Future Voice Chips on the Road to AIoT

From the ZTE incident last year to today's Sino-US trade dispute, one word has always been at the center of the storm of public opinion: chips. Behind the voice, there are also many secrets about chips.

From last year's ZTE incident to today's Sino-US trade dispute, one word has always been at the center of the public opinion storm: chips.

Surrounding the foundation of this information society, the crystallization of human technology that makes people love and hate it, any Chinese person you pick today can talk to you about "bottlenecks", rare earths, the global industrial chain, and the "entity list".

However, generally speaking, the chips that people often refer to nowadays refer to the CPU, GPU, etc., which are the lowest level of a computing device. In fact, there are many types of chips, and the chips that are in the midst of the Sino-US trade dispute and global technological changes today are far more than these "basic models".

For example, the "voice chip" that we use every day but don't pay much attention to in daily life. Its technological upgrade and trade status have actually quietly come to the forefront.

Behind the sound, there are also many secrets about chips.

Small chips hiding behind the sound and trade with big countries

The so-called voice chip, or voice IC, refers to the processing module in an electronic device that is responsible for emitting and recording sounds.

In fact, this thing is everywhere in your home. From TVs, computers, and stereos to mobile phones, voice recorders, and music players, even the "beep" sound of induction cookers, refrigerators, and washing machines, all require special voice chips to control them.

This field may not sound very technical, after all, which device today doesn’t make sounds? But in fact, the design technology and customized production capabilities behind it, plus the low-cost requirements attached to the huge number of applications, have led to the fact that today China’s voice chip market still relies mainly on imports.

At present, only about 16% of the domestic voice chip market demand can be met by domestic manufacturers. my country imports more than 200 billion US dollars of voice chip products every year, which exceeds the total amount of crude oil imports in my country each year.

If we put this situation into the context of today's Sino-US trade dispute, it is not difficult to find that this is another unsafe "bottleneck" industry. Due to heavy reliance on imports, lack of underlying design technology and large-scale mass production capabilities, once voice chips are involved in trade uncertainties, it is likely to bring a chain reaction to the semiconductor industry and even the entire electronics industry.

In order to get rid of the downstream position of the industrial chain, China's related industries have also been carrying out a series of actions in recent years, such as large-scale acquisitions of voice chip companies and patented technologies in the United States and Europe, but such acquisitions are often interrupted by various factors, and the abortion rate is much higher than the success rate.

On the other hand, most of the voice chips purchased in the mainland market today are actually produced in Japan, South Korea and Taiwan. Against the backdrop of the Sino-US trade war, it has become relatively important to consolidate and develop industrial linkages with these countries and regions.

The flow of talents in this field has also become a new trend in the development of voice chips in the mainland in recent years. For example, as Taiwan's economic development slows down, a large number of Taiwanese voice chip talents choose to work in the mainland, which objectively becomes a booster for the development of the mainland's own industrial chain.

However, among many factors, the one that is most likely to change the global trade structure of voice chips is the rise of new technologies. From AI to 5G , behind these familiar names, voice chips have ushered in a lot of "variables".

A springboard from AI to 5G

We know that traditional voice chips are only responsible for recording and playing sounds, and their main difficulty lies in transmission accuracy, signal stability, etc. However, a newly emerging market in the past two years has allowed the "old" voice chips to see the possibility of change.

That is AI and smart speakers. In 2018, global shipments of smart speakers reached 56 million units. This also means a chip market worth tens of millions.

In the beginning, smart speakers also used general-purpose CPUs to complete computing tasks, but this approach soon exposed its drawbacks. Since there were not many tasks that required CPU processing in speaker products, most of its energy efficiency was wasted, and consumers had to bear a considerable cost. Soon the industry reached a consensus that dedicated AI voice chips should be the standard configuration of smart speakers.

Compared with traditional voice chips, the main difference of AI voice chips is that they must add voice recognition function. This involves the understanding of AI algorithms, high compression of sound files, and new control relationships with microphone arrays, etc. These capabilities involve multiple disciplines and are brand new functions that were not originally involved in voice chips. The good news brought by the new demand is that companies of all sizes and countries in the industry have come to almost the same starting line.

If there are only tens of millions of speakers in this market each year, then compared with the global voice chip market, it is just a drop in the ocean. The factor that really makes AI technology more imaginative in the voice chip market is 5G.

We all know that the goal of 5G is to enable industrial and business applications, allowing a large number of IoT devices to access large bandwidth and low latency networks. However, it is actually useless to allow these devices to access the network, or at least the usefulness is very limited. The real value of 5G is to enable IoT devices to access intelligent interactions, allowing the IoT system to understand human needs, and to bring overall value enhancement based on the real-time interoperability brought by the 5G network.

For example, connecting the air conditioner at home to the Internet has no value in itself, since no one would climb up to the air conditioner to access the Internet anyway. At most, it can allow people to remotely control the air conditioner, which is probably a useless function. However, if AI is added to the air conditioner, allowing humans to control the air conditioner by voice, and allowing the air conditioner to understand the owner's more complex needs, such as "My shoulders are a little cold and my legs are a little hot", then the irreplaceable role of the Internet of Things will emerge. The human-computer interaction capability here comes from AI, and the network infrastructure is built on 5G.

Following this logic, it means that although speakers are the core market of AI voice chips today, countless electronic devices may need AI voice chips in the future. Is it even possible that today we are actually used to all screens being touch screens, and one or two non-touch screens must be clearly written next to them. In the future, every voice chip may need to be an AI voice chip - humans need every device to not only be able to speak, but also to communicate and understand.

If this logic holds true, the voice chip industry chain will face a large-scale reshuffle, and companies that can only produce traditional voice ICs will not be able to gain a foothold. Today is the initial stage of industrial change, and it is also one of the biggest opportunities for the Chinese voice chip industry to change its position in the industry chain.

Of course, Chinese companies are not the only ones eyeing this opportunity, but at least everyone has a relatively equal chance. At this moment, several groups have rushed into the nascent AI voice chip market.

First of all, chip manufacturers. With the rise of smart speakers, chip manufacturers who seized the opportunity have ushered in another round of growth. Currently, the company that has made the most profit is MediaTek in Taiwan, which has occupied more than 70% of the global smart speaker chip market share. Chip companies such as Hangzhou Guoxin and Amlogic Technology in mainland China are also making efforts in this field.

Generally speaking, the advantage of chip manufacturers is that AI voice chips do not require particularly difficult technical barriers, and have very strong requirements for mass production capabilities and cost control capabilities, which are the strengths of chip companies. However, the understanding of new technologies, especially the integration of AI algorithms, has become a shortcoming of chip manufacturers, so cooperation with AI companies has often become the mainstream today.

On the other hand, the famous domestic AI unicorns are also on the road to AI voice chips. Companies such as KEDA Feixun, Unisound, and Mobvoi have launched their own voice chip plans. The advantage of AI companies is obviously in their understanding of the software layer, such as sound modeling, NLP, semantic understanding, etc. Unicorns often have their own secrets. On the other hand, since the industry generally believes that the business of algorithm companies is too "light" and it is difficult to achieve long-term development in the future, the mainstream story this year is that algorithm companies are moving towards chips and hardware, integrating algorithms into chips, and getting stuck in the upstream of the industry. This also strengthens the motivation of AI companies to enter the chip market.

However, AI companies are also clearly weak in chips. AI voice chips are not high-net-worth products and often rely on mass production and large-scale shipments to be profitable. In terms of integration and engineering capabilities, AI companies are clearly at a disadvantage. Currently, AI unicorns' voice chip plans are still mostly at the stage of affirmation and commitment to mass production.