Smart home is popular but not profitable, on-device AI can help

The smart home market was once considered one of the largest market segments in the Internet of Things, and voice is an AI technology widely used in smart homes. With the acceleration of domestic consumption transformation and upgrading, industry insiders have optimistically predicted that smart homes will usher in explosive growth. However, the Chinese smart home market, which was once highly anticipated, has always faced a situation of "good reviews but poor sales." Those dazzling smart home appliances that once appeared in exhibition halls and news reports have not really entered the homes of ordinary people.

From a technical perspective, voice technology has entered the commercial stage; from the perspective of user mentality, consumers have also become aware of smart home appliances; for home appliance manufacturers, they hope that smart manufacturing and AI will increase the premium ability of their products and establish a second curve of corporate growth.

This market containing a trillion-dollar gold mine seems to be right before our eyes, but what is the correct approach to opening up the market?

AI implementation: two major barriers between manufacturers and users

Lu Yong, CEO of Beijing Tanjing Technology Co., Ltd., believes that the current difficulty in implementing smart homes is due to two major barriers between manufacturers and users.

The first is the R&D threshold of manufacturers. Most home appliance manufacturers are more concerned with the R&D of the functions of home appliance products themselves, and are not good at AI voice. Large and powerful manufacturers often set up intelligent departments to be responsible for the R&D and implementation of intelligent voice solutions, while small and medium-sized manufacturers have neither AI R&D capabilities nor intelligent departments, and can only passively follow suit.

In addition to product research and development itself, home appliance manufacturers also need to solve problems such as ecology, cloud platform, APP, and interoperability, which further increases the difficulty of product implementation.

The second is the user threshold. From the user's perspective, smart products connected to the cloud must meet the rigid requirement of Wi-Fi networking. An unstable network environment will affect the user experience. In addition, they need to download APP, register, set up, connect to a third-party central control, and other operations.

Finally, users also have to face the incompatibility issues between various smart home ecological platforms. Various platforms and brands are scrambling to seize user entrance traffic, resulting in a situation of fragmentation, but ultimately consumers have to pay the bill.

All of the above have invisibly built a "user" threshold: that is, only digital experts or R&D engineers with networking or maintenance capabilities can play with it, while ordinary people are inevitably discouraged.

"I spent an hour studying it and half a day teaching my parents how to figure it out, but they only have a vague idea of ​​what it means. I don't think this is a smart home, it would be more appropriate to call it an 'intelligence home'." The users' voices are a vivid reflection of this reality.

These visible and invisible obstacles seem complicated and confusing, and perhaps they can only be solved by clarifying the essence of "intelligence".

Client-side voice, the sword to solve the "Golding knot"

There is such a story in the ancient fable book: whoever can untie the strange Gaulding knot is destined to become the king of Asia, and all those who try to untie this complicated knot end in failure. Alexander the Great's way of untying this rope is to establish my own rope-untying rules, no longer try to untie the rope, but draw the sword and split it in two.

It seems that the "last mile" of smart home implementation is just like the Golding's knot, with multiple factors intertwined to form an insurmountable "level". 

In essence, the difficulties in implementing smart home are caused by the "inertia of thinking" in the smart home industry. Lu Yong explained that since the development of AI voice capabilities requires the assistance of cloud computing power, many people in the industry naturally think that cloud computing is the only direction of intelligence, and the intelligence of products is to connect with the cloud, without considering the inherent use of the product itself, and forcibly binding "intelligence" with "cloud and ecology".

Therefore, in Lu Yong's opinion, the best technical path to solve the implementation of smart home is the offline solution. Looking back at the historical evolution of technology, we can find that computing power was first released from large computer rooms to personal PCs, and then from computers to mobile phones. All computing power has evolved from the initial centralized to the final division of responsibilities. In the AI ​​era, it is against market value to concentrate all computing power in the cloud.

Especially in the field of smart home, although the categories of home appliances seem numerous, after detailed analysis, it is found that only a few home appliances that require content, such as TVs and smart speakers, need to be connected to the Internet and the cloud; while for products such as lamps, switches, air conditioners, fans, etc., the required computing power can be achieved on the terminal side.

Looking further, the advantage of end-side voice is that it does not need to rely on the network, does not need to develop an APP, does not need to be adapted to third-party central controls such as speakers, and does not need to consider ecological interconnection issues. It can be used by plugging in, and a single machine can achieve intelligence, which will greatly reduce the user's usage threshold.

In addition, as users pay more and more attention to privacy, personal data privacy has also become a factor for users to consider when purchasing and using smart homes. The data processing method of localized voice on the terminal side also enables smart home appliances to improve response speed while avoiding the problem of product collection of user data, which can effectively protect the user's data privacy security.

Since there is no need to connect to the Internet, a single machine can be intelligent, and the end-side voice can easily free home appliance manufacturers from the invisible constraints of various ecological platforms, eliminating the need to worry about being strangled by Internet giants.

If you understand one thing, you will understand everything.

Moreover, while the edge-to-edge voice solution helps B-end manufacturers reduce costs and increase efficiency as well as enhance their premium capabilities, it also lowers the usage threshold for C-end users, forming a positive cycle between the two and truly moving smart homes from "geek intelligence" to "people's intelligence."

Exploring solutions to help manufacturers achieve killer products

However, there are many voice solutions on the market. How can we find a suitable voice solution?

In Lu Yong's opinion, a qualified voice solution must simultaneously meet the requirements of recognition rate, noise robustness, low false wake-up rate, low false recognition rate and cost.

"We have seen products on the market that have maximized the recognition rate, but the false wake-up rate of such products is often high. If you only focus on one indicator, the product may seem to be very sensitive, but other hidden problems will seriously damage the user experience." Lu Yong said bluntly. 

Tanjing's end-side voice solution can achieve a balance among the five indicators. Taking Tanjing's smart lighting solution as an example, it can achieve a 97% recognition rate, a false wake-up rate of less than once a week, and a far-field voice recognition capability of 10m. Even in harsh environments with a signal-to-noise ratio of less than 5dB, it can achieve accurate recognition results.

The reason behind such "outstanding performance" is obviously due to the integrated hardware and software R&D strategy that Tanjing has adhered to since its inception.

Tanjing has its own SFA (storage priority) architecture with independent intellectual property rights. It uses storage to drive computing, integrates storage, computing, and scheduling algorithms, and uses software and hardware to work together to allow the controller to perform optimal mapping between storage and computing nodes. The SFA architecture naturally supports high-concurrency computing and fundamentally solves the bandwidth problem caused by AI computing.

In terms of algorithms, Tanjing has developed its own AI noise reduction algorithm based on deep learning convolutional neural networks. A single or dual microphone can achieve the effect of a microphone array, with advantages such as good noise immunity and high remote recognition rate. It can not only process steady-state noise, but also efficiently filter some non-steady-state and sudden noise.

According to Lu Yong, Tanjing's end-side voice solution has the same process as the traditional MCU import process and is very cost-effective. Its price is almost the same as that of ordinary MCUs. Its voice solution has covered most home appliances and can empower a large number of home appliances, saving manufacturers valuable R&D time, achieving rapid intelligence, and further seizing market opportunities. 

Competition in the red ocean has always been an inevitable development that the home appliance market has to face. As the AI ​​technology foundation, industrial foundation and market demand become more mature, the evolution of smart home appliances will continue to lead the industry. Smart products such as lamps, switches, fans, and air-conditioning environment control equipment will hopefully become "hot products" of black technology due to their simple development logic and obvious user perception.

From a development perspective, Lu Yong mentioned that the empowerment of home appliances by voice is not just as simple as intelligent hot products. The current interaction mode of home appliances has evolved from buttons to touch in 10 years, but the distance between human hands and machines is still 0cm, while voice gives the product a new control interface of at least 5-10m. In the future, when users become accustomed to voice, voice will inevitably replace touch buttons, and once again subvert the top UI design of the entire home appliance manufacturing industry, thus triggering a revolution in the entire industry.

How far is that day? When users are used to being able to sit still and center themselves, without having to move their feet for the location of home appliances or lift their fingers for a button, it means that the voice era has finally arrived.