Recently, “MIT Technology Review” China released the “2021 Artificial Intelligence Innovation Research Institute (AI Labs) Report”, which depicts the annual picture of AI technology innovation and future technology trends in a diverse and objective manner, thereby foreseeing changes in China and even the world technical strength. At the same time, the report selected 10 corporate AI research institutes with technological innovation and leadership, and Xiaomi AI Lab was successfully selected. This is a recognition of Xiaomi’s technical strength.
Xiaomi AI Lab was listed on the list with “AI penetrates C-end life scenarios and continues to improve the strategic layout of AIoT”. “MIT Technology Review” China announced the reason for Xiaomi AI Lab’s listing: “Xiaomi provides users with a rich smart life experience around the “mobile phone × AIoT” strategy and core business needs. Its self-developed technology is widely used in smart Smart hardware products such as mobile phones and speakers, as well as Internet service businesses such as e-commerce, have created Xiao Ai, an industry-leading smart life assistant, and fully integrates AI technology into C-end users’ segmented life scenarios.”
Xiaomi established Xiaomi Discovery Lab, the predecessor of Xiaomi AI Research Institute, in 2014, and officially established Xiaomi AI Lab in early 2016. Its research directions include computer vision, acoustics, speech, NLP, knowledge graph, machine learning, and other fields.
The report pointed out that Xiaomi, as a consumer electronics and smart manufacturing company with smartphones, smart hardware, and IoT platforms as its core, pays attention to hardware-based AI technology support, and builds voice interaction as a basic capability. Its advantages are reflected in diverse application scenarios, data richness is high. Personalization, refinement, multi-modality, and security privacy are the development directions of Xiaomi technology. AI technology will be more integrated into the subdivision life scenarios of C-end users. Through the combination of software and hardware technologies, while ensuring user privacy and security, provide refined services, understand life better, and get closer to life.
As the “experimental field” and “munition depot” for Xiaomi AI technology, the Xiaomi AI Lab aims to develop the industry’s first-class AI technology and deliver key AI capabilities for the company’s key businesses. At the same time, combined with the company’s major strategic needs, exploratory pre-research work will be carried out to reserve core AI technology for the company’s future and enhance the company’s technology brand influence.
Xiaomi AI has independently developed technology and has achieved a series of results in research fields such as computer vision, acoustics, speech, NLP, knowledge graph, and machine learning.
JOIN US ON TELEGRAM
Computer vision field
Xiaomi AI Lab has implemented more than 40 visual core algorithms on mobile phone cameras and AIoT devices, creating the ultimate photo-taking experience for users. Among them, more than 20 algorithms will be implemented on more than 10 models in 2021, and more than 20 pre-research algorithms are in progress and will gradually be implemented on flagship mobile phones.
On the DXO Mark list, the world’s most authoritative third-party evaluation agency, AI Lab’s HDR, portrait ultra-clear, and edge image quality enhancement technologies helped Xiaomi phones reach the top three times. The magic series such as magic day change, magic avatar, etc. have good segmentation effects, fast speed, and novel and interesting experiences.
In addition, the Mi 12 computational photography technology “Following Focus of Everything” jointly developed by the AI Lab and the Camera Department has been widely praised by users.
Acoustic field
Xiaomi AI Lab is committed to the research and engineering implementation of artificial intelligence acoustic technology, creating an industry-leading audio experience for Xiaomi mobile phones and IoT full ecological hardware products.
At present, technologies such as AI call noise reduction, collaborative wake-up, collaborative sound playback, and spatial audio have been implemented in more than 60 products such as mobile phones, TVs, Xiaoai, and headphones.
In 2018, Xiaomi built the first fully automatic far-field acoustic laboratory in China and became the first joint laboratory of the China Artificial Intelligence Industry Alliance. The Xiaomi far-field acoustic test specification was established as the alliance test standard. In recent years, the team has led the initiation of 1 international standard, 2 national standards, 2 industry standards, 4 group standards, and 2 enterprise standards, laying a leading edge for Xiaomi in the field of far-field acoustics.
Voice field
Xiaomi AI Lab has made major breakthroughs in speech recognition, voice wake-up, voiceprint recognition, spoken language evaluation, speech synthesis, singing synthesis, AI composition and arrangement, etc. At present, Xiao Ai undertakes 190 million voice requests every day.
Xiaomi AI Lab has implemented AI voice technology on Xiaomi mobile phones, tablets, watches, bracelets, headphones, TVs, speakers, watches, quadruped robot dogs, and other products. At the same time, the team has successively published more than 10 academic papers at top international speech conferences such as Interspeech and ICASSP and won 5 international competitions for speech recognition, voice wake-up, and voiceprint recognition technologies for the company.
A new generation of KaIdi
Led by Dr. Daniel Povey, Group Chief Speech Scientist, the Next Generation Voice team is committed to developing a new generation of Kaldi projects while maintaining existing Kaldi projects. As the most eye-catching open-source voice project in the industry in the past 10 years, Kaldi has achieved great success in both scientific research and industrial circles.
The new generation of Kaldi is a new future-oriented open source speech recognition framework independently developed by Xiaomi. While promoting the development of speech recognition technology, this project will also improve the international competitiveness and influence of Xiaomi’s speech recognition technology.
After more than a year of intensive development, the three sub-projects of the current new generation of Kaldi, K2, Lhotse, and Icefall, have been basically developed and have been officially released at the InterSpeech conference in August 2021, which means a new generation of Kaldi projects. The layout has been completed. It is believed that the release and widespread use of a new generation of Kaldi, will completely change the way people implement speech recognition.
NLP field
Research and develop core technologies for natural language processing, and create leading technology platforms and innovative products. The team focused on creating directions such as search, recommendation, advertising, multimodality, machine translation, human-machine dialogue, and text creation.
Search, recommendation, and advertising support Xiaomi Mall, Youpin, new retail, advertising recommendation, and other businesses. Using multi-modal technology to automatically generate advertising creative, successfully support advertising. Machine translation is widely used in Xiaoai classmates, mobile phones, MIUI portal and web page translation, Xiaoai teacher, Xiaoai mouse, and other devices.
The device-side translation experience is industry-leading. Simultaneous interpretation has been launched in Xiaoai face-to-face simultaneous interpretation, conference simultaneous interpretation, AI subtitles, call translation, and other scenarios, and has become a function of Xiaomi’s high-end product promotion pages.
In terms of man-machine dialogue, the chatting function has been developed, which is one of the most popular functions of Xiao Ai in important equipment such as mobile phones and stereos. Based on the latest pre-training generation technology, machine writing has achieved near-human-level effects in the fields of couplet and poetry generation, and has been applied to Xiao Ai’s festival operation and communication; language style conversion technology has successfully supported Xiao Ai’s language style customization of various custom tones.
Knowledge Graph Domain
In addition, Xiaomi AI Lab has made major breakthroughs in the field of graph construction and graph application. In terms of graph construction, Xiaomi AI Lab and OpenKG have carried out the open-source work of knowledge graphs.
In addition, in terms of entity linking technology, it won first place in the Chinese short text entity linking competition twice. In terms of graph applications, knowledge graphs are widely used in Xiaoai, accepting hundreds of millions of query requests every day, and winning first place in the 2021 CCKS knowledge quiz competition. The intelligent customer service business relies on the knowledge map platform to create online customer service robots, intelligent outbound calls, intelligent IVR, and other products, saving thousands of customer service manpower for the company every year.
In the insight section of the report, MIT Technology Review China pointed out that the AI research institute of an enterprise is the main position for creating a new business growth curve, and it is invisibly brewing new breakthrough points and growth points. In the camera direction, Xiaomi AI Lab improves the clarity of photos, focusing accuracy, and fun operation by applying computer vision-related algorithm technology.
In the direction of voice interaction, Xiao Ai has gradually upgraded from a voice assistant to an intelligent life assistant. NLP and knowledge graphs are widely used in the field of Internet content. In general, around Xiaomi’s core strategy of “mobile phone xAIoT”, Xiaomi AI has made a complete layout in key AI products and technologies.
Only when technological innovation is closely integrated with society, industry, and market can it maximize commercial achievements and truly change people’s production and life. In the future, Xiaomi AI Lab will continue to explore new heights of science and technology, and use artificial intelligence technology and products to allow more people to enjoy the beautiful life brought by technology.
(VIA)
