Voice recognition technology, also known as automatic speech recognition, is（Automatic Speech Recognition，ASR）， It is a technology that focuses on speech as the research object, and uses speech signal processing and pattern recognition to enable machines to automatically recognize and understand human spoken language or text.

In today's rapidly developing technology, voice recognition technology has gradually penetrated into every aspect of our lives. From intelligent voice assistants to security monitoring systems, from smart home control to financial transaction authentication, voice recognition technology is bringing unprecedented convenience to our lives with its unique advantages.

However, voice recognition is not limited to recognizing human speech as electrical signals. In non biological object recognition, objects can also be identified by analyzing the sound characteristics emitted by objects, such as the knocking sound of different metals, the sound of normal and abnormal operation of power equipment. 

In this field, Anhui Zhongke Haoyin Intelligent Technology Co., Ltd. has become a leader in the industry with its outstanding technical strength and innovative spirit. This issue of Qingqiao International JournalThe "Ningdian Interview" program invites Liu Min, founder and chairman of Zhongke Haoyin, jointlyExplore the mysteries of object sound recognition.

In 2019,Liu MinAnhui Zhongke Haoyin Intelligent Technology Co., Ltd. was jointly established with the Chinese Academy of Sciences, focusing on voiceprint technologyIn the field of AI.At the same time, he also serves asPCCP pipe gallery broken wire voiceprint monitoring and evaluation expertMember of the Hefei Municipal Political Consultative ConferenceSpecial Researcher at the Counselor's Office of Anhui Provincial People's GovernmentVice President of Wuhan University Anhui Alumni Association, Vice President of Anhui Hubei Chamber of CommerceMember of Anhui Province Urban Safety and Emergency Industry Alliance、Member of Anhui Provincial Association of Scientists and Entrepreneurs、Executive Director of Anhui Big Data Enterprise Association、Chinese Civil Aviation Fatigue Testing System Review ExpertEqual positionAwarded patent 24 projects.

voice recognition technologyWhat did it go throughdevelopment history？

Liu Min:In the 1950s, Bell Labs successfully developed an experimental system capable of recognizing 10 English digits, marking the beginning of research in speech recognition technology. With the advancement of computer technology, dynamic programming (DP) and linear predictive analysis (LP) techniques have been applied to the acoustic model construction of speech signals, enabling speech signals to be converted into digital form for computer processing. In 1970, the concept of pattern recognition was introduced into the field of speech recognition, and linear predictive coding (LPC) technology emerged and was widely applied. In 1978, the Dynamic Time Warping (DTW) algorithm solved the matching problem of speech with different durations.

In the 1980s, speech recognition research shifted from small-scale independent word recognition for specific individuals to speaker independent continuous speech recognition. The Jelinek team at IBM developed a speech activated typewriter that, although requiring separate training and pauses, proved the effectiveness of statistical methods. By the mid-1980s, the typewriter's vocabulary recognition reached 20000 words, driving technological development. In 1989, the proposal of Hidden Markov Model (HMM) shifted speech recognition from template matching to probability based modeling, laying the theoretical foundation.

After 2000, human-computer voice interaction became the focus, with improvisational speech recognition, natural speech dialogue understanding, and multilingual simultaneous translation becoming the key. In 2011, Apple launched Siri, which changed the way people interact with each other and brought speech recognition technology into people's lives. In 2012, Google first used deep neural networks in speech recognition, greatly improving accuracy and speed, and promoting the widespread application of speech recognition in fields such as the Internet of Things and smart homes. Afterwards, companies such as Baidu, iFlytek, and Alibaba also proposed their own end-to-end models to promote the development of technology to a higher level.

When it comes to voiceprint recognition, most people's first thought is often human voice recognition. So, how did Zhongke Haoyin enter the relatively niche field of object sound recognition?

Liu Min: Early on In 2010, the company team began to focus on the research and application of technologies such as voiceprint recognition, acoustic imaging, mechanical noise recognition, and fault predictive analysis. In the initial research process, the team mainly focused on human voice recognition technology, but with the deepening of research and continuous exploration of market demand, we found that the field of object voice recognition contains enormous development potential.

Compared to visual technology, there are very few companies specializing in the research and development of voiceprint, especially in the field of object sound recognition. Our competitors are indeed not many. Moreover, with the advancement of industry The rapid development of 4.0 and intelligent manufacturing has put forward higher requirements for the status monitoring and fault warning of industrial equipment. Traditional monitoring methods often rely on manual inspections and complex sensor technologies, which are not only inefficient but also difficult to ensure accuracy. Voiceprint recognition technology can achieve real-time monitoring of device status and early warning of faults by analyzing the sound emitted by the device during operation, and has advantages such as low cost, high efficiency, and strong accuracy

Based on such market demand and technological advantages, Zhongke Haoyin has gradually shifted its research focus from human voice recognition to object voice recognition, and continuously increased research and development investment, committed to creating object voice recognition technology and products with independent intellectual property rights. After years of effort, the company has achieved a series of important results in the field of object sound recognition and has become a leading enterprise in this field.

Accurate through sound What technologies, hardware, and software support are needed to achieve monitoring of device status and diagnosis of faults through auscultation?

Liu Min: In terms of technology, Zhongke Haoyin has multiple core technologies with independent intellectual property rights, among which the most representative is Cmfmc3.0 channel and format conversion engine. This technology can accurately identify various sounds emitted by equipment during operation and determine whether they are normal by extracting and analyzing acoustic features such as frequency spectrum, amplitude, cepstral, and waveform of sound signals. It is reported that the Cmfmc3.0 technology independently developed by Zhongke Haoyin has a voiceprint recognition accuracy of 95% to 99%, far exceeding the 40-60% of domestic and foreign peers.

In addition to core technology, a series of hardware devices are also required to achieve sound collection and transmission. Zhongke Haoyin adopts advanced sensor technology, which can collect the sound signals sent by the device during operation in real time, and transmit these signals to the back-end core engine for analysis and processing through the edge computing gateway. In the selection of hardware devices, Zhongke Haoyin focuses on the stability, reliability, and anti-interference ability of the equipment to ensure the accuracy and correctness of the collected sound signals.

In terms of software, Zhongke Haoyin has independently developed a complete voiceprint recognition software system, which includes multiple modules such as sound signal processing, feature extraction, model training, and fault diagnosis. By collecting and analyzing sound data from a large number of devices, the system can continuously learn and optimize, improving the accuracy of identifying various fault sounds. At the same time, the software system also has a good human-computer interaction interface, which is convenient for users to operate and manage.

If an object cannot produce sound, will its voiceprint still be recognized?

Liu Min: We have proposedThe concept of "everything is interconnected, everything has sound" allows us to use special methods to make objects that do not actively emit sound and recognize their voiceprints.

For example, for transmission towers, we can emit sound in a fixed frequency band and then use ourThe 'artifact' is used to collect sound. Because we know the frequency of the emitted sound, we can better analyze its bolt looseness and check the stability of the iron tower. This method is not only cost-effective, but also effective, and is currently at the forefront of the industry.

In addition, for large infrastructure such as bridges, Zhongke Haoyin has also adopted a similar method for monitoring. By installing special sensors on the bridge, transmitting and receiving sound signals, analyzing changes in sound to determine the structural health of the bridge. This non-contact monitoring method can not only detect potential problems in the bridge in a timely manner, but also will not have any impact on the normal use of the bridge.

In which industries is object sound recognition irreplaceable?

Liu Min: Firstly, the power industry. In the power system, the operating status of equipment such as transformers, switchgear, and transmission lines directly affects the stability and reliability of power supply. Through object sound recognition technology, it is possible to monitor the operating sound of these devices in real time, detect potential equipment malfunctions in a timely manner, and avoid power outages caused by equipment failures. For example, Zhongke Haoyin refers to the Three Gorges of the Yangtze River The 700 MW water turbine unit provides monitoring services by installing sensors through magnetic attraction, which can monitor the operating noise of the unit in high voltage and strong magnetic field environments without the need for power outages or shutdowns, and predict the possibility of faults in advance.

Next is the oil industry. The operation status of various mechanical equipment and pipelines is crucial for production safety in the process of oil extraction and transportation. Object sound recognition technology can monitor the power, power, and pipeline components of petroleum equipment, accurately capturing the sound of equipment faults in complex working conditions, providing early warning and prediction, and reducing personnel and property losses. Zhongke Haoyin has successfully detected faults such as belt conveyor breakage through object sound recognition technology in five application scenarios in Shengli Oilfield, providing strong guarantees for the safety production of enterprises.

In addition, object sound recognition technology has broad application prospects in fields such as transportation, industrial machinery, and new energy. In the field of transportation, it can be used to monitor the operating status of vehicles and detect vehicle faults in advance; In the field of industrial machinery, various mechanical equipment can be monitored for status and fault diagnosis, improving production efficiency and product quality; In the field of new energy, equipment such as wind turbines and solar panels can be monitored to ensure a stable supply of new energy.

Object sound recognitionDoes technology also face some limiting factors in its development process?

Liu Min: On the one hand, the complexity and variability of sound signals pose great challenges to object sound recognition. The sound emitted by different devices in different operating states varies greatly, and is also affected by environmental noise, electromagnetic interference, and other factors, which requires object sound recognition technology to have higher accuracy and robustness.

On the other hand, the application of object sound recognition technology also faces the problems of data shortage and inconsistent standards. Due to the relatively new nature of object sound recognition technology and the insufficient accumulation of relevant data, this limits the training and optimization of the model. At present, there is no unified standard for object sound recognition in the industry, and the data and algorithms between different enterprises and institutions are difficult to be compatible, which also restricts the promotion and application of object sound recognition technology.

In addition, the public's awareness and acceptance of object sound recognition technology are relatively low, which to some extent affects the market expansion of this technology.

What is the future development direction of Zhongke Haoyin in the face of these challenges and opportunities?

Liu Min:Our goal is to become a global leader in voiceprint technologyAI technology providers not only maintain technological leadership, but also continuously expand their application scenarios. We hope to provide more intelligent and efficient solutions for health monitoring and fault warning of industrial equipment through voiceprint technology.

In the future, we will continue to expand the application of object sound recognition technology in fields such as electricity, water conservancy, petroleum, transportation, industrial machinery, and new energy, and strive to createA comprehensive solution for industrial monitoring of object voiceprint. At the same time, we will continue to increase research and development investment, improve our technological level, and enhance the accuracy and robustness of object sound recognition. In terms of data, establish the largest voiceprint fault database in China to provide richer data support for model training; In terms of standard setting, actively participate in the formulation of industry standards, promote the standardization and development of object sound recognition technology; In terms of expanding into overseas markets, we strive to build our company into a leading enterprise in the field of voiceprint fault recognition.

In terms of innovation, Zhongke Haoyin will continuously explore new application scenarios and business models, promote the deep integration of object voice recognition technology with emerging technologies such as artificial intelligence, big data, and the Internet of Things, and provide customers with more intelligent and personalized solutions. For example, combining artificial intelligence technology to achieve automatic diagnosis and intelligent warning of equipment failures; Combining big data technology, deeply analyzing equipment operation data, mining potential value information, and providing support for enterprise decision-making.