Then by hardware dream, ByteDance enters the AI ​ ​ headset

Our reporter Li Jing reports from Beijing.

ByteDance officially entered the AI intelligent hardware market by leveraging the big model.

Recently, ByteDance’s Bean Bag released Ola Friend, the first AI intelligent headset. The introduction shows that this headset is connected to the bean bag model and deeply integrated with the bean bag App. After the user puts on the earphone, he can arouse the bean bag to have a conversation through voice without turning on the mobile phone.

Since the official release of ChatGPT on November 30, 2022, the domestic large-scale model market has also been surging in the past two years, and many enterprises have invested in research and development, striving to occupy a place in this emerging market. At present, there are many basic large models and vertical large models in China. In the B-end market, large models are gradually integrated into daily business and workflow, accelerating the empowerment of medical, financial, manufacturing and other industries; In the C-end market, applications such as chat bots, search and content generation are becoming more and more popular.

In fact, in addition to the rapid development of software applications, many enterprises are also exploring the combination of large models and hardware.

Some PC manufacturers have launched AI PC products, mobile phone manufacturers are also developing AI phones, and many companies are exploring the combination of AI and wearable devices. At the end of last year, Meta released Ray-Ban Meta, a smart glasses with built-in Llama model. Companies such as Apple and Midjourney are also exploring the combination of head display and AI. In the domestic market, Iflytek has released several AI headsets.

For Ola Friend, the first AI agent headset released this time, the person in charge of Bean Bag said: "This headset is an exploration and attempt of Bean Bag in the AI ​ ​ scene. I hope Ola Friend can become a friend who can accompany users at any time. The various abilities of bean bags will continue to iterate in the future to help users in various scenes in life. "

How about headphones, AI?

From the point of view of starting the AI function of headphones, users often need to use mobile phones or computers to input traditional large-scale AI dialogue products. While Ola Friend, the AI agent headset released by ByteDance this time, users can communicate with each other through voice by simply touching the headset or saying the wake-up word "bean bag bean bag" after wearing Ola Friend headset.

According to the person in charge of bean bag, in the headset user test, many users gave feedback. Because Ola Friend is very light and comfortable to wear, it has universal intelligence after accessing bean bag, so it has a good experience in tourism, English learning, chat and other scenes. For example, when visiting museums and art galleries, users can ask them about the origin and background of exhibits and artworks. It can also be extended to topics such as dynasty changes and artists’ ideas, and to some extent, it acts as a tour guide, which is very convenient. In addition, for some "whimsy" chat and emotional expression, such as "I have an important speech, what should I do if I am a little nervous?" "I just ate the eight-treasure rice, as if I had returned to my childhood", and its reply was also very kind.

In addition, Ola Friend has made some optimizations to enable users to communicate with AI just like chatting with friends. In the earphone, the tone of the bean bag can show happiness, surprise and other emotions. Moreover, users can "talk while listening, interrupt at any time" and switch topics at any time.

According to reports, compared with smart speakers and other products, the use environment of headphones is more complicated. In order to do a good job of speech recognition, Ola Friend headphones are connected to the Seed-ASR (Speech Recognition) technology model of the Byte Bean Bag model, which can recognize Chinese and English accents with high precision, and even "intelligently" recognize all kinds of information through context.

Judging from the overall AI function of Ola Friend headphones, it is more inclined to be used in consumers’ personal life scenes. In April this year, Cleer, an intelligent acoustic brand, also released the world’s first open AI headset CleerARC3 sound arc, which is also aimed at consumers’ personal life scenes. The built-in AI motion algorithm in the headset can help users monitor sports physiological data in real time. In the interaction, Mobvoi’s voice control technology is integrated, and the AI voice control is upgraded. You can use shortcut passwords such as "Next" and "Answer the phone" to perform corresponding operations without prompting the voice assistant. In addition, the AI noise reduction effect is achieved.

AI headphones launched by Iflytek in recent two years are mainly aimed at office scenes of individual consumers. In 2023 and 2024, Iflytek successively introduced several AI conference headphones with real-time transcription, multilingual transcription and text processing.

From the price point of view, both Ola Friend headphones from ByteDance, Cleer ARC3, and AI headphones from Iflytek are priced above 1000 yuan, belonging to mid-to high-end products.

Pan Xuefei, research director of IDC China, told reporters: "At present, the combination of AI big model and headphones mainly drives the development of the middle and high-end market of Bluetooth headsets. If we cut into specific scenes effectively, we can find and reach the target users and solve certain pain points."

Why do you enter from headphones?

With the development of AI model and the release of terminal AI chips by AMD, Intel, Qualcomm and MediaTek, the development of AI has turned to hardware+software parallel driver, and many manufacturers have explored the combination of AI and hardware devices.

On the PC side, AI PC of Lenovo, Hewlett-Packard and Dell are being shipped one after another. On the mobile phone side, flagship models such as Samsung Galaxy S24 series and Xiaomi 14 have integrated AI functions, such as instant call translation and AI photo taking. On the wearable hardware side, at the end of last year, Meta released Ray-Ban Meta, a smart glasses with built-in Llama model. Not long ago, Solos, an American smart glasses company, said that it would launch the world’s first smart glasses AirGo Vision with integrated GPT-4o, and companies such as Apple and Midjourney are also exploring the combination of head display and AI. In addition, there are AI headphones mentioned above.

As can be seen from the products, the main force of AI PC and AI mobile phone production is still the traditional PC manufacturers and mobile phone manufacturers, while Internet vendors such as Meta and ByteDance are more concentrated in the field of AI wearable devices.

Why did ByteDance bean bag model choose AI earphone as the entrance?

According to the report of Southwest Securities Research and Development Center, compared with end-side devices such as PC, mobile phone and VR/AR head display, headphones have obvious portability advantages in edge-side AI terminals. At present, headphones are mainly connected through Bluetooth, and they need to be connected to the cloud by mobile phone or PC for interaction. In the future, headphones may develop to Wi-Fi connection, and they will be able to access the edge AI model anytime and anywhere.

From the perspective of incremental market, the incremental market of Bluetooth headset is relatively more than that of PC and mobile phone. Huaxin Securities reported that smartphone shipments decreased by 3.3% year-on-year in 2023. On a quarterly basis, global smartphone shipments narrowed quarter by quarter in the first three quarters of 2023, and turned positive in the fourth quarter of 2023. The reasons behind it come from two aspects. On the one hand, there is a lack of innovation in mobile phone hardware. After the end of the 5G cycle in developed economies and China, the consumer replacement cycle is lengthened; On the other hand, after 2022, consumer demand is weak.

As for wireless headsets, according to customs export data, there has been a recovery trend since 2023. Since February 2023, the monthly growth rate of wireless headsets has continued to be positive, and the year-on-year growth rate has continued to expand since September 2023.

"Because the wireless headset technology is fully mature, there is still room for the popularization of wireless headsets compared with the consumption of mobile phones. With the increase of wireless headset sensors, the product experience will be even better, the superposition value will be smaller than that of mobile phones, and the replacement cycle will be significantly faster than that of mobile phones." Huaxin Securities believes that with the opening up and economic recovery in China, we will continue to be optimistic about the growth of wearable devices such as wireless headsets.

"With the increasing demand of consumers for intelligent services, AI headphones, as an emerging intelligent hardware product, have strong market demand. Domestic large model manufacturers can meet market demand and expand new market space by launching AI headsets. " Wang Peng, an associate researcher at the Beijing Academy of Social Sciences, also pointed out to reporters that compared with other wearable devices or hardware devices, China’s headset technology is mature and the industrial chain is complete. Domestic large model manufacturers have accumulated rich experience in AI technology, and can combine AI technology with earphone hardware to create competitive AI earphone products.

ByteDance’s Hardware Dream

Internet giants have more or less some dreams of making hardware, and ByteDance is no exception. Although ByteDance already has well-known software products such as Today’s Headline, Tik Tok, Feishu, etc., so far there is no hardware product with a relatively high market share.

At the end of 2018, ByteDance acquired the mobile phone business of Hammer Technology and some patent rights for more than 100 million yuan, and established the Xinshi Laboratory, with Wu Dezhou, the former head of mobile phone of Hammer Technology Nut, as the president. After the acquisition, Xinshi Lab released two generations of nut mobile phones, new TNT displays and speakers and other peripheral products.

In 2020, ByteDance established the brand of "Vigorously Educate" and began to lay out table lamp products in educational scenes. In January 2021, Xinshi Lab was merged into the educational hardware team, giving up the mobile phone business and gathering educational hardware. The reasons behind this are, on the one hand, the concentration of the mobile phone market was further improved at that time, and on the other hand, the market scale of educational hardware was expanding due to the epidemic situation and the outbreak of online education. At that time, according to the prediction of Duowhale Capital Education Research Institute, by 2022, the market scale of K12 intelligent education hardware alone may reach 57 billion yuan.

Unfortunately, in August 2022, due to the influence of the national "double reduction" policy and the "three children" policy, vigorously education began to lay off employees, and some people changed jobs or invested in new projects. ByteDance’s educational hardware products have also been shelved.

In September, 2021, ByteDance also acquired Pico, the head display manufacturer, for 9 billion yuan. Although ByteDance has high hopes for Pico and invested a lot of money in it, the market performance of Pico has not reached expectations. In 2023, ByteDance adjusted the organizational structure of Pico, drastically laid off employees, and lowered its sales expectations. Pico continues to introduce new products, such as the PICO 4 Ultra MR mixed reality integrated machine released in 2024. But overall, Pico has been gradually marginalized in ByteDance’s strategy.

Today, ByteDance’s hardware exploration is more focused on the AI field. It is understood that the exploration of ByteDance’s AI hardware direction is divided into two product lines internally: one product line is code-named "D line", with emphasis on wearable devices with AI capabilities; The other product line is "O-line", which focuses on handheld AI hardware devices. Hand-held devices have not yet taken shape.

At present, Ola Friend smart headphones, which are positioned in the mid-to high-end market, have been launched on major e-commerce platforms, and officially shipped from October 17, with a price of 1199 yuan. At present, on the JD.COM platform, Ola Friend’s self-operated flagship store in JD.COM has sold more than 500 headphones; On the Tmall platform, Ola Friend intelligent headphones have sold more than 1,000, and ranked second in the hot-selling list of bone conduction headphones above 300 yuan on the Tmall list. It remains to be seen whether more consumers will like it in the future.

Pan Xuefei pointed out to reporters: "At present, the combination of large models and hardware is still mainly in the form of access. If we want to achieve the end-side, on the one hand, we have higher requirements for chip size and space design in terms of volume and ergonomic design, and on the other hand, we have extremely high requirements for battery size and endurance. At present, these aspects still face great challenges."

(Editor: Zhang Jingchao Review: Li Zhenghao Proofread: Zhai Jun)