Unable to sell smart speakers, waiting for AI to save lives

Source: Shen Ran, Author: Wang Min, Editor: Jin Yufan

Image source: Generated by Unbounded AI tool

Smart speakers are falling all the way.

According to relevant data from Luotu Technology, the sales of domestic smart speakers have declined for two consecutive years. In the first quarter of this year, it fell by more than 20% year-on-year.

Since Amazon's smart speaker Echo was born and became popular all over the world in 2014, smart speakers have become a rising star in the field of consumer electronics, and major domestic manufacturers have entered the market one after another. By 2018, the domestic smart speaker track will gather three major players, Baidu Xiaodu, Ali Tmall Genie, and Xiaomi Xiaoai, and the competition will intensify. In the round after round of price wars by major manufacturers, the sales of smart speakers continue to rise, and will reach their peak in 2020.

But entering 2021, smart speakers are becoming less and less sellable. After the periodical demand rise brought about by the epidemic gradually dissipated, the decline of smart speakers became more and more obvious.

On the one hand, many users believe that smart speakers are not so much "artificial intelligence" as "artificial mental retardation", and it is difficult for users to have smooth voice interaction with the speakers. The role of "super entrance" is favored by major manufacturers, but in the smart ecology with obvious "island effect", it is difficult to take on the heavy responsibility of the home smart center.

Therefore, smart speakers are often bought back, and it doesn't take long for them to eat dust. However, new changes are taking place.

In 2023, ChatGPT will become popular, and the transformation of AI large-scale model technology will change all walks of life, including the smart speaker track. The three leading players in the smart speaker track have all announced that they will be connected to large-scale model technology, and Baidu and Ali are even ahead.

It is foreseeable that the advantages of AI large-scale model technology in logical reasoning, semantic understanding, and dialogue interaction will greatly improve user experience. Practitioners are looking forward to AI technology changes that can bring smart speakers back to life.

Plummeted 25%, smart speakers shut down

Smart speakers are getting cooler.

In March of this year, the Dingdong speaker, known as the first real smart speaker in China, officially announced that it will terminate its operation service, and all intelligent voice functions, background cloud resources, skills and other services of the Dingdong smart speaker will be discontinued. The former industry leader, released its first product in 2015, has gone through eight years, and finally fell into the cold wind of the market.

Sales data also verifies the desolation of smart speakers. Data from Luotu Technology shows that in the first quarter of 2023, the monitored retail volume of smart speakers in China was 1.57 million units, a sharp drop of 40.6%. From the perspective of omni-channel statistics, the overall sales volume of smart speakers in the first quarter was 5.709 million units, a year-on-year decrease of 25.2%.

This is not the first time the smart speaker market has declined. According to previous data from Luotu Technology, the shipments of smart speakers reached their peak in 2020, and have been in a state of decline since then. Among them, the sales volume of China's smart speaker market in 2021 will be 36.54 million units, a year-on-year decrease of 3.5%; in 2022, the sales volume of China's smart speaker market will be 26.31 million units, a year-on-year decrease of 28%; the market sales will be 7.53 billion yuan, a year-on-year decrease of 25%.

Source/ Luotu Technology

Once the smart speaker market was in full swing, it is still vivid.

In 2017, Ali and Xiaomi entered the market successively, and by 2018, Baidu also quickly followed up. At that time, big manufacturers entered the market one after another, and manufacturers started a price war. The original market price of smart speakers could reach four to five hundred, but the preferential price of Baidu Xiaodu’s first speaker product dropped to 89 yuan, and the scene of staking land became more and more intense.

By the end of 2018, the total data from all channels of Aoweiyun.com showed that the retail volume of China's smart speaker market was 16.25 million units, an increase of 823% year-on-year, and the retail sales were 3.65 billion yuan, an increase of 645% year-on-year.

Capital has catalyzed the rise of smart speakers, and the concentration of market share has also increased. The market structure of Baidu, Xiaomi, and Ali Tmall Genie has gradually formed. According to relevant data from Luotu Technology, by 2022, the market share of these three companies in China's smart speaker market will exceed 90%.

However, the low-price strategy doesn't always work because smart speakers don't provide a satisfying user experience.

"Deaf and dumb", a user who bought a smart speaker of a certain head brand in 2022 complained, calling the smart speaker, often getting no response, and when asking questions, the answers received were mostly "don't know", "Go to study", and sometimes suddenly make a sound in the middle of the night, and stage a "horror movie".

Due to the weak voice interaction ability of smart speakers, manufacturers have launched "smart speakers with screens", which use screen manual control to make up for the problem of poor voice interaction experience, but the convenience of use has been greatly reduced.

Locke Capital analyst Deng Xintao told Shenran that the decline in sales of smart speakers may be due to insufficient product power. Smart speakers are not smart, especially the inaccurate and unstable voice control makes the consumer experience not good, and a large number of smart speakers are more like a gimmick.

Wang Chao, founder of Wenyuan Think Tank, told Shenran that the decline in smart speakers is partly due to product iterations mainly due to software and cloud updates, hardware iterations are slow, and users have a relatively long replacement cycle. From the perspective of the supply side, smart speakers are no longer a business with a high strategic priority in the business ecology of the top three companies.

At the same time, manufacturers are also continuing to use the core "voice assistant" function of smart speakers in various smart home products, including air conditioners and TVs. However, most smart speakers cannot be interconnected with other smart homes to establish a smart ecology, which cannot meet the expectations of manufacturers and become an important entry point for smart home ecology.

Yuan Shuai, a senior data analyst, said that the biggest reason why smart speakers can't be sold is that consumers' cognition has returned to rationality, and this kind of consumer goods that are not just needed are not consumers' preferred products. In the past, several giants believed that smart speakers would be the entrance of the future "smart Internet of Things". After early rough subsidies and share grabs, they found that smart speakers are rarely used for the Internet of Everything, and play the role and value of Internet product company positioning. In the preparation of a full set of smart homes, smart speakers even play a "dispensable" role. Although smart speakers have been launching new products, the speaker products themselves have not seen substantial leapfrog upgrades except for the expansion of the screen size.

Under the influence of various factors, the smart speaker market began to decline. Not only in China, but even overseas, Amazon, which once set off a boom in smart speakers, is not having a good time in this business. In November 2022, it was revealed that Amazon planned to lay off 10,000 employees. According to market feedback, at that time, the Worldwide Digital department, where its voice assistant Alexa was located, was the hardest hit area for layoffs.

Behind the layoffs is related to its losses. Amazon's Worldwide Digital unit, which includes everything from Echo smart speakers and Alexa voice technology to its Prime Video streaming service, posted an operating loss of more than $3 billion in the first quarter, according to internal data obtained by Insider.

AI large model, a life-saving straw

Entering 2023, there will also be new undercurrents in the smart speaker market. Under the wave of ChatGPT, how the AI model will stir up the smart speaker market has become the focus of the market.

Big factories have already run ahead. Ali and Baidu, which run faster on AI large models, have also disclosed their plans to the outside world. In February, Xiaodu announced that it would integrate Wenxin Yiyan to create an AI model "Xiaodu Lingji" for smart device scenarios, with functions such as super assistant and smart housekeeper.

In the demonstration video, faced with the changing schedule and deliberate embarrassment of the staff, for example, after inputting "I am going to see my mother on weekends, she said that I will bring her peanut oil and dark soy sauce", the staff then asked Xiaodu for inspiration, "I went to see if my mother brought soy sauce or dark soy sauce", Xiaodu Lingji could accurately capture the main point, "I should bring peanut oil and dark soy sauce".

In April, Tmall Genie connected to the personalized model of "Niao Niao Di Niao" to create an "AI mouth replacement", which has been able to realize the tone and intonation very similar to the talk show actor Niao Niao, with anthropomorphic timbre, tone, and expression. In addition, Tmall Genie also announced the access to Alibaba Tongyi Qianwen.

Xiaomi also mentioned in its financial report for the first quarter of this year that Xiao Ai is a typical scenario for implementing AI large-scale model capabilities. In mid-April, Lei Jun also issued a special article pointing out that some technologies and products are already being developed, and he will firmly embrace large-scale model technology. Although Huawei has not made a clear statement on smart speakers, it has a large model of the Pangu series, and the market share of smart speakers is second only to the top three in China, and it will inevitably keep up with this wave in the future.

Overseas major manufacturers have accelerated their deployment early. Google, Amazon Alexa, and Apple have all made it clear that they will increase investment in AI technology. It is only a matter of time before the combination with AI large models in the future. Among them, Google was revealed to be reorganizing the reporting structure of its virtual assistant department Assistant to focus on the research and development of its previously launched chat robot Bard.

ChatGPT technology has greatly improved in terms of context understanding, multi-round dialogue, content generation, etc., which is also the direction of improvement for smart speakers.

Wan Yulong, chief architect of OPPO Xiaobu Assistant, told Shenran that the voice interaction link can be roughly divided into four stages: perception-cognition-expression-execution. Among them, perception is to understand sounds and images through hearing and vision; cognition is to judge user intentions, context logic, etc. by understanding semantic information, and determine the reply content; expression is to express the reply content through voice, vision, etc.; execution refers to While expressing, certain commands can be executed, such as playing a specific song, calling up a specific application, and so on. **The emergence of large language model technology is mainly due to changes in the cognitive stage, which can better understand user intentions and generate more reasonable responses. **

Deng Xintao believes that the most essential change of the AI model for smart speakers is that there will be certain improvements in voice input, recognition and output. Smart speakers can accurately identify the user's intentions through artificial intelligence analysis of language and context, which may further expand the application scenarios of smart speakers.

Wang Chao pointed out that in the past, it was difficult for smart speakers to achieve multiple rounds of dialogue, but in the future, the dialogue interaction will be smoother, and even the realization of specific functions, including teaching children to do math problems, will become a reality.

In the future, how can users experience smarter smart speakers? Wan Yulong believes that the service capability of the big language model can be switched in the cloud, and users may be able to experience it directly through software version upgrades. However, due to the relatively high cost of computing power of large language models, manufacturers may use this as a new commercial growth point. As for whether to charge according to the number of services in the future, or to adopt a subscription system similar to that of video website members, manufacturers may find the best way after calculating the business model. It will take some time to achieve a smooth transition from product experience to user payment.

However, Deng Xintao pointed out that if you want to access stronger AI capabilities, smart speakers may use new chips, and smart device terminals need to have stronger computing power, decision-making capabilities, and recognition capabilities. What used to be a "smart speaker" is unlikely to plug directly into it. It is also difficult for merchants to update and upgrade sold products.

But in any case, the AI model will become a new variable in the smart speaker track. As Wang Chao said, "With the maturity of AI large-scale model technology, the smart speaker industry will surely usher in a second rise in the future."

The relevant research report of Zheshang Securities also pointed out that the "personalized large model" is expected to become the key to detonating the next round of AIoT product innovation, not only voice assistants, but also various ChatGPT products such as text, image, voice, video, etc., are expected to achieve more The innovation of application scenarios and the empowerment of AIoT by various ChatGPT products are expected to bring the development of new AIoT products into the era of "Cambrian Explosion".

Can smart speakers still catch fire?

Even though the access to the AI model will bring some product changes to smart speakers, the product form of smart speakers still faces many challenges.

**One of the cores of the smart speaker competition is its voice interaction technology. **

Wan Yulong believes that for users, voice assistants mainly assume three types of roles, namely: tool-assistant-friend. Many users are accustomed to using voice assistants as a tool to conveniently control devices or query information; many users expect voice assistants to better understand themselves and provide services according to personal preferences, such as itinerary recommendations, schedule reminders, etc.; Some users have even begun to regard voice assistants as friends in the digital world, obtaining emotional companionship and sustenance through chatting and other means.

After the smart speaker is connected to the large language model technology, it will make the user's dialogue experience more intelligent. However, it is unknown whether the smart speaker product can meet the deeper needs of users and generate greater value for users. And when users regard it as an assistant and friend, they will pay more attention to privacy and security. Whether this experience can be realized through smart speakers, a family public device, remains to be verified.

At the same time, after the user experience is optimized, the voice assistant will also be carried on more hardware products. At that time, the competitors of smart speakers will not only be the previous smart speakers, but also include all hardware devices equipped with voice assistants in the smart home ecosystem.

Smart speakers will have a place in the entire smart home ecosystem, but they are more likely to be replaced. "Large language model technology can make smart speakers smarter, but it is difficult to make them irreplaceable." Therefore, the large language model technology can improve the interactive experience of the device, but the opening of the smart home ecology is still a problem. The improvement of smart speaker experience may hardly affect the ecology of the entire smart home in the short term.

Jason Low, chief analyst at Canalys, told Shenran that there are still certain hidden dangers in the content generation of the current AI large model. For example, if consumers want to use smart assistants, they often require that the content must be accurate, but now large models including ChatGPT still give some things out of nothing. It will take some time for manufacturers to improve these problems before they can be implemented on a large scale.

In addition, smart speakers are currently facing many core product issues, including data security, privacy security, and the adaptability of content link recommendations.

As early as 2017, Amazon's smart speakers were exposed to a large number of vulnerabilities. Although they were repaired urgently at that time, they did not fundamentally solve the problem of privacy and security. In the following years, smart speakers have been exposed many times, and there are technical defects. Hackers can use such defects to activate and hijack the speakers, eavesdrop and steal the user's conversation content.

Although technology has been improving, the risks of data security and privacy security have not been eliminated. In the future, if smart speakers are to assume an increasingly important role, the user experience will become more and more in-depth, and the loss of personalized data will definitely bring serious losses to users, and even endanger personal safety. Problems loom.

Yuan Shuai believes that after the staking of the past few years, the smart speaker market has gradually become saturated. Next, track manufacturers should focus on improving product functions and experience rather than grabbing market share. Smart products with single form or single function and poor user experience will be eliminated by the market.

*At the request of the interviewees, Xiaobai and Lili are pseudonyms in the article.