“The beautiful picture outlined by the future Internet of Things is slowly unfolding, and the smart home that has become the main driving force is in its prime. Among all the “keys” that open the door to smart homes, voice has become the most popular. According to ReportLinker, by 2024, the global smart voice market will reach US$21.5 billion.
The beautiful picture outlined by the future Internet of Things is slowly unfolding, and the smart home that has become the main driving force is in its prime. Among all the “keys” that open the door to smart homes, voice has become the most popular. According to ReportLinker, by 2024, the global smart voice market will reach US$21.5 billion.
However, if you equate voice with a smart speaker, that would be a thousand miles away. As a central control device, smart speakers are just an entrance to voice. From the living room to kitchen appliances, bathroom appliances and other application scenarios, the voice can actually achieve “single machine intelligence” without the speaker, and the voice chip is its enabler.
Break the inherent thinking voice ≠ cloud
Although when talking about voice solutions, most of the industry’s inertial thinking is focused on the level of networking and cloud, but in fact, applications can only stimulate new insights when they reach the level of market segmentation requirements.
Lu Yong, CEO of Exploration Technology, once deeply analyzed the demand for voice in smart home products. Taking the common Internet TV as an example, the technology path is divided into two. One is the voice recognition link to obtain user instructions, and the other is the content acquisition link, which executes instructions to obtain cloud-side audio and video.
Further analysis, we will find that, in fact, among all kinds of household appliances in sub-scenarios from smart living room to smart kitchen, smart bathroom, etc., only a few household appliances such as TVs and speakers need to obtain audio-visual content. Most products such as lamps, switches, and air conditioners may add such functions to some high-end product lines, but they are not just needed.
It can be seen that the voice needs of smart homes can be divided into two categories, one is to obtain content through voice, and the other is to control home appliances through voice. There are not many categories of home appliances that need to obtain content, and the content acquisition link has nothing to do with voice technology. The quality of audiovisual content depends on the quality of the film source on the cloud platform.
The technical path of controlling electrical appliances can be divided into two types: “remote control” and “voice control”. Among them, remote control realizes app networking through wireless technologies such as Wi-Fi or Bluetooth, and further controls electrical appliances. The key node of the technology is cloud networking, and remote control is actually the extension of control functions through networking.
It can be seen that in the field of smart home, the ultimate goal of smart interaction is to control home appliances, and the only thing strongly related to it is “voice control”.
“Voice control” itself can be subdivided into “command type” and “natural type” (that is, NLP natural language processing). “Command type” can be implemented offline or online. At present, NLP is mainly implemented through the cloud side.
“Just like everyone strongly associates cloud with speech, in fact, many people have mixed the two problems of’cloud’ and’NLP’ together, thinking that natural language recognition has to go to the cloud.” Lu Yong believes that the future will follow the algorithm model With smaller and smaller chips and more powerful chips, the realization of offline recognition by NLP will be just around the corner.
It can be seen that speech recognition is not equal to the cloud. Speech recognition and cloud computing are two levels of things, one is specific technology, the other is basic computing facilities. The implementation of speech recognition technology is divided into two types: offline and online. The difference between the two is whether the calculation is performed on the end side or in the cloud.
In fact, in the field of smart home, offline voice solutions have their specific advantages.
The universal path of offline voice
In the not-too-distant future, the Internet of Everything will give birth to hundreds of millions of devices. If all computing is placed in the cloud, it is not only expensive, but also difficult to guarantee efficiency. Especially in the special scene of smart home, it has extremely high requirements for real-time, stability and privacy. In consideration of cloud data processing capabilities, network latency, and data security, the “decentralization” of computing Power to edge computing close to the terminal will see rapid development.
If this is a product design consideration, then from the perspective of the ecological chain, for the majority of home appliance manufacturers, going to the cloud means either opening the traffic portal or big data to third-party ecological chain companies, or building their own private cloud portal , It is necessary to consider ecological compatibility, and there is a certain threshold for research and development, and the end-to-side solution does not have to consider many concerns, and can let it go.
For C-end users, the benefits of offline voice are also obvious. Users do not need to purchase and use centralized control hardware such as smart speakers, nor do they need to consider the compatibility of different brands with cloud platforms, nor do they need to consider privacy issues. More importantly, the offline solution does not need to rely on the network, there is no delay, and the recognition is accurate, thereby reducing the threshold for users.
Lu Yong believes that the development direction of smart home should be “realize intelligence first, and then consider ecology”. The ecology that should guarantee product ease of use should not become an obstacle to restrict users’ use.
Offline voice makes smart home appliances become like USB, plug-and-play, without any barriers to use, which will fundamentally improve the final user experience and greatly eliminate the end consumers’ worries. In addition, offline voice can be applied to almost all home appliance categories, making every home appliance a truly intelligent device, and transforming smart home from a niche product into a universal product.
It is precisely by virtue of its deep technical background and keen market judgment that the off-line voice recognition chip of Exploration Technology-Yinxuanfeng VOI611 quickly opened the door to the market. The competitive advantage of VOI611 in the market is very obvious: it supports 200 command words, has a wake-up rate of 99% and a recognition rate of 97%, the false wake-up rate is less than 1 time/24 hours, and the response time is less than 0.2s. 10 meters far-field recognition distance, with accurate recognition effect.
In addition, the price of the voice chip of Exploration Technology is almost the same as that of ordinary MCUs, and there is no research and development threshold, which will undoubtedly greatly accelerate the speed of intelligentization of home appliance companies.
At present, the smart home products covered by Tanjing Technology include: smart lamps, smart switches, air conditioning companions, voice fans, air purifiers, drying racks and other categories. Cooperating manufacturers include well-known companies such as Midea, Haier, Xinyi, and Airmate.
At the same time, Lu Yong emphasized that end-side and cloud are not all-or-nothing single-choice questions. Under the premise of existing end-side voice solutions to achieve stand-alone intelligence, whether to combine Wi-Fi modules or add cloud recognition? For manufacturers, it is just a multi-choice question that can be freely combined.
At present, the industry has been aware of the many drawbacks of online solutions. Off-line solutions such as “offline + app” and “offline + cloud recognition + app” have already seen signs. Lu Yong believes that as the algorithm model of the voice solution becomes smaller and smaller, and Chip performance is gradually improving, and the technical barriers of end-side NLP can be breached. When offline NLP technology matures in the future, smarter and customizable offline NLP voice solutions will be the best choice for home appliance manufacturers.
There are counts in the future “cores”
Seeing the anchor point of the offline voice chip, the exploration environment has already been deployed in a long-term, and preliminary research and development results have been achieved.
In the fourth quarter of 2020, the chip product codenamed Yinxuanfeng II internally by Tanjing Technology has been successfully taped out. Compared with the first generation, the second generation of Yinxuanfeng has stronger computing power, can run a larger neural network model, and has lower power consumption and lower price. In addition, Tanjing’s end-to-side NLP products have also emerged.
Relying on the “Storage First Architecture” (SFA architecture) to solve the unique hardware architecture of the storage wall, supplemented by the algorithm refinement to open up the full link of speech recognition, the voice chip of Tanjing has a good experience, low power consumption, high cost performance, and practicality. And other advantages.
Lu Yong mentioned that Tanjing can not only provide customers with a variety of voice solutions such as chips, algorithms, and Turn-Key, but also support customers in secondary development and algorithm migration. At the system software level, different levels are provided for different customers, including instruction sets, SDK development environments, or application-level voice/image algorithms, and the deployment process can be personalized for customers.
In Lu Yong’s view, humanity must be respected for any product. People need companionship and emotional communication. This is also the ultimate development direction of exploration. Voice and vision are the most convenient and most human interaction methods in nature. Environment Technology will not simply pursue a certain AI technical index, but will make a chip solution with a temperature, and promote life with AI technology.
“How each drop of water will enter the valley when it rains, this route is definitely unknown. But you must know the direction-because there is gravity, it must go downward. Similar to the inevitable “gravity”, the business trend is also Inevitably, the overall trend must be predictable. “Kevin Kelly, the father of the Internet of Things, summed up the importance of trends in this way.
After being tempered by the market, Tanjing will also be more “core” in the future. In the future, Tanjing upholds the concentrated insight and analysis of the market, coupled with the continuous innovation of hardware and algorithms, it will also usher in a new bright moment.