How do AI voice chip startups go through the evolutionary path

In recent years, AI chips have been experiencing the pain of landing. Even for image recognition and speech recognition AI chips with a large demand, it is also necessary to break the barriers of algorithms, data and application scenarios in different applications. With the in-depth application, the players competing for the deer will continue to evolve in continuous iterations to form a closed loop and accelerate the landing. AI chips are like marathon running. They compare endurance, not sprint speed. In this competition, it is not who runs fast, but who has the potential to reach the end and who doesn't take a fork in the way.
With the gradual opening of the voice chip market, forces from all parties are pouring in to take the lead when the market pattern is uncertain. The four mainstream forces are also relatively short and long. The strengths of Internet and communication giants lie in the support of data and scenarios, as well as sufficient funds and talents; Mature chip design companies have obvious advantages in cost control, chip definition, supply chain management and channel; Start up chip companies can move forward with light packaging in front of chip architecture and customization; Algorithm and solution companies have advantages in integrating chip downstream algorithms and application solutions.
Although there are different strategies in the integration mode and function positioning of the advanced AI speech chip, the trend from general to special is clear, that is, whether the CPU directly accelerates the AI algorithm, or the way of DSP + NN module, and the AISC chip designed for speech recognition, in terms of flexibility, model support, cost, power consumption and other comprehensive indicators, In the future, special AI chips will become mainstream products.
In AI chips, scale was indeed very important in the past, but today, it may be more important than scale, but the speed of change. On the track of special AI chips, various "architecture fever" have sprung up. There are indeed differences in memory computing, analog computing, optical computing, brain like computing, etc. For start-ups, if they only adopt "dedicated architecture", there is no high threshold. The final competition is not only the comprehensive test of engineering ability and customer ability, but also the ability of continuous iteration and evolution.
Since the release of sound whirlwind 611 in the middle of the year, the shipment has tended to be one million, which is not easy for the exploration area established for only more than two years. After getting the "ticket" to the next level of competition, the exploration continued to make efforts. At the end of December, it released a product matrix composed of three series of low-power series, flagship series and flagship series and six AI chips, including the first off-line integrated speech recognition solution and a speech recognition solution with AI dual microphone noise reduction function.
In the application of smart home, its voice control technology still faces many challenges, including the low signal-to-noise ratio caused by high noise and far-field recognition environment, the impact of unsteady noise on noise reduction algorithm, and the problem of multiple sound sources caused by playing TV dramas and music at the same time during speech recognition.
The complete speech recognition link can be divided into four links: microphone input, noise reduction processing, speech recognition and recognition result input. First of all, we should work hard on the noise reduction algorithm. On the one hand, the voice AI chip will continue to be optimized. In the future, the voice chip will be upgraded in high dimension and more online and offline integration schemes will be launched; On the other hand, it has extended its tentacles to the next dividend market (image recognition).
AI chip cake is very large. We hope those who have the ability can cooperate with upstream and downstream, open SDK and tool chain, continue to build ecology, land more scene development with a large number of partners, and create value together. "
Shopping malls are like battlefields, and enterprises in different locations have different life and death lines; For different positions, there should be corresponding resource allocation and combat posture. In the charge, we are attacking in an all-round way and striving to activate the stock and increment. Stock is the core voice chip business, and strive to be thorough; On this basis, go beyond the original core business and find new increment, namely image recognition and cloud chip.

Commax-Tech Electronic Co., Ltd      Electronic component specialist

B23, second floor, ASEAN building, Longhua District, Shenzhen

sales@commax-tech.com

https://commax-tech.com

Keyword：AI Voice Chip   Image Recognition   Start-up Chip Company   Chip Downstream Algorithm   Chip Application Solution   Commax-Tech Electronic