Since the 1980s, speech recognition technology has made many breakthroughs, especially the speech recognition technology based on Hidden Markov Model (HMM), which has matured and become the mainstream of speech recognition. However, the basic HMM model also has some inherent defects. These defects are reflected in the model parameters, except that they are not reflected directly in the model parameters.
(1) Using the independent assumption of state output, the output of each moment is only related to the state it is in, but has nothing to do with the previous output. However, the actual speech signal has a strong time correlation, which affects the description of the HMM model. The ability of inter-frame correlation of speech signals.
(2) The continuous HMM model assumes that the state output probability density function is a mixed Gaussian distribution function, and the actual speech signal distribution is very complicated and difficult to characterize with a simple Gaussian distribution combination. To compensate for these deficiencies, many improved methods have been proposed.
Speech recognition technology is a technology that has developed rapidly in recent years. Due to its important theoretical value and broad application prospects, it has received extensive attention. Speech is a complex nonlinear process, and the limitations of speech recognition methods based on linear system theory are becoming more and more prominent. In recent years, with the gradual deepening of the research and application of nonlinear neural networks, fuzzy logic, particle swarm optimization algorithms, etc., these theories have begun to be applied independently or cross-over to the field of speech recognition.
Language is one of the main sources of human access to information. It is not only the most convenient, effective and natural means for human beings to exchange information with the outside world, but also an important tool for communication between people and machines. Whether it is human-to-human or human-to-human language communication, speech signal processing, especially digital processing of speech signals, has a particularly important role.
With the rapid development of computer technology, the use of modern methods to study speech signal processing technology, enabling people to more efficiently generate, transmit, store and obtain voice information, which is of great significance for the promotion of social development.
Digital speech signal processing, including three aspects, namely the digital representation of speech signals, various methods and techniques of digital signal processing theory of speech signals and the practical application of digital speech processing theory and technology in various fields.
Application of Fuzzy Neural Network in Speech RecognitionNeural network is an emerging science based on the results of modern scientific research to simulate the structure of human brain. It is not a true comprehensive description of the human brain, but the abstraction, simulation and simplification of such biological neural networks. Exploring the information processing, storage and search mechanisms of the human brain, opening up new avenues for the research of artificial intelligence and information processing. An artificial neural network is a system that uses a physically achievable system to simulate the structure and function of human brain nerve cells. It is interconnected by many processing units for parallel operation. Its processing unit is very simple, but its work is "collective". Its information transmission and storage are similar to neural networks. It has no arithmetic unit. The basic unit of modern computers such as memory, controllers, etc., but a combination of identical simple processors whose information processing is stored on the connection of the processing unit.
Fuzzy logic is a way of imitating the uncertainty of the human brain, the way of thinking, the description system of the model that is unknown or uncertain, and the control object of nonlinear and large lag. Apply fuzzy sets and fuzzy rules to reason and express the transitional boundary. Or qualitative knowledge experience, simulating the human brain mode, implementing fuzzy comprehensive judgment, reasoning and solving the problem of regular fuzzy information that is difficult to deal with by conventional methods. Fuzzy logic is good at expressing qualitative knowledge and experience with unclear boundaries. It uses the concept of membership function to distinguish fuzzy sets, handle fuzzy relations, simulate human brains to implement rule-based reasoning, and solve various kinds of problems caused by the logical break of the "discharge law". Identify the problem.
With the deepening of research on fuzzy information processing technology and neural network technology, the fuzzy technology and neural network technology are organically combined to construct a neural network or adaptive fuzzy system that can "automatically" process fuzzy information to become fuzzy. An inevitable trend of in-depth research and development of technology and neural network technology. The neural network technology and the fuzzy technology each have their own advantages. The former uses the biological neural network as the simulation basis, and tries to move forward in the aspects of simulation reasoning and automatic learning, making artificial intelligence closer to the self-organization and parallel processing functions of the human brain. New prospects and new ideas have been shown in many aspects such as pattern recognition, cluster analysis and experts. Based on fuzzy logic, the latter grasps the fuzziness of human thinking and imitates the fuzzy comprehensive judgment reasoning of human beings to deal with the problem of fuzzy information processing that is difficult to solve by conventional methods. Combining fuzzy technology with neural network technology can effectively leverage their respective advantages and make up for the shortcomings. The specialty of fuzzy technology is to expand the scope and ability of neural network to process information, so that it can not only process accurate information but also process fuzzy information and other inaccurate information, not only to achieve accurate association and mapping, but also to achieve inaccurate Lenovo and mapping, especially fuzzy association and fuzzy mapping.
Speech recognition usually involves multiple factors in the implementation process and needs to be considered at the same time. Due to the large amount of computation, coupled with the randomness of speech signals, and our shallow understanding of human hearing mechanisms, the current ability of machines to automatically recognize speech is much worse than that of humans, especially for non-specific people. This is especially true for recognition. Using the fuzzy neural network model as a classifier or clusterer, some new speech recognition methods have been developed.
Because fuzzy neural network not only has the ability of knowledge extraction and expression in fuzzy systems, it is suitable for expressing fuzzy or qualitative knowledge, can use similar human thinking mode for reasoning, and also has neural network with parallel computing, distributed information storage, fault tolerance. Strong ability and a range of capabilities with adaptive learning capabilities. The fuzzy neural network model is used in the speech recognition system, and the system has the following characteristics:
1. The useful information in the sample set can be utilized as much as possible to achieve multi-factor comprehensive evaluation, and the advantages of the neural network can be exerted.
2. It can introduce the experience knowledge of domain experts well, use fuzzy rules to guide the network training, and make the network training more in line with human reasoning habits.
3. After special fuzzification on the input and output forms, the information contained in the finite sample set can be better, and the original distribution is reflected in the approximate true distribution.
Traditional speech recognition and speech recognition using fuzzy neural networks are different. In the traditional speech recognition method, the pattern matching method is to perform recognition by extracting feature patterns and pattern matching after pre-processing the speech. Due to the high degree of variability of the speech signal, it is almost impossible to match the input pattern to the standard mode. Therefore, the rules for calculating the similarity or distance of the input speech feature pattern and each feature pattern are determined in advance, and the smallest distance is the most similar mode. When the syntax pattern recognition method considers that the input position pattern belongs to an object, it is necessary to check the input mode and the structure of the recognition object. When the structure of the object pattern is the same or the structure is consistent within a certain range, then the unknown mode is determined. Identify the voice of the object. The difference between the speech recognition method of fuzzy neural network and the traditional method is that after the feature parameters of the speech are extracted, unlike the traditional method, the input mode is compared with the standard mode, but the fuzzy neural network is based on expert knowledge or prior knowledge. First, the input feature data is fuzzified to generate membership degrees for different rules. Then, according to the standard, a large number of connection weights in the network are used to perform nonlinear operations on the input mode, and the input points that generate the most excitement represent the classification corresponding to the input mode.
Fuzzy control was born in the United States in the 1960s, and was born in Europe in the 1970s. When Westerners did not like the "fuzzy theory" in the 1980s, it was developed in Japan and widely used in the automatic control of household appliances. Since the age and neural network, it has been widely recognized around the world and has become an important branch of intelligent systems. Although the study of fuzzy neural networks is not as long as neural networks, it combines the advantages of fuzzy control and neural networks and is now widely used in various fields. At present, the application of fuzzy neural networks in speech signal processing is very active, and the application in speech recognition has made great progress. Similar to neural networks, fuzzy neurons are mainly inspired by auditory neural models to form artificial systems with similar capabilities, so that they can get better performance when solving speech signal processing (especially identification) problems. It is an important direction of current speech recognition research to study fuzzy neural network to explore human auditory nerve mechanism and improve the performance of existing speech speech recognition system.
CL-2H Copper Connecting Terminals
Our company specializes in the production and sales of all kinds of terminals, copper terminals, nose wire ears, cold pressed terminals, copper joints, but also according to customer requirements for customization and production, our raw materials are produced and sold by ourselves, we have their own raw materials processing plant, high purity T2 copper, quality and quantity, come to me to order it!
CL-2H Copper Connecting Terminals
Taixing Longyi Terminals Co.,Ltd. , https://www.longyicopperlugs.com