit. “Voice interface is a critical part of DuerOS and we are committed to working closely with Conexant to quickly grow the DuerOS ecosystem by offering product developers a solution to help them quickly fulfill consumer demand for top-performing AI-infused devices. We’re working closely with Conexant to ensure their voice solutions provide optimal speech recognition performance with our AI system and are excited to provide device makers tools to jumpstart the creation of new hardware applications,” he added.
Baidu and Conexant's development kit and reference design for companies that want to build AI-powered devices that can be controlled by voice will be available starting July 5, 2017.