On February 13, Xiaomi Mi 10, which has been warmed up for a long time, finally officially debuted. Due to the pneumonia epidemic, the press conference adopted a purely online live broadcast mode, but this did not affect its extremely high attention. As the first Snapdragon 865 flagship in China, Xiaomi Mi 10 has achieved self-breakthrough and brought many surprises.
In addition to the dazzling parameter information such as the Snapdragon 865 processor and LPDDR5 memory, the achievements of Xiaomi’s self-developed AI technology are also worthy of attention. They have penetrated into every corner of Xiaomi’s products, and have a lot of impact on cameras, systems, voice, etc. Aspects of the use experience have had a profound impact.
Xiaomi AI voice has become stronger!
1. Xiaomi voice AI technology is rapidly popularizing
At a Xiaomi conference in July 2017, the Xiaomi AI speaker was officially released, and Xiaomi started the test of the self-developed voice AI technology in the market.
In just over two years, Xiaomi’s self-developed voice AI technology has penetrated into various types of Xiaomi products. In addition to smart speakers, Xiaomi products such as Xiaomi mobile phones, Xiaomi TVs, and Xiao Ai teachers are all equipped with “Xiao Ai classmates”. A device that integrates Xiaomi’s voice intelligent assistant, which makes the product more convenient to use through the new interaction method of voice.
Voice interaction liberates users’ hands, reduces learning costs, improves user experience, and enhances product competitiveness.
2. The smart assistant on Mi 10 is more interesting
At present, the voices issued by the voice assistants on smartphones are all artificially synthesized, and basically do not provide personalized services. In actual use, it is inevitable to feel mechanical and unnatural, and the gap between the voices produced by real people is still relatively large.
The big surprise on the Mi 10 is that it is equipped with a personalized speech synthesis service. It adopts the latest synthesis technology. Users only need to record a small number of sound samples in a quiet environment and upload them. After the server recognizes, trains and models, they can obtain a customized AI voice assistant.
The voices issued by the voice assistants of other people’s mobile phones are the same, but Xiao Ai on the Xiaomi mobile phone can make thousands of voices, and it feels like there is a real assistant in the mobile phone.
3. What skills does Xiaomi show in AI voice technology?
The mainstream speech synthesis technology on the market has many deficiencies, which are manifested in the fact that the voice produced is too mechanical, like a robot speaking, and in the mixed Chinese and English speech, the pause rhythm and excessive feeling of bilingual switching are unnatural, etc.
The personalized speech synthesis service on Xiaomi Mi 10 aims to improve these problems. Specifically, its implementation process can be divided into several steps.
First, the user needs to record the target sound in a quiet environment;
Then, the system will perform noise reduction, error detection and other processing on the collected sound information;
Then, the processed target sound will be extracted features;
Finally, the Xiaomi cloud server trains and deploys the collected information online, and generates a speech synthesis engine.
This technical process looks relatively clear, but there are many difficulties. For example, it has relatively high requirements on the data quality of the target sound, the online training model is time-consuming and labor-intensive, and whether a small amount of data can be trained to achieve satisfactory results.
However, Xiaomi has solved these problems perfectly, the model training time is greatly shortened, and users only need 20-30 minutes to complete the whole process. In addition, its synthesis effect is stable, and it can synthesize simple English even without English corpus. voice.
With AI blessing, Xiaomi Mi 10 plays a new trick with the camera
The improvement of the camera is one of the most important selling points of the Xiaomi Mi 10 series. The 100 million pixels and four rear lenses have attracted the attention of countless people. In addition to the crazy stacking of materials on the hardware, the camera of Xiaomi Mi 10 also shows the power of software algorithms.
1. Millet 10 is stronger
The “one-click sky change” function of Xiaomi CC9 has been welcomed by many users. After the photo is taken, you only need to tap the editing options to change the sky in the photo to the effect you want, such as sunny day, Sunset, twilight, etc.
This time, the Xiaomi Mi 10 day change function has added rainy and snowy days on the basis of the previous one, which is more powerful. From the comparison of Xiaomi’s demonstration, it has won a great victory in PK with various third-party APPs. The effect of changing the sky is very natural, and it can basically reach the level of falsehood.
Behind the Xiaomi Mi 10’s day changing function is Xiaomi’s continuous investment in visual imaging technology. When MIUI 10 was launched, it brought the function of AI Selfie blur, and the single-camera can also have a portrait blur effect, which is achieved through excellent algorithms.
Xiaomi applied the previous Selfie blurring algorithm experience and technology to the sky changing function, marked tens of thousands of actual sky pictures for training, and optimized the sky segmentation model, and finally achieved the current amazing effect. It has to be said that the actual collection of proofs and the use of AI technology to train the model is quite time-consuming and labor-intensive, and requires a lot of long-term investment.
Of course, the help of this algorithm for imaging is not comparable to that of general software optimization. In fact, with the excellent imaging effect of the Google Pixel series, the camera algorithm is also the idea of training the AI model through actual sample collection. In this regard, Xiaomi and Google thought of going together.
2. Shooting vlog can also be done with one click
The vigorous development of short videos has made vlogs popular, but it is not easy for ordinary people to cut out cool and delicate vlogs. First of all, video post-production is more complicated than pictures, and vlog production also involves music, subtitles, special effects, etc.
And Xiaomi has been keenly aware of the needs of users, and has invested in multiple teams to help everyone get started vlogs. Simply put, it optimizes several aspects.
First of all, the automatic mirror movement function allows users to have excellent results without moving their mobile phones, without the need to learn very professional skills such as slide rails and hand cranking. However, in the specific implementation process, the technical difficulty is not small.
During automatic mirror movement, the angle of each frame of the video must be accurately matched, and the 4K resolution data must be calculated in real time, such as scaling and transitions, which requires high requirements and tests for the accuracy of the algorithm. Through algorithm optimization and adjustment, Xiaomi AI Lab combines multiple computing units such as CPU, GPU, and decoder to “squeeze out” the computing performance of mobile phones.
In addition, the function of voice to subtitles is also worthy of praise. Manually adding subtitles to videos is very troublesome. With the voice AI technology, Xiaomi phones can automatically convert the voice in the recorded video into text subtitles, saving vlog producers a lot of work.
3. The “little thing” of document scanning, Xiaomi Mi 10 has achieved the ultimate
Another very bright feature on the Mi 10 camera is “Xiaomi Documents”, which solves many pain points in the scanning process of mobile phone documents: inaccurate frame recognition, loss of details, unclear Display, deformation, shadows, etc.
What impresses people about Xiaomi Documents is that AI technology has greatly helped the document scanning function. For example, its cropping and correction ability is very strong. Even if the background of the photographed document is very messy, it can accurately locate the document boundary and accurately Crop and straighten document areas to remove distracting extra elements.
In addition, in many cases, the ambient optics for photographing documents are not ideal, and the imaging area is partly dark and partly bright, which greatly affects the final scanning effect. Xiaomi Paid Document solves the problem through targeted shadow removal technology, which greatly improves the success rate of document scanning.
Technological innovation has created the excellent experience of Xiaomi document shooting. The neural network is used to distinguish the light and shadow parts of the image, the edge algorithm is used to help users to accurately locate the document, and a variety of document usage scenarios are actually collected for training optimization.
It is worth mentioning that Xiaomi’s documentation is completely independently completed by the Wuhan Vision team of Xiaomi AI Lab. During the epidemic, their work and contributions are worthy of praise.
Xiaomi’s self-developed AI technology is worthy of further expectations
The Xiaomi Mi 10 conference this time has shown us a lot of things that Xiaomi has easily overlooked in the past, and self-developed AI technology is an important part of it.
First of all, the products represented by Xiaomi Mi 10 show the strong strength of Xiaomi in self-developed AI technology. Whether it is the specific functions of AI assistance such as personalized speech synthesis service, one-key change of sky, vlog automatic subtitles, or the AI research and development ideas of training models by collecting real data, we can see Xiaomi’s intentions in self-developed AI technology.
Secondly, there are many manufacturers researching AI technology, but it is rare that Xiaomi precisely aims at user experience. The functions such as scanning documents and making vlogs that we mentioned earlier can actually be implemented by third-party applications, and mobile phone manufacturers generally do not put too much effort into these details.
But these seemingly inconspicuous details are closely related to user needs. Xiaomi has invested the achievements of its self-developed AI technology in these fields, so that its users can most directly enjoy the dividends of the development of AI technology. Xiaomi AI technology pays more attention to the implementation of scenarios related to user experience, and each technological innovation can identify the actual needs of users and improve user experience. Of course, this kind of gene that focuses on user experience has existed since the birth of MIUI, and now it has been engraved into Xiaomi’s AI technology research and development.
In addition, Xiaomi is involved in far more fields than mobile phones, home appliances, smart home, PC and other industries, you can see Xiaomi’s active figure. And Xiaomi’s self-developed AI technology is not limited to adding bricks and mortar to mobile phone functions. In the future, we are expected to see the explosive results of Xiaomi’s self-developed AI technology in imaging, voice, 5G, IoT and other fields, which is very worth looking forward to.