AI special voice chip will start to explode this year? What are the advantages of AI special voice interaction chip compared with general-purpose chip? Let's have a look.
Recently, several voice technology start-ups in China have successively launched AI voice dedicated chips. On May 16, yunzhisheng released the first AI series chip Unione and the first generation chip "swift" for the Internet of things in Beijing. On May 24, we went out to ask about the release of mobvoi A1, the first voice chip module of our company, in Beijing. Yesterday, rokid released its special SOC chip kamino18 for AI voice in Hangzhou. At the same time, CEO Gao Shixing confirmed that the company is building an AI voice chip, which is expected to be streamed in the second half of this year.
Cloud Zhisheng, go out to ask, rokid, and simichi, the top start-ups in the field of AI voice, almost all started betting on AI voice chips at the same time.
So, why did AI speech chips start to explode this year?
After a small climax in 2017 (global sales of smart speakers exceeded 30 million units), the domestic smart speaker market is in the period of sales blowout this year, with a large number of 100 yuan smart speakers. Canalys, a market analysis company, predicts that in 2018, the global shipment of smart speakers will exceed 56 million, making the voice interaction industry the focus of the market.
Amazon echo series products have exploded the smart speaker Market
Behind such a large market, the chip scheme in the smart speaker began to change from a general-purpose chip to a dedicated voice chip. For example, in 2017, the tmall Genie launched by Ali used the mt8516 voice chip of MediaTek, while the Xiaomi Xiaoai speaker used the crystal morning A113 chip.
In addition to the smart speaker, more hardware devices in the home and office scenes also began to be voice and intelligent, which led to the outbreak of AI special voice chip.
In this situation, a number of domestic voice technology companies, with their own accumulation in voice recognition, natural language processing, voice interaction design and other technologies, began to transform into AI voice chip integration and provide voice interaction solutions. And the four voice start-ups mentioned above -- cloud Zhisheng, go out to ask, rokid and Spitzer almost all started the layout of voice chips under this situation.
So, what are the advantages of AI special voice interaction chip compared with general chip?
Whether it's swift, the first generation of Unione chip released by yunzhisheng for the IOT field, mobvoi A1, the voice chip module core released by go out, or kamino18, the SOC chip released by rokid, it's featured by high integration, low power consumption, low cost and customization.
Swift chip architecture
Yunzhisheng's "swift" chip adopts CPU + udsp + deepnet architecture, and the company claims that these architectures are independently developed. And go out to ask and rokid announced that their chips are based on Hangzhou Guoxin technology chip in-depth customization. In an interview with Netease intelligence, Rosa, CEO of rokid, confirmed that kamino18 is manufactured by Guoxin gx8010 based on 40 nm process.
Go out and ask mobvoi A1 chip
Rokid kamino18 chip
In addition, Guoxin technology and sipic are also partners. If there is no accident, the sipic AI voice chip, which will be streamed in the second half of the year, will also be built based on Guoxin gx8010.
At the end of last year, Guoxin technology released two NPU chips, gx8008 and gx8010, which are mainly used for AI voice interaction. The latest tensilia hifi 4 DSP core of cadence is built in, which is mainly used for low power consumption, low cost, offline and integration.
In other words, the AI chip developed by Guoxin technology has provided standard interfaces such as digital signal processor DSP, neural network processor NPU, and USB / IIS / IIC / UART. Ask outside, rokid and other manufacturers do not need to do IP design, only need to carry out architecture integration, most of which are microphone array signal processing, noise reduction, wake-up technology, voiceprint recognition and some voice skills. Although yunzhisheng is a self-developed udsp and deepnet architecture, it is basically equivalent to the above two chips in function.
Recently, CEO Gao Shixing also revealed that the AI voice chip to be released in the second half of the year will be an ASIC chip, with acoustic signal processing and voice capabilities, ultra-low power consumption, and strong acoustic signal processing and expansion capabilities.
In addition to chips, these companies focus on providing overall voice interaction solutions. Among them, cloud Zhisheng has proposed a solution of cloud core integration, docking with AI cloud services, AI software solution providers and chip manufacturers, and also providing certain open source capabilities, providing corresponding customized tools; going out to ask for a one-stop voice solution combining hardware and software; rokid has also said to provide a series of voice solutions.
Where is the market for AI speech chips?
When talking about the market of AI chips, they are generally customized according to specific products in specific scenarios. For these companies that make special voice chips for AI, smart speakers, children's story machines and home appliances become their main products. Finding a manufacturer that can mass customize AI voice chips is the most critical step in commercialization.
At present, yunzhisheng is on the road of to B, and its partners are JD alpha platform and Yikatong technology. The goal of cooperation with the former is to build customized intelligent benchmark products, while the latter is to jointly develop pre car loading specification level AI chips.
At the rokid conference, CEO misca said that rokid is not a to B company, but a to C (community) company. In the future, through R & D and understanding of products, technologies and markets, platforms and solutions will be launched to build ecological and industrial empowerment.
Rokid me portable speaker
Go out to ask and define itself as a company that combines software and hardware. It is launching a variety of consumer grade intelligent hardware products of different categories. In addition to its layout in the field of smart speakers, it is also involved in smart watches, smart headphones and other fields.
Small question speaker fashion tickasa nano
But can voice interaction products represented by smart speakers really support the AI chip dream of these companies?
Whether AI dedicated voice chips can continue to explode depends on whether these chips can be applied to products on a large scale, and on the other hand, whether the voice interaction ability of these products can be favored by users and tested by the market.
Wei Shaojun, director of the Institute of microelectronics at Tsinghua University, said in an interview that the current AI chip market is over hyped. This is because the killer application of AI hasn't appeared yet, no matter it's smart speakers or other products, it hasn't become a just needed product.
Therefore, only by making speech a real interface of human-computer interaction can we promote the outbreak of AI speech chip.
While Qualcomm, NVIDIA, Intel and other chip giants have not yet entered the voice chip market, this is a good time for startups to run in the voice chip field blindfolded.
In a word, cloud Zhisheng, go out to ask, rokid and sipic may take advantage of this wave of AI voice chip boom to become the leader in the field of AI voice, but they may also become a show.
Everything is waiting for the market and time to test.