Yunzhisheng AI chip manager: The future will enter the image field

Interpretation of goods: It is certain that Yunzhisheng must not only focus on the voice field, the image field will definitely be involved. Li Xiaohan said: 'Artificial intelligence equipment is to make the machine more human, so it must provide a variety of perception, image It is an indispensable link. '

Founded in six years, Yunzhisheng began to work in the field of AI chips.

First released the UniOne series of the first generation of AI chip 'Yuyan', and then announced the 600 million C + round of financing, recently, Yun Zhisheng co-founder, chip technology leader Li Xiaohan officially announced the open source full stack voice interaction program.

As an artificial intelligence company specializing in the voice field, Yunzhisheng had the idea of ​​building chips as early as 2015, and began to form a team. In 2016, Yunzhisheng officially announced the development of chips, which was released in May this year.

Li Xiaohan also said that a large proportion of the new round of financing of Yunzhisheng will be put on the chip. The chip open source voice interaction program is to expand the scope of partners faster.

What kind of effect does open source bring to Yunzhisheng? Pinway Business Review interviewed Li Xiaohan, bringing his thoughts on the field of AI chips.

UniOne Series AI Chip

Li Xiaohan believes that due to advanced EDA tools, FPGA simulation tools, mature IP business ecosystem, and many excellent design service companies, the digital chip design process is becoming more and more mature; and many open source design frameworks and algorithms make The threshold of the chip is greatly reduced, but the threshold for doing well is still very high.

At the same time, the understanding of the application scenario will exceed the digital circuit design capability, which will become the decisive factor for the success of the chip. The understanding of the application scenario, including the understanding of the application and the understanding of the business, is also obvious between the chips. Part of the difference.

In Li Hanhan's view, Yunzhisheng has three key elements in algorithms, scenes and chip design, so it is the best AI chip in the IoT scenario.

Swift is designed and developed by Yunzhisheng. It also includes general-purpose CPU, AI accelerator (DeepNet) and digital signal processor (uDSP) architecture. It adopts autonomous AI command and is oriented to voice AI scene, supporting 6 analog/digital microphone access. Li Yuhan specifically mentioned that the performance of deep neural network is 50 times higher than that of the general scheme.

Swift belongs to UniOne's first generation chip. At the previous chip launch conference, Yunzhisheng mentioned that UniOne will also launch the second generation chip 'Snow Leopard' and the third generation 'Sailfish' for smart car and smart city scenes. Upgrade.

From the current point of view, Swift is divided into two options in the direction of smart home, corresponding to smart speakers and smart home.

Providing customers with software and hardware cloud + end integration solutions is the most common way of cooperation with Yunzhisheng. Previously, intelligent hardware modules shipped in large quantities in the white electricity field served many large companies in this way. Such as the United States, Gree and so on.

After the release of Swift, the solution provided by Yunzhisheng is more three-dimensional, from chip to solution to form a complete solution for customers, and not limited to air conditioners, smart speakers and other equipment. All smart home hardware products can try to access Artificial intelligence technology of Yunzhisheng.

In addition, Yun Zhisheng still wants to play differently.

Open source full stack voice interaction solution

In the smart home industry, both brand manufacturers and suppliers will encounter various difficulties.

For example, if a manufacturer wants to build a smart speaker product, the first difficulty encountered is the supplier's choice.

Because it involves all aspects of speech, noise reduction, recognition, synthesis, etc., not to mention the design of the speaker, sound adjustment... A speaker must be tested after a long time to find a number of suppliers '攒' Products, if a supplier does not achieve the best results, then the experience of the speaker will be greatly reduced.

As a smart speaker, it means that it must be closely related to AI. At present, most people don't have much experience with AI products. There are bound to be many uncertain events in the development process, which is time-consuming and labor-intensive.

'I hope there is a supplier to get all these things done.' This is the conclusion that Yunzhisheng has drawn after investigating many partners.

Correspondingly, due to the cumbersome customer type and product form, it is impossible for the solution provider to support many customers at the same time. Yunzhisheng also thinks of a new solution: Open source.

'A lot of partners have said that we are special 'independent'. ' Li Xiaohan said. Yunzhisheng provides a one-stop solution for many planners who hope to cooperate with them. They feel that Yunzhi is not willing to play with everyone. But Li Hanhan thinks ' Independent' is responsible for the partners.

Because AI landing for smart homes involves a lot of links, such as the need to accumulate structural experience that can be mass-produced; for example, through engine, hardware platform selection and system optimization, to meet the overall power requirements of home appliance manufacturers; A universal chip selection that adapts to the cloud-aware engine and achieves optimal configuration in terms of price and performance.

These need to go through the daily close cooperation between the teams, and sometimes even need to make corresponding engine code level changes for certain hardware features, in order to achieve the best results.

'If you only provide one engine to your partner, and then provide some SDK level adaptation and support, you are irresponsible to your own partners, including your own team. Because everyone has limited resources. Valuable, the engine factory does not have enough hardware, system and product experience, can not effectively support the partners, and finally everyone has done a lot of cooperation, may just be a lively, and can not mass production shipments.

Yunzhisheng will implement the experience and parameters accumulated in the actual landing scene of IVM into the design of its own AI chip UniOne. It is hoped that through the chip, the key parts of the home scene will be cured as much as possible, and then the chip will be The full-stack voice interaction on the open source, greatly reducing the technical threshold, shortening the time to market, thus ensuring the cooperation between partners and Yunzhisheng.

Li Xiaohan uses MediaTek mobile phone solution for comparison: MediaTek provides all the solutions based on MTK mobile phone chip. If you do not modify the outer casing, you can ship it directly. If you want to modify it, you only need to make a simple change. 'The best experience, Can be highly customized, and it is our three major advantages to be able to ship quickly. ' He said.

AI chip era guarantees efficiency

The open source of 'Turnkey's solution will definitely promote the development of the whole intelligent hardware products'. When talking about this, Li Xiaohan is full of confidence, mainly due to the following three aspects:

First, the product manager of intelligent hardware is very scarce, especially the product manager who understands the design of voice interaction. The voice interaction (VUI) is very different from the graphical interface interaction (GUI) of the previous screen. The former is a flat structure, a direct sentence Any graphical interface of the system can do any operation. The latter is a tree structure and needs to be clicked step by step.

These two interactions have their own advantages, and VUI currently has few talents on the market, and because of its flat structure, it needs to be considered from the overall level of the system when designing, rather than simply a single App level. , greatly increased the difficulty of VUI design.

'If the product interaction design is not good, the final product development effect can be imagined.' And as the founding team of Yunzhisheng, Li Xiaohan has more than 10 years of human-computer interaction related research experience, from the voice interaction on Motorola mobile phones. To Yunzhisheng car to Gree air-conditioning, Fibonacci speakers, Yunzhisheng team has accumulated rich experience in VUI design and development, the overall voice interaction program with UniOne as the Turnkey solution as a whole open source, all this will greatly reduce the industry threshold.

Second, the voice interaction scheme open source will greatly shorten the development cycle. As a system-level function, the voice interaction scheme will handle audio drivers, handle interactions with other applications of the system, handle individual cases and wake-up events, and must be robust and stable. With the cloud knows the open source solution is solved.

Partners can only do shallow-level customization, such as awakening word modification; can also do deep-level development, can be completely rewritten in the case of understanding the overall solution.

Third, Yunzhisheng's tried and tested implementation team. In the 'core era', this team will provide technical support for the whole open source solution for partners who are willing to adopt Yunzhisheng UniOne, including code training, tool development, etc., to do everything possible. Reduce the steepness of the learning curve that partners are familiar with in the overall open source approach.

According to Yunzhisheng, the Turnkey program is expected to be officially open source on September 15.

When talking about competition, Li Xiaohan also told the product business review that there are many voice open platforms, but most of them are aimed at cloud service functions. Cloud capabilities are not very helpful for developers. The key path lies in the edge side. The relationship belongs to the upstream and downstream, and will not produce competition.

This set of solutions is not only for the partners that have been missed before, but also wants to absorb companies that have not had similar ideas before, let them know how low the threshold for products to become smart hardware.

The release of the AI ​​chip also changed the positioning of Yunzhisheng: It used to be a technology provider, and now it has become an AI cloud service provider, software solution provider and chip manufacturer.

As for the future positioning of Yunzhisheng, no one can predict. Li Xiaohan told the product business review, one thing is certain, Yunzhisheng must not only focus on the voice field, the image field will definitely involve. 'Artificial intelligence equipment is let The machine is more like a human being, so you have to provide a variety of perceptions, and images are an essential part. '

Li Xiaohan said that in the fast-developing stage of the Internet of Things, there are many possibilities for future development. While greatly increasing the investment in chips, the Yunzhisheng team will also look for new opportunities for innovation, regardless of voice or image. From the perspective of the Internet of Things.

At present, Yunzhisheng began to plan the future a few years ago, and can pay for future results or risks. 'As long as you are determined to move forward, this is the guarantee of efficiency. 'In the form, the opportunity is fiercely evolved In the process, efficiency is especially important for Yunzhisheng.

2016 GoodChinaBrand | ICP: 12011751 | China Exports