Promote cooperation, promote promotion and build a platform, the National Intelligent Voice Innovation Center——

Analyze the mystery of sound to help the industry upgrade

2024-05-21 10:01 Source: People's Daily Font size: large in Small Print

General Secretary Xi Jinping stressed that "we should strengthen major scientific and technological breakthroughs, strengthen the deep integration of scientific and technological innovation and industrial innovation, actively cultivate new business forms, new models and new drivers, and develop new quality productivity in line with local conditions."

Let scientific and technological achievements move from laboratory to production line, and the National Manufacturing Innovation Center plays the role of innovation highland. In the 29 national manufacturing innovation centers that have been built, the innovation chain, the industry chain, the capital chain, the talent chain, and the upstream and downstream cooperate closely. With the integration of industry, education, research, and innovation, original and disruptive scientific and technological innovation achievements have emerged, accelerating the transformation to real productivity, and providing strong support for promoting new industrialization, building a modern industrial system, and developing new quality productivity.

Our reporter walked into the national manufacturing innovation center to ask for results, talk about experience, explore prospects, and see how new industries, new models, and new drivers can grow.

——Editor

To the east, it is the High tech Campus of University of Science and Technology of China; To the north, it is the Hefei Innovation Institute jointly established by the Hefei Municipal Government and the Hefei Institute of Physical Sciences of the Chinese Academy of Sciences - China Sound Valley in Hefei, Anhui Province. It is the first national industrial base in the field of artificial intelligence in China. More than 2000 enterprises, including iFLYTEK and Huami Technology, have settled here, with an annual output value of more than 200 billion yuan.

At the north gate of Sound Valley, the brand of "National Intelligent Voice Innovation Center" is particularly eye-catching. Relying on the local intelligent voice and artificial intelligence industry cluster, the innovation center focused on the field of intelligent voice to carry out research on key common technologies, and produced a number of scientific and technological innovation achievements. Here, how to strengthen scientific and technological innovation, especially original and disruptive scientific and technological innovation? How to apply innovative achievements to industry in a timely manner? Journalists visit the innovation center on the spot.

Intelligent unmanned laboratory——

Intelligent voice interactive detection for 24 hours

"Hello, air conditioner, the room is too hot." "OK, I have turned on the cooling mode for you." Human computer dialogue is increasingly appearing in families. Compared with household appliances such as refrigerators and washing machines, people have a stronger demand for intelligent voice interaction of air conditioners. However, it is not easy to ensure that the "ear" of the air conditioner is sensitive enough.

"In the past, only in a closed room, testers played sounds and observed and recorded the response of the air conditioner." Gao Ru, director of the testing center of Shandong Qingdao Haier Air Conditioner Co., Ltd., said that manual testing was not only inefficient, but also difficult to simulate complex use scenarios. Last March, Gao Ru accidentally heard that the National Intelligent Voice Innovation Center was building an intelligent unmanned laboratory for intelligent voice interaction, and immediately went to the field to learn about the situation.

Entering the intelligent unmanned laboratory is like being in a professional recording studio - surrounded by sound absorption diffusion plates and acrylic hemispheres for adjusting reverberation, in which various speakers are distributed. "Through reverberation adjustment, it can simulate the sound field environment of 10 square meters to 300 square meters, and 19 speakers can imitate the background noise of various scenes. A laboratory of 50 square meters can restore more than 95% of the voice interaction use scenes." Li Menghui, the development engineer of the public detection service platform of the National Intelligent Speech Innovation Center, introduced, The laboratory can realize 24-hour intelligent voice interaction detection. Relying on the millions of corpora in the center, all kinds of voices cover nearly 200 kinds of voices, languages, and accents of all ages and people.

Taking the air conditioning detection as an example, the staff only need to set the relevant parameters, and the intelligent robot can reach the designated place and play the sound through the bionic artificial mouth. The microphone beside the test bench will automatically identify the feedback results of the air conditioner. The camera above the laboratory will take pictures of the air conditioning display panel. At the end of the detection task, the detection report will be automatically generated to give feedback on the response success rate, response time, failure reason, etc.

Through cooperation with the Center, Haier "copied" the experimental environment in Qingdao. "It was put into use this year. According to the calculation of 20 seconds per test, more than 4000 tests can be completed in one day." Gao Ru said that with the help of an intelligent unmanned laboratory, the air conditioners produced by Haier today can not only carry out voice interaction in Mandarin, but also "understand" many dialects. Some export products have mastered multilingual ability.

It is reported that the center is established in the form of "company+alliance", which gathers leading enterprises and scientific research institutions in the field of intelligent voice in China. The center acts as an engine to drive the cooperative operation of shareholders and the alliance. "This model is conducive to promoting cooperation and exchange between the Center and enterprises, and between enterprises, and promoting the application and development of scientific and technological innovation achievements in the manufacturing industry." Wu Jiangzhao, general manager of the National Intelligent Voice Innovation Center, said.

Industrial AI scheme——

The single station patrol inspection time of the substation is reduced to less than 30 minutes

When a substation equipped with more than 20 sets of 10kV switchgear has abnormal noise, how to quickly identify the fault area? "It is difficult to identify the source of abnormal sound directly with your ears, and in the past, you could only check one by one," said Wang Longzhen, a specialist in operation and maintenance of State Grid Maanshan Power Supply Company. "Now, using voice print recognition devices, you can quickly lock the fault location."

The so-called voiceprint recognition device is jointly developed by the Center, iFLYTEK and State Grid Anhui Electric Power Research Institute. "The center has been exploring how voiceprint technology can be applied to the industrialization scene. After communication, it found that Anhui Electric Power Research Institute also has this demand, so it 'hit it off'." Li Jun, vice president of iFLYTEK Industrial Intelligence Research Institute, said.

The voiceprint recognition device can also determine the cause of the fault. "Our professional and technical personnel join in the research and development, analyze what kind of fault various sound samples represent, and then train the algorithm model of the device," said Zhang Chenchen, an electric power operation inspection engineer of Anhui Electric Power Research Institute.

Today, voiceprint recognition devices have been applied to more than 40 substations in Zhejiang, Anhui, Guangdong, Ningxia and other places. They can accurately detect partial discharge, short circuit impact, loose clamps, cooler abnormal noise and other problems, shorten the patrol inspection time of a single station to less than 30 minutes, and also reduce the frequency and safety risks of on-site operations.

Voiceprint recognition is only one of a variety of industrial AI (artificial intelligence) solutions provided by the Center. Various plans have been put into practice at an accelerated pace, which is a new driving force for the traditional industrial agglomeration.

Huang Wei, the person in charge of the AI project of the center industry, has been busy installing intelligent quality inspection equipment for the air-conditioning assembly line of Hefei Haier Industrial Park with his colleagues. "Previously, an intelligent quality inspection equipment was installed in one production line, and the effect is good. Now it needs to be laid on more production lines," said Huang Wei.

In the past, a worker had to test more than 1000 air conditioners a day. "When checking the brand logo, it is likely to cause visual fatigue because of repeated watching." Dai Yongsheng, general manager of Hefei Haier Air Conditioner Co., Ltd., said, "If it is a product with voice interaction function, workers need to give voice instructions." To improve the detection efficiency, Dai Yongsheng found the National Intelligent Voice Innovation Center and jointly developed intelligent quality inspection equipment.

The reporter saw at the scene that the intelligent quality inspection equipment was like adding a semi enclosed rectangular iron box to the production line. When the air conditioner enters from the production line, the internal speaker of the equipment sends voice commands, and the camera and recording equipment will judge whether the response given by the product is correct. The code scanners and cameras distributed in other locations will also confirm the trademark, energy efficiency level, model nameplate and other information. When the product "walks" out of the quality inspection equipment, the test results will be displayed in the background.

"Intelligent quality inspection equipment can complete more than 20 quality inspection tasks in 7 categories, such as product functions, voice interaction, logo appearance, with an accuracy rate of 98.5%." Dai Yongsheng introduced that the quality inspection equipment of a production line can complete the inspection of more than 4000 products every day, and it is planned to be fully applied in the company's home appliance production line in the future.

AI model full custody cloud service platform——

Support more than 1000 algorithm models online

As the leading enterprise in the voice field, iFLYTEK has a large number of algorithm models for speech recognition and speech synthesis. "Different languages involve different algorithms. In the past, each set of algorithms was implemented separately, which took a long time and required a lot of repeated construction, operation and maintenance work." Zheng Wei, project director of the AI model full custody cloud service platform of the National Intelligent Voice Innovation Center, said.

Wu Jiangzhao also agreed: "If an innovative enterprise or scientific research institute wants to implement an algorithm, it needs not only algorithm engineers, but also engineering framework designers, testers, operation and maintenance personnel, as well as the support of computing resources."

At the beginning of 2020, the Center and the Voice Cloud Platform R&D Department of iFLYTEK jointly developed the AI model full custody cloud service platform. By importing the designed algorithm into it, we can achieve the landing of scientific research achievements, and the whole process generally does not exceed two days.

Today, the types of algorithm models hosted by the platform are not limited to the field of intelligent voice. Application oriented enterprises can choose the algorithm model they need. "Just like shopping in the supermarket, everyone can find the corresponding capability engine on the platform for the needs of natural language understanding, image recognition, voice recognition, etc." Wu Jiangzhao said.

The platform provides hosting services for Shangtang Technology, Maverick Translation, University of Science and Technology of China and other manufacturers and universities, and supports the operation of more than 1000 algorithm models. The total number of applications accessed by the platform exceeded 2 million, covering nearly 4 billion end users in total, and the average daily total service volume exceeded 2 billion.

Focusing on the implementation of innovative achievements, a series of policies and measures have been implemented successively: Hefei Municipal Bureau of Economy and Information Technology held an industrial integration docking conference, invited more than 20 key manufacturing enterprises to participate in the conference, and 6 enterprises initially reached cooperation intentions; Anhui Province printed and issued Several Policies for Building Innovation and Application Highlands of General Artificial Intelligence Industry, proposing to accelerate the application of all time and all area scenarios and build a good industrial ecology.

"With the continuous support of various measures, the innovation momentum of the Center will continue to increase," Wu Jiang said. Reporter Luo Yangqi

Scan to open the current page