After the easy defeat of Li Shishi by AlphaGo last year, people have been amazed that human beings have been surpassed by robots. It is obvious that Ke Jie, who ranked first in the world, has once again lost the robot of artificial intelligence (AI). The fact that artificial intelligence surpasses human beings in many fields, more rationally explores the opportunities for the transformation of artificial intelligence-based technology to human society. According to today's headline report on the first artificial intelligence impact, the Chinese national AI confidence index is as high as 83, and Chinese consumers' attitude towards artificial intelligence has become very positive.
"In fact, artificial intelligence is gradually entering our lives, especially as the research on the intelligent recognition of speech that is very important for artificial intelligence. The world has already made very successful innovative applications, and China is also at the forefront." Xu Zhengdi, product manager of X-Powers Technology, a leading supplier of power and analog semiconductor technology, told the media recently. Due to the optimistic application of voice-based artificial intelligence technology, the company has recently released the first voice recognition multi-array microphone front-end ADC chip AC108 for popular applications of artificial intelligence. "As the introduction of the AC108, a key entry signal for intelligent voice applications, we expect a breakthrough SNR (Signal to Noise Ratio, 108dB) performance and more optimized design to help companies seize market opportunities in artificial intelligence applications." Xu Zhengdi pointed out.
Figure 1: Chinese National AI Confidence Index is as high as 83
Consumers are accustomed to voice interaction, Smart Home is the first breakthroughAccording to the latest "Internet Trends Report 2017", in the smartphone users who use Google Assistant (Google Voice Assistant), only 20% of mobile queries were completed by voice in May 2016, and by May 2017 Nearly 70% of the inquiries are made in natural/conversational language. Great changes have taken place in just one year, and most consumers' mobile query habits have begun to favor intelligent voice interaction.
Figure 2: Consumers are accustomed to using voice interaction to complete queries and other operations
"Intelligent voice will first land in the vertical market. In the application scenario where the interaction is relatively simple, the scene is relatively fixed, and the user has just needed, such as home appliance control and car electronics, there are a lot of development opportunities." Xu Zhengdi said in an interview. The needs of such scenarios are relatively simple, and the most needed is the identification of a series of command words, and the technology is relatively easy to implement. At the same time, before the autopilot matures, the driver in the car can't release his hands, so the voice control inside the car is just needed. For now, the voice control of smart homes has clearly gone a little further. The obvious example is the popularity of Amazon Echo!
Figure 3: Speech recognition as one of the intelligent entrances to the Internet of Things
In the second half of 2016, Amazon lowered the price of Echo Dot from $99 to $49, which means that shipments of Echo-related products will increase significantly in the coming year. At the same time, the upstream supply chain revealed that in 2017, Amazon has increased the order volume of smart speakers to 10 million units, three times that of 2016! In this smart speaker competition, global leaders such as Amazon, Google and Apple are pushing more and more Internet companies to release similar smart speaker products, including Chinese companies joining the competition, such as Jingdong Intelligent's series. This also brings huge opportunities for suppliers in all aspects of upstream and downstream.
Signal pickup front-end processing is critical, and 108dB SNR solves the biggest challenge of far-field speech designIn fact, close-range smart voice applications such as Siri and Google Assistant on smartphones are relatively mature (because they are more biased towards algorithms). "At the moment, the most important intelligent voice applications usually have an interaction distance of 3 to 5 meters, or a longer distance. The more noisy far-field intelligent voice signal processing is the key to Smart Home applications, which is the main target market for AC108.†Xu Zhengdi said, “These applications require multi-array microphones, which must solve multi-channel voice acquisition and high signal-to-noise ratio ( SNR) and low complexity design requirements."
The realization of intelligent voice application requires three steps: first, picking up - converting natural voice signal into digital signal; followed by pre-processing - denoising, canceling, reverberating, etc. the collected voice signal To form a "clean" audio signal; finally, a local or cloud speech recognition engine recognizes and semantically analyzes "clean" speech signals to achieve intelligent feedback. Obviously, long-distance pickup is needed before the speech recognition at the algorithm level is completed. The microphone array must be used first, and the voice pre-processing algorithms, such as NS, AEC, De-reverb, etc., can be used to realize natural language human-computer interaction.
In the entire intelligent voice system, front-end voice collection and processing plays a decisive role in the high accuracy of speech recognition. From the perspective of hardware components, the performance of the front-end voice ADC chip is a key factor! In general, one of the important parameters that determine the performance of an ADC chip is the signal-to-noise ratio SNR, which characterizes the ratio of the maximum undistorted sound signal, the subjectively considered useful signal strength to the noise level, and the signal-to-noise ratio SNR. The higher the noise, the better the performance of the chip.
Analysis of several reports on the mainstream smart speaker products in the market can find that the front-end voice ADC chip has an SNR of 98dB and 101dB, while the AC108 claims 108dB, which is the highest signal-to-noise ratio ADC chip in the industry. “The AC108 model name highlights its performance advantage – 'Audio Codec 108dB(SNR)', 108dB is currently the highest performance for microphone array applications. At the same time AC108 is designed, produced and tested in strict accordance with home appliance standards to ensure good quality requirements. Xu Zhengdi introduced.
Figure 4: Domestic consumer electronics products based on intelligent speech recognition technology
The entire ecological industry chain has matured, and the market is detonating soon.There is no doubt that voice intelligence applications are rapidly evolving. "From our contact with domestic and foreign companies, this market is currently facing a big development opportunity. Currently, major home appliance companies including Gree, Midea, Haier and Changhong have densely arranged intelligent voice control products, from air conditioners and refrigerators. Color TVs, even small appliances have related product layouts, and some products have been launched in the market.†Xu Zhengdi pointed out that “especially with the changes in cost, program maturity and design complexity, market explosive growth is just around the corner.â€
According to Xu Zhengdi's analysis, in the application of large-sized home appliances such as air conditioners, the current cost of voice recognition solutions accounts for about 5% to 10% of the overall cost, and the cost pressure is still relatively large. “There is a lot of compression space for this part of the cost. The cooperation between the company and the domestic voice recognition industry chain is rapidly promoting low-cost, high-performance solutions.†Xu Zhengdi further explained: “The current SoC for most intelligent voice applications. Generally, the standard I2S interface is reserved. The traditional method of supporting multiple ADC inputs is to implement channel conversion by FPGA or MCU. The AC108 considers this factor in the design, and realizes the multi-channel speech signal by adopting the upsampling rate. The standard I2S supports multi-channel audio data transmission, which can save FPGA or MCU in the solution, greatly reducing the cost and complexity of the design. In addition, some solutions need to use multiple ADCs to support different microphone arrays, based on AC108 cascade two A multi-array solution can be implemented with chips (up to four cascading).
Figure 5: AC108 high performance, low cost far field voice picking solution
This customized optimization solution accelerates the time-to-market of vendors while significantly reducing costs, which is especially critical for the fast-growing consumer market. It is understood that the current core protocol AC108 has provided EVM Board and friendly PC tools to potential customers, which is convenient for engineers to quickly evaluate the ADC. “We also provide professional and detailed guidance documents and reference drivers, as well as babysitting services to help customers quickly complete design development and speed up product launch.†Xu Zhengdi introduced. High-performance analog signal processing is often a nightmare for engineers. For intelligent speech recognition applications dominated by the consumer electronics market, such "nanny" technology services are critical to achieving high performance and fast time to market.
Whether ADC or SoC, it is only a part of the voice recognition ecosystem, and the mature development of the entire ecosystem is crucial. At present, the microphone array algorithm, the supporting voice pre-processing algorithm, and the cloud semantic platform have matured. “A large number of R&D work in the industry is still rapidly improving the speech recognition effect, including multi-language support under a large number of AI training, rapid replication and mass production multi-link coordination, etc. The introduction of AC108 once again compensates for the high performance ADC for microphone arrays. Short board, we have cooperated with major domestic platform solution providers and cooperated with international platforms such as Amazon Alexa. The good interaction of these industry chains is boosting the explosive growth of intelligent speech recognition. It is expected that more and more will be seen in one year. Related products are available." Xu Zhengdi pointed out optimistically.
ZGAR AZ MC Disposable
ZGAR electronic cigarette uses high-tech R&D, food grade disposable pod device and high-quality raw material. All package designs are Original IP. Our designer team is from Hong Kong. We have very high requirements for product quality, flavors taste and packaging design. The E-liquid is imported, materials are food grade, and assembly plant is medical-grade dust-free workshops.
Our products include disposable e-cigarettes, rechargeable e-cigarettes, rechargreable disposable vape pen, and various of flavors of cigarette cartridges. From 600puffs to 5000puffs, ZGAR bar Disposable offer high-tech R&D, E-cigarette improves battery capacity, We offer various of flavors and support customization. And printing designs can be customized. We have our own professional team and competitive quotations for any OEM or ODM works.
We supply OEM rechargeable disposable vape pen,OEM disposable electronic cigarette,ODM disposable vape pen,ODM disposable electronic cigarette,OEM/ODM vape pen e-cigarette,OEM/ODM atomizer device.
Disposable E-cigarette, ODM disposable electronic cigarette, vape pen atomizer , Device E-cig, OEM disposable electronic cigarette
ZGAR INTERNATIONAL TRADING CO., LTD. , https://www.zgarvapepen.com