Connect with top gaming leaders in Los Angeles at GamesBeat Summit 2023 this May 22-23. Register here.
Yesterday evening, Baidu held a three-hour virtual keynote to kick off its Baidu World 2020 conference. In addition to voice-activated earbuds dubbed the Xiaodu Pods, the company showcased the newest generation of its natural language processing platform, DuerOS, and its PaddlePaddle machine learning development framework.
PaddlePaddle and Baidu Brain
PaddlePaddle 2.0 beta makes substantial changes to the underlying API structure. Baidu says the directory format has been optimized to better support high-level API functions, with scaffolding for mixed-precision training under a dynamic graph. (Mixed-precision training is the combined use of different numerical precisions in a computational method.) There’s a total of 140 new APIs, 155 modified or improved APIs, and upgraded C++ APIs for the framework’s inference library.
The PaddlePaddle upgrades dovetail with the release of Baidu Brain 6.0, the newest version of Baidu’s machine learning platform supporting AI industrial applications. The company says Baidu Brain now boasts more than 270 core AI capabilities used by over 2.3 million developers, with an API upgrade that enables AI models to be deployed more efficiently.
Baidu also unveiled Xiaodu Pods, dual-microphone earbuds that are imbued with the company’s Xiaodu assistant and claim to last 28 hours on a charge. The wireless Xiaodu Pods have a noise-canceling feature and can translate between several languages. Via Xiaodu, they also relay information like the weather, turn-by-turn directions, answers to math problems, and local news on command. A special Wandering Earth mode available in English and Chinese allows two users wearing one earbud each to translate conversations into their preferred language in real time.
Intelligent Security Summit On-Demand
Learn the critical role of AI & ML in cybersecurity and industry specific case studies. Watch on-demand sessions today.
The Xiaodu Pods arrived with DuerOS 6.0, the latest suite of tools developers can use to plug Baidu’s voice platform into speakers, refrigerators, washing machines, infotainment systems, and set-top boxes. DuerOS 6.0 brings support for low-power voice processing chips and neural beamforming, a technique that employs algorithms to amplify the sound recorded by microphone arrays. Voice recognition error rates are now 46% lower compared with DuerOS 5.0, Baidu claims, and DuerOS 6.0 incorporates a more efficient text-to-speech model (WaveRNN).
Baidu said over 60 automakers and more than 40,000 developers are working to integrate products with DuerOS. As of March, DuerOS was handling 6.5 billion voice queries per month and 3.3 billion from Baidu’s Xiaodu smart speakers and displays alone. As of July 2019, the platform was on 400 million devices, in 500 vehicle models (and over a million vehicles), and in over 100,000 hotel rooms.
DuerOS hasn’t quite reached the storied heights of Amazon’s Alexa and Alexa Voice Service — they have more than 100,000 third-party apps and thousands of brands signed on, not to mention compatibility with 60,000 smart home devices. But Baidu, which claims DuerOS now has over 4,000 third-party apps, continues to work with heavy hitters like Huawei, Vivo, and Oppo to build the operating system into future flagships. Baidu also works with automakers like BMW, Daimler, Ford, Hyundai, Kia, Chery, BAIC, FAW, and Byton and hotel chains like InterContinental.
Baidu this summer inked a strategic partnership with China’s largest smart device manufacturer Midea, which has 70 million smart home appliances on the market, to sell bundles of appliances with Xiaodu products. In the future, Xiaodu-powered devices will be able to control Midea appliances via the cloud or infrared tech.
Beyond PaddlePaddle 2.0, DuerOS 6.0, and Baidu Brain 6.0, Baidu announced developments regarding Apollo, its open source software solution for driverless vehicles. In 2021, Weltmeister will launch a car incorporating Apollo’s valet parking that will be able to identify vacant slots in parking garages and allow people to use Tesla-like smart summon functions, Baidu said. In addition, Baidu demoed what it calls 5G Remote Driving Service, a 5G-powered teleoperation service that allows human operators who have completed at least 1,000 hours of cloud-based training to remotely control multiple vehicles simultaneously in emergency situations.
The fifth generation of Baidu’s autonomous driving kit will be released soon, the company said during the keynote. This will arrive alongside the latest version of Apollo: Apollo 6.0. As of today, Apollo has over 600,000 lines of open source code, 45,000 contributors, and 210 ecosystem partners. (That’s up from 400,000 lines of code, 12,000 contributors, and 130 partners in July 2019.) Notably, Baidu recently launched its own robotaxi service leveraging the Apollo platform — Apollo Go Robotaxi — in Beijing with roughly 100 pickup and drop-off areas covering residential and business zones in Yizhuang, Haidian, and Shunyi.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.