Author WEI Shiwei
Editor SHI Yaqiong
It is reported that CCVUI, the audio NPU chip and algorithm company, recently announced receiving round A financing of tens of millions of RMB， invested solely by Forebright with Geyang Capital acting as the sole financial advisor for this round. This round of funds will be mainly used to expand R&D personnel and technical support teams, as well as R&D investment in chip iterations.
CCVUI, founded in September 2018, focuses on developing AI chips and algorithms for the field of intelligent speech recognition mainly based on the self-developed reconfigurable neural network engine, with features of high performance, ultra-low-power consumption, high recognition rate, low cost, strong algorithm scalability and etc., suitable for smart phones, earphones, white household appliances and other categories.
The application of acoustics is very broad. All sound-related applications in listening to music, making phone calls, and voice interaction on devices of daily use, such as smartphones, Bluetooth headsets, smart speakers and etc, need to process the sound signal in an interlocking manner supported by a long component supply chain. For example, ASR, voice assistant, and application of noise reduction for calls mainly involve the pre-processing of sound, which is related to the microphone, while devices’ sound quality, active noise reduction, and sound effect mainly involve the post-processing of sound, which is related to the speaker.
However, Mr Tang, CEO of CCVUI, told 36Kr that he found many players in the industry would focus more on the technological innovation in pre-processing when designing voice chips, leaving post-processing designed in traditional ways. Therefore, the company's original intention is to design a voice chip that takes into account both pre-processing and post-processing of sound. Then why did the company choose to start with NPU (embedded Neural-network Processing Unit)? On one hand, the long cycle and high cost of the R&D of large-scale SoCs can be high thresholds for start-ups. On the other hand, since the products are mainly made for low-power, lithium-ion power supply scenarios, and the power consumption of the main control chip to run the algorithm is too high, it is more suitable to develop an NPU chip as a co-processor to match with the main control chip.
At present, x5200x, CCVUI’s low-power voice chip, has been mass-produced and delivered. This is a dedicated voice recognition front-end chip, based on the RISC-V architecture, adopting miniaturised neural network design, breaking through difficulties such as ultra-low-power ADC, low-power PLL system design, low-power digital system design, and miniaturised packaging, and etc.
Specifically, the XS200X chip has four major features. First, it adopts ultra-low-power design, supporting Always on wake-up mode with wake-up power consumption less than 1mW; second, it has far-field speech recognition and high-intensity computing capabilities, and supports 14 algorithms, such as AGC, AEC, beamforming, de-reverberation, complex noise reduction, and multi-command speech recognition, and etc., among which the KWS wake-up rate is above 98% in noise-free cases; third, based on the DNN+CNN neural network noise reduction algorithm, it can better achieve ENC and ANC; Fourth, equipped with 4 microphone interfaces, it supports multi-channel analog MIC input or multi-channel digital MIC input, as well as multi-channel TDM (I2S) input and output.
CCVUI's ultra-low-power voice wake-up and recognition chip
With these features, CCVUI's noise reduction solution for speech can well solve the issue of noise reduction in strong wind noise scenarios, which is also the main issue faced by many audio products today. Tang gave an example that, when the speed of the motorcycle reaches 120 km/h, the single-microphone noise reduction solution of CCVUI’s chip can greatly filter wind noise and road noise to achieve clear calls, while the Bluetooth chip solutions of well-known manufacturers in the industry can only achieve clear calls with the maximum speed of 30 km/h. It is the neural network noise reduction technology applied by CCVUI and trained under over 6000 kinds of noises in scenes like strong wind, road, metro, square dancing and etc. that greatly solves various noise problems.
From the market perspective, currently the benchmark of CCVUI is mainly the US DSP Group, who is the world's leading provider of communication voice and wireless chipset solutions, having rich technical experience in audio digital signal processor, AI core, audio IoT wireless chip and etc. Meanwhile, the company has been acquired by Synaptics for $528 million announced in August 2021.
Tang mentioned that the core team of CCVUI has started to develop chip solutions based on the RISC-V architecture four years ago, and has accumulated a lot of experience in cost control and technology development, which is precisely the differentiated advantage of the company compared with others players. On the one hand, players in the market mostly use the Arm Cortex-M4F architecture or the Cadence architecture, required to pay for IP patent fees on product development, while CCVUI uses the open source RISC-V architecture, with all analog IPs independently researched and developed, which can greatly reduce the costs for product development. On the other hand, the homogenisation of products on the market is very serious, and the company developing chips based on the RISC-V architecture can achieve product differentiation while ensuring the consistency of product power consumption, computing power, and performance.
In terms of business, CCVUI has four business segments: travel, intercom, voiceprint unlocking, and sound source positioning, with each segment combined together to achieve implementation. For example, travel and intercom are mainly targeting at the smart helmet market. The company has cooperated with Eleme Takeout to develop smart helmets for riders, so that riders can directly process orders and make phone calls through voice interaction; and sound source positioning is mainly targeting at areas such as live streaming and video teleconference.
At this stage, the company's XS200X chip has completed the joint optimization and reference design of multiple chip and software platform vendors, and has launched on markets such as smartphones, smart headphones, smart helmets, car audio, low-latency live streaming, and teleconferencing. Its partnership has reached domestic and overseas leading companies such as Vodafone, 3M, Harman, Wuling Technology, and Eleme Takeout.
The smart helmet developed with Eleme Takeout
Since its establishment, the cumulative R&D investment of CCVUI has exceeded 30 million RMB. Currently there are two profit models: chip sales and algorithm sales, among which the chip shipment has achieved hundreds of thousands of pieces in 2021, mainly used in the fields of smart helmets, intercom/conference speakers, and headphones, while algorithm sales involve single-mic neural network noise reduction algorithm and dual-mic noise reduction algorithm, which, at this stage, are only suitable for the Bluetooth chip platform of Bestechnic and Airoha Tecnology Crop.
In the next year, CCVUI will focus on promoting the offline voiceprint recognition function, covering applications such as identity authentication and auxiliary unlocking. At the same time, the company expects that the chip shipments for the fields of smart helmets and walkie-talkies will achieve millions of shipments respectively in this year.
The core team of CCVUI has possessed key technical capabilities and mass production experience in multiple fields from chips to algorithms, as well as rich technical knowledge and sales experience of staff whose average working experience is over 15 years. Mr Jiang, the chairman of the company,is a doctor of engineering from Tokyo Institute of Technology, and has over 20 years of experience in chip development; he was the GM of Fujitsu Semiconductor Limited, the CTO and director of Goke Microelectronics Co., Ltd., and the person in charge of relevant major national science and technology projects; he has applied for more than 100 patents so far, and developed and achieved the mass production of more than 20 chips. Mr Tang, the company's founder and CEO, used to be the CMO of Goke Microelectronics Co., Ltd., and served as the sales director in Innofidei, China Unichip Technologies Inc., and ROFS Microsystem (Tianjin) Co., Ltd.; he has more than ten years of experience in chip sales and team and channel management.
This article is only for information sharing. If there is any infringement, please contact the editor for deletion.