If you want to summarize the status quo of AI ventures in the past two years, financing and core building are inseparable topics. The word chip is very likely to be selected for the 2018 annual keyword.
The heat that started last year, under the catalysis of the ZTE incident, has a more rapid chemical reaction. Among them, the most intriguing is the rapid rise of AI voice chips:
From May to July, in just two months, from the publicly reported data, more than five companies announced that they have made AI voice chips:
So, why do you all agree to release the AI voice chip at this time node? What is the logic behind this?
Fuse: A small explosion in the smart speaker market
This year's 618 promotion, Tmall Elf played a price marketing war, only need 99 yuan to buy a smart speaker. Soon, on July 5, the Tmall Elf announced a year, their Total channel sales totaled more than 5 million units.
The behavior of this giant to drive down the price of the price quickly caused a chain reaction. The discussion about the smart speaker outlet was very rampant. The gust brought by Amazon finally reached the domestic market.
According to the latest research report released by Strategy Analytics, in the first quarter of 2018, the total sales of global smart speakers reached 9.2 million, an increase of 278%.
'When the amount of equipment just got up, everyone suddenly realized that the chip is a very important part, put its necessity to a higher position.' As the earliest beginning to lay out the AI voice chip, Yunzhisheng, its founding Ren Huang Wei talked about the recent chip heat.
This round of smart speaker market explosion, let many people see the potential AI voice chip market opportunities.
According to the foreign media Information report in March this year, Amazon is designing an AI chip customized to support the smart speaker Echo. At the time, it was reported that Amazon had 449 employees with chip expertise and skills.
Coincidentally, Zhongtianwei, which was just acquired by Ali, also announced that it would release a smart voice chip in early July.
The potential action of the giant is one of the most important market vane, and this fuse has naturally burned the AI voice chip, exploring the logic behind it, and also the advantage of the AI voice chip compared to the traditional general-purpose chip.
In fact, the earliest general-purpose chips play a small role in voice. Usually, multimedia digital encoder + digital signal module processing is combined.
At the end of 2014, Amazon's Echo came out. Some semiconductor manufacturers aimed at this market and started to introduce voice chips. The most typical one is MediaTek. It is understood that at that time, some people speculated that nearly 80% of the chips in the smart speaker market in 2016. They are all provided by MTK, and this is thanks to their deep cooperation with Amazon Echo.
When the requirements of intelligent hardware for voice interaction are getting higher and higher, many things need to be implemented on the end, such as wake-up, data signal processing. Considering security, network conditions and other factors, the emergence of AI voice chip is an inevitable result.
Compared with voice chips, AI voice chips have high integration, low power consumption and low cost, which can achieve the perfect combination of algorithms and terminals.
When Rokid co-founder Wang Yude answered why he would do AI voice chip, he mentioned 'because we know the pain of making products, knowing the price of the chip will drive the whole product, and the chip at that time is very power-consuming and low in integration. After experiencing these pain points, we want to optimize the design of the chip and use our front-end algorithm.
Algorithm-Chip-Hardware: The Necessity of Commercialization
Carefully sort out the ideas of several major AI ventures to do AI voice chips, most of them choose to cooperate with experienced chip companies.
For example, when asked, Rokid announced that his chip is based on the deep customization of Hangzhou Guoxin Technology Chip. Among them, Rokid's KAMINO18 is based on the 40nm process Guoxin GX8010. The GX8010 is the main AI interactive NPU chip released by Guoxin last year. Designed for IoT applications, it has the advantages of low power consumption, offline, and mobile. When the company announced the creation of chips, it also mentioned that they will cooperate with a chip giant in the AI chip to form a joint venture company.
The chip company provides low-power, low-integration design architecture. AI Voice Technologies will work on microphone array signal processing, voice interaction SDK and voice noise reduction, wake-up, recognition and understanding, and will have its own AI voice interaction technology. Integrated into it.
Why is the algorithm technology landing, starting from the chip and the subsequent hardware, the reason is inseparable from China's hardware and software environment.
Xu Zhijun, Huawei's rotating CEO, mentioned at the 2018 Soft Expo: 'Domestic customers especially do not accept software charges, which makes domestic software product companies unable to form a business model.'
The software sales model that everyone accepts is a hardware-like model. Domestically, hardware is considered valuable, software is not worth much, and its cost is low.
Therefore, in order to achieve rapid growth of business and business in China, the outbreak of scale, only algorithms, software is very difficult. The latest technologies, solutions, including products through hardware carriers are more likely to achieve large-scale growth and replication. .
Therefore, AI's algorithm is integrated into the chip company's voice chip, which can be said to be a labor-saving and pleasing cooperation, and the AI voice chip is accompanied by a variety of intelligent hardware heat is also a matter of course.
In addition, removing these technical factors and telling the story of the chip will also help AI companies to obtain financing, and have more capital to exert their own strength. Especially the current time node: The chip is both a performance of technical strength and a national sentiment. Symbolic body.
In the context of such a good time and place, the AI voice chip is on the rise of the explosion, which is expected.
Of course, it is not excluded that there are still some followers who want to make a 'net red' in the impetuous market. The heat of the AI voice chip is not the soap. The scale of the industry is not only the technical strength. , there are commercial landing capabilities and risk tolerance.
Is the virtual fire still hot?
In addition to the butterfly effect of the smart speaker and the cause of the chip heat, if you understand the AI voice chip from the demand and industry, you have to start from a broader application scenario and commercial landing.
Wei Shaojun, director of the Microelectronics Institute of Tsinghua University, said in an interview with the media that the killer application of AI has not yet appeared. Whether it is a smart speaker or other products, it has not yet become a necessity. Therefore, only the voice is truly human-computer interaction. The mainstream, in order to promote the outbreak of AI voice chips.
So even though the AI chip is hot, the rational voice that comes with it will ask the real demand for voice technology. Where is the market?
Take smart speakers as an example. Before the home Internet of Things was formed, many people think that it is more like a gimmick hardware. In the current situation, the consumer market is not suitable for voice interaction and recognition. Just need to be discovered yet.
To this end, we have compiled the products and solutions of several major AI voice technology companies:
Looking back at foreign technology giants, they are following a similar path in development. They use hardware or open application platforms through investment or acquisition.
As can be seen from the above table, at present, whether it is like Spirit, Yunzhisheng, go out to ask, AI ventures like Rokid, or giants like Google, Amazon, Apple, they have on the landing of the application scene. Many crosses, mostly biased towards the Internet of Things, centered around smart homes, cars, and robots. The smart home, outside the main battlefield of the Internet of Things, like smart medical care, is also the new frontier that these AI companies are expanding.
At the same time, according to Analysys' report, the intelligent voice market is in a high-speed development period, and the vertical fields based on voice interaction, such as smart car, smart home, and smart wearable, will mature.
In these scenarios, artificial intelligence speech technology is not a very core and indispensable technology, but following the development trajectory of consumption upgrade and technology iteration, the speech recognition and interactive technology carried by the AI voice chip is definitely the trend of the times.
Based on such development path planning and the prediction of a huge consumer market, the AI voice chip is also taken for granted.
Just as a person's body is composed of multiple organs, in many intelligent application scenarios, the role played by the AI voice chip is more of an explicit manifestation of algorithmic technology. The chip acts as a 'hardware' to match its own software solution. Finally, to complete the ecological closed loop.
The key to thorns: technology + data
Doing AI voice chips is a huge investment project. Rokid co-founder Wang Yude said, 'The most important point of the chip is quantity. The key profit of the chip is more than five million.'
So if the enterprise wants to have the ability to self-create blood, what is the biggest bottleneck currently facing?
Wang Haode put forward two points: data and interaction. Among them, the interaction refers to 'now the voice technology even the general white user's industry ideals have not reached', which also means that voice technology is still at a very early stage.
Taking data as an example, one of AI's competitive performances is data. How to achieve deep reflow in the industry is a problem that AI voice technology companies need to solve. Because only after deep data reflow is implemented, the algorithm will be implemented in the industry. More precise, more competitive products.
But in addition to the core algorithms and computing power, the entire artificial intelligence is also very important: the technology, the program, the product should be able to be promoted in the core application scenarios, and ultimately bring the company a realistic revenue.
Indeed, in addition to the integrated solution, the deep integration with the scene is the real test of the future. Yang Yuxin, co-founder of Anchuang Space, said, 'If AI company only makes chips, there will be no algorithms and scenes. Ecosystem issues. Now with algorithms and chips, the key question is how to drill down into the scene to create an excellent voice interaction experience.'
In addition to technology, Si Bi Chi's Gao Shixing also emphasized the importance of the industry's landing. 'Technology and industry must form a cycle, and we must grasp the window period. If the opportunity is over, there will be no more.'
In the window of AI's traditional industry, once a strong enough AI company cuts into an industry, it can rely on data and accumulated industry experience to build its own barriers.
This is also the competitiveness of AI companies in the era of Internet big data: Technology + Data.
Out of the comfort zone, facing the real market
'Beginning a lot of teams want to do what they do best, the best they can do is better, the team is more comfortable, you go from the algorithm to the chip or hardware, you have to break through and get out of your comfort zone, this may be Need a challenge to the self. '
For example, the AI chip only strengthens the deep learning ability, sensor access, signal processing, detection and identification, and software-level decision-making and feedback. The algorithms and computational characteristics required for each link are also different.
From algorithms to chips, hardware, for many start-ups, it can be a big leap, which is why some AI companies will choose to cooperate with the chip company. Because to escape from the comfort zone, you have to put more energy into it. , licked more pits.
Then there is the status quo of the market. It is undeniable that the Tmall Elf sells very well, but behind it is the huge funds of Ali to support it, but undoubtedly this state will not last. When this ecology is removed, many hardware costs are met. For the real cost, go back to a normal stage.
So on the landing of the AI voice chip, everyone will look at the entire Internet of Things field. Yun Zhisheng Huang Wei mentioned in the interview, 'Today, it seems that the number of smart speakers is more, in fact, it is a giant. In desperate subsidies, but that is not true market behavior.'
He mentioned that other intelligent voice scenes made by Yunzhisheng are not like smart speakers, but the vertical contrast still has a substantial increase.
Indeed, if you put the smart speakers in the millions of meters, the order of magnitude is placed in the intelligent voice market, it is only a drop in the ocean. In the high-rise of the company, the amount of smart speakers can not be considered as 'explosives', 'our China and even The global population, each person has several smart hardware in each family. In addition to some industry application scenarios, the terminal of the IoT intelligent hardware will far exceed the smartphone.