Decoding network system, speech recognition method and device, equipment and medium

Abstract

The application provides a decoding network system, a speech recognition method, a device, equipment and a medium. The decoding network system comprises a first direction network unit and a second direction network unit. The first direction network unit comprises N first keyword transmission channels. The phoneme sequence on the first keyword transmission channel comprises a first phoneme sequence of a keyword. The second direction network unit comprises M second keyword transmission channels. The phoneme sequence on the second keyword transmission channel comprises a second phoneme sequence of the keyword. The second zero-in-degree node is connected to at least one first zero-out-degree node. The decoding network system provided by the application can be used only for recognizing and matching a keyword in speech, so as to reduce information capacity and occupied memory while ensuring keyword recognition efficiency. Furthermore, since the decoding process does not need to be backtracked, the efficiency of keyword recognition in speech recognition is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, and in particular to a decoding network system, a speech recognition method and device, equipment and a medium. BACKGROUND

[0002] Speech recognition technology, also known as automatic speech recognition (ASR), aims to convert the lexical content in human speech into computer-readable input, including key presses, binary codes or character sequences, to achieve human-computer interaction. Speech recognition technology has a wide range of applications in modern life and can be applied to scenarios such as car navigation, smart home, voice dialing, simultaneous interpretation, etc. As the core of the speech recognition system, the speech decoding process based on the decoder plays an important role in the entire speech recognition process and directly affects the accuracy of the recognition result.

[0003] A weighted finite-state transducer (WFST) is usually used as a decoder for speech recognition, but this method is bulky and inefficient for keyword decoding. Currently, the large decoding network size is reduced to facilitate keyword decoding, but this brings a series of negative effects, such as low keyword recognition efficiency and low accuracy.

[0004] Therefore, the present application proposes a decoding network system, a speech recognition method, a device, equipment and a medium to improve the efficiency and accuracy of keyword recognition in speech recognition. SUMMARY

[0005] The present application proposes a decoding network system, a speech recognition method, a device, equipment and a medium to improve the efficiency and accuracy of keyword recognition in speech recognition.

[0006] In a first aspect, the present application provides a decoding network system, comprising: a first direction network unit and a second direction network unit; the first direction network unit comprises N first keyword transmission channels, the phoneme sequence on the first keyword transmission channel comprises the first phoneme sequence of the keyword, and the N first keyword transmission channels have the same first zero-in-degree node and different first zero-out-degree nodes, and N is a positive integer; the second direction network unit comprises M second keyword transmission channels, the phoneme sequence on the second keyword transmission channel comprises the second phoneme sequence of the keyword, and the M second keyword transmission channels have different second zero-in-degree nodes and the same second zero-out-degree node, and M is a positive integer; the second zero-in-degree node is connected to at least one first zero-out-degree node.

[0007] The beneficial effects are that: the first phoneme sequence on the first keyword transmission channel includes the first phoneme sequence of the keyword, and the second phoneme sequence on the second keyword transmission channel includes the second phoneme sequence of the keyword, so that the decoding network system can only be used to identify and match the keyword in the voice, so as to reduce the information capacity and occupied memory of the decoding network system while ensuring the keyword recognition efficiency. Meanwhile, different keywords can be identified through different keyword transmission channels, so that the accuracy of identifying the keywords in the voice can be improved, and because different keyword transmission channels correspond to different keyword phoneme sequences, the physical meaning of the nodes on each transmission channel is clear, so that the decoding network can be optimized; and because the decoding process does not need to be backtracked, the efficiency of keyword recognition in voice recognition is improved.

[0008] Optionally, the decoding network system further comprises L bridges, wherein the bridges are provided with keyword identifiers, the second zero-in-degree node is connected to the first zero-in-degree node through the bridges, and L is a positive integer. The beneficial effects are that: through the identifiers of the bridges, the second zero-out-degree node and the first zero-in-degree node can be more accurately matched and connected.

[0009] Further optionally, the line structure of the first directional network unit is a directed acyclic graph, the line structure of the second directional network unit is a directed acyclic graph, and the directions of the first directional network unit and the second directional network unit are opposite. The beneficial effects are that: because the line structures of the first directional network unit and the second directional network unit are directed acyclic graphs, backtracking decoding does not occur in the process of voice recognition of the first directional network unit and the second directional network unit, so that the efficiency of keyword recognition in voice recognition is improved.

[0010] Still further optionally, the line structure of the first keyword transmission channel is a straight line path or a tree-shaped path, and the second keyword transmission channel is a straight line path or a tree-shaped path. The beneficial effects are that: different decoding network systems can be established according to different situations to adapt to actual needs.

[0011] Still further optionally, the sum of the first phoneme sequence and the second phoneme sequence includes all phoneme sequences of the keyword identified on the bridges. The beneficial effects are that: all phoneme sequences in the same keyword can be identified separately, so that the recognition efficiency is ensured while the size of the decoding network system is reduced.

[0012] Optionally, the decoding network system is connected with a phoneme sequence module, and the phoneme sequence module comprises an acoustic model composed of a hidden Markov model, a Gaussian mixture model or a deep neural network model, so as to obtain phoneme sequences to be identified by the decoding network system.

[0013] Further optionally, the all-phoneme sequence of the keyword comprises at least one of a bi-phoneme sequence or a tri-factor sequence of the keyword.

[0014] In a second aspect, the present application provides a method for speech recognition, comprising: obtaining a speech and the decoding network system according to any one of the first aspect; the decoding network system obtaining an all-phoneme sequence of the speech; and the decoding network system matching the all-phoneme sequence with the first phoneme sequence and the second phoneme sequence to realize speech recognition.

[0015] The beneficial effects are that the all-phoneme sequence of the speech is matched with the first phoneme sequence and the second phoneme sequence to realize speech recognition, which ensures the recognition efficiency of the decoding network system and reduces the specification of the decoding network system.

[0016] In a third aspect, the present application provides a speech recognition device, which comprises a module / unit for executing the method according to any one of the second aspect. The module / unit can be implemented by hardware or by executing corresponding software by hardware.

[0017] In a fourth aspect, the present application provides an electronic device, which comprises a processor and a memory. The memory is used to store one or more computer programs, and when the one or more computer programs stored in the memory are executed by the processor, the electronic device can implement the method according to any one of the second aspect.

[0018] The beneficial effects of the third aspect to the fourth aspect can be referred to the description of the first aspect or the second aspect. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A schematic diagram of an embodiment of the decoding network system provided by the present application;

[0020] Figure 2 A schematic diagram of another embodiment of the decoding network system provided by the present application;

[0021] Figure 3 A flowchart of a method for speech recognition provided by the present application;

[0022] Figure 4 A schematic diagram of a speech recognition device provided by the present application;

[0023] Figure 5 A schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0024] Before the embodiments of the present application are described in detail, the following first explains some terms in the embodiments of the present application, so as to facilitate understanding by those skilled in the art.

[0025] 1. A phone is the smallest unit of speech divided according to the natural properties of speech, and is analyzed according to the pronunciation actions in a syllable, one action constituting one phone. Phones are divided into two categories: vowels and consonants.

[0026] 2. A weighted finite-state transducer (WFST) is used for large-scale speech recognition, and the change of states can be marked by input symbols and output symbols.

[0027] 3. A token is a data structure recording the score and information of a certain state at a certain moment in the decoding process. Starting from the initial state of the weighted finite-state transducer, the token is transferred along the edges with directions, and the change of states can be reflected by the change of input symbols. During the state transition from the initial state to the terminal state, the token records a series of paths composed of states and edges.

[0028] 4. An acoustic model is a representation of the differences in acoustics, linguistics, environmental variables, speaker gender, accent, etc. The acoustic model includes an acoustic model based on a hidden Markov model (hmm), such as a Gaussian mixture model-hidden Markov model (gmm-hmm) and a deep neural network-hidden Markov model (dnn-hmm), and also includes an acoustic model based on an end-to-end model, such as a connectionist temporal classification-long short-term memory (ctc-lstm).

[0029] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. In the description of the embodiments of the present application, the terms used in the following embodiments are only for the purpose of describing the specific embodiments and are not intended to be limiting on the present application. As used in the specification and the appended claims of the present application, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that “at least one” and “one or more” as used in the embodiments herein indicates one or two or more (including two), unless otherwise indicated. The term “and / or” is used to describe the association relationship of the associated objects, which means that there can be three relationships; for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character “ / ” generally represents an “or” relationship between the associated objects.

[0030] In this specification, the reference “one embodiment” or “some embodiments” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases “in one embodiment”, “in some embodiments”, “in other embodiments”, “in additional embodiments”, etc. in various places in this specification are not necessarily all referring to the same embodiment, but can refer to one or more but not all embodiments, unless otherwise specifically stated. The terms “comprising”, “including”, “having” and their variants mean “including but not limited to”, unless otherwise specifically stated. The term “connected” includes both direct connection and indirect connection, unless otherwise stated. “First”, “second”, etc. are used only for the purpose of description and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0031] In the embodiments of the present application, the words “exemplary” or “for example” are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as “exemplary” or “for example” in the embodiments of the present application is not necessarily to be construed as preferred or advantageous over other embodiments or design solutions. Rather, the use of the words “exemplary” or “for example” is intended to present concepts in a particular manner.

[0032] Speech recognition technology, also known as automatic speech recognition (ASR), aims to convert the lexical content in human speech into computer-readable input, including key presses, binary codes or character sequences, etc., so as to realize human-computer interaction. Speech recognition technology has a wide range of applications in modern life and can be applied to scenarios such as car navigation, smart home, voice dialing, simultaneous interpretation, etc. As the core of the speech recognition system, the speech decoding process based on the decoder plays an important role in the entire speech recognition process and directly affects the accuracy of the recognition result.

[0033] A weighted finite-state transducer (WFST) is usually used as a decoder for speech recognition, but this method is bulky and inefficient for keyword decoding. Currently, the large decoding network size is reduced for keyword decoding, but this brings a series of negative effects, such as low keyword recognition efficiency and low accuracy.

[0034] To solve the above problems, the present application provides a decoding network system, a speech recognition method, device, equipment and medium. The speech recognition method provided by the embodiments of the present application can be applied to various scenarios that require the use of speech recognition functions, such as smart home scenarios, voice input scenarios, car navigation scenarios, simultaneous interpretation scenarios, etc. In the above application scenarios, the implementation environment involved in the embodiments of the present application can include a terminal and a server.

[0035] The present application provides a decoding network system, which comprises a first direction network unit and a second direction network unit; the first direction network unit comprises N first keyword transmission channels, the phoneme sequence on the first keyword transmission channel comprises a first phoneme sequence of a keyword, and the N first keyword transmission channels have the same first zero-in-degree node and different first zero-out-degree nodes, and N is a positive integer; the second direction network unit comprises M second keyword transmission channels, the phoneme sequence on the second keyword transmission channel comprises a second phoneme sequence of the keyword, and the M second keyword transmission channels have different second zero-in-degree nodes and the same second zero-out-degree node, and M is a positive integer; and the second zero-in-degree node is connected to at least one first zero-out-degree node.

[0036] Exemplarily, as Figure 1As shown, the decoding network system comprises a first direction network unit and a second direction network unit. The first direction network unit comprises 6 first keyword transmission channels, which are a first A keyword transmission channel 101, a first B keyword transmission channel 102, a first C keyword transmission channel 103, a first D keyword transmission channel 104, a first E keyword transmission channel 105 and a first F keyword transmission channel 106. The phoneme sequence on each first keyword transmission channel corresponds to the first phoneme sequence of a keyword, and the first keyword transmission channels have a unique first zero-in-degree node 100 and 6 different first zero-out-degree nodes (not shown in the figure). The second direction network unit comprises 6 second keyword transmission channels, which are a second A keyword transmission channel 107, a second B keyword transmission channel 108, a second C keyword transmission channel 109, a second D keyword transmission channel 110, a second E keyword transmission channel 111 and a second F keyword transmission channel 112. The phoneme sequence on each second keyword transmission channel corresponds to the second phoneme sequence of a keyword, and the second keyword transmission channels have 6 different second zero-in-degree nodes (not shown in the figure) and a unique second zero-out-degree node 113. Each second zero-in-degree node is connected to one first zero-out-degree node, and the connection relationship is as shown in the connection relationship between the first keyword transmission channels and the second keyword transmission channels in the middle of FIG. 1. Figure 1

[0037] ​Exemplarily, the first phoneme sequence comprises a front several phoneme sequence of the keyword, and the front several factor sequence comprises zero phoneme, one phoneme or multiple phonemes. The second phoneme sequence comprises a rear several phoneme sequence of the keyword, and the rear several phoneme sequence comprises zero phoneme, one phoneme or multiple phonemes. The sum of the front several phoneme sequence and the rear several phoneme sequence is the whole phoneme sequence of the keyword. The front several phoneme sequence refers to the sequence corresponding to the phonemes which are uttered first by a person, and the rear several phoneme sequence refers to the sequence corresponding to the phonemes which are uttered last by the person. When a piece of pronunciation is completed, the decoding net system obtains the whole phoneme sequence in the pronunciation through the processing of the acoustic model, and then screens out the phoneme sequence matching the front several phoneme sequence in the whole phoneme sequence and the phoneme sequence matching the rear several phoneme sequence, so as to realize the recognition of the keyword contained in the speech. Meanwhile, the keyword can be recognized through different keyword transmission channels, which can improve the accuracy of the recognition of the keyword wrapped in the speech, and because different keyword transmission channels correspond to different keyword phoneme sequences respectively, the physical meaning of the nodes on each transmission channel is clear, so the decoding net system can be optimized in a targeted manner. Because the decoding net system does not need to backtrack the decoding process, the efficiency of the keyword recognition in speech recognition is improved. Alternatively, different keyword transmission channels correspond to the same or different phoneme sequences.

[0038] In another possible embodiment, the decoding net system further comprises L bridges, wherein the bridges are provided with keyword identifiers, the second zero-in-degree nodes are connected to the first zero-in-degree nodes through the bridges, and L is a positive integer. Exemplarily, each of the bridges is provided with an identifier of a keyword, and each of the second zero-in-degree nodes is connected to one of the first zero-in-degree nodes through the bridge. In this embodiment, the identifiers of the bridges can be used to more accurately match and connect the second zero-out-degree nodes and the first zero-in-degree nodes.

[0039] In a possible embodiment, the line structure of the first directional net unit is a directed acyclic graph, the line structure of the second directional net unit is a directed acyclic graph, and the directions of the first directional net unit and the second directional net unit are opposite. Exemplarily, in the simplified state, the first directional net unit is a tree, and the second directional net unit is a reverse tree. In this embodiment, because the first directional net unit and the second directional net unit are directed acyclic graphs, the backtracking decoding phenomenon does not occur in the process of speech recognition of the first directional net unit and the second directional net unit, and the efficiency of keyword recognition in speech recognition is improved.

[0040] In yet another possible embodiment, the line structure of the first keyword transmission channel is a straight path or a tree path; and the line structure of the second keyword transmission channel is a straight path or a tree path. Thus, different decoding network systems can be established according to different situations to adapt to actual needs.

[0041] For example, when the N, the M and the L are all 1, the structure of the decoding network system is as shown in Figure 2 The decoding network system as shown in Figure 2 The decoding network system as shown in

[0042] In another example, although a plurality of phonemes can be provided on the first keyword transmission channel and the second keyword transmission channel, the first keyword transmission channel and the second keyword transmission channel can not take a certain phoneme as a fixed starting recognition point in the process of voice recognition. In some embodiments, the first keyword transmission channel and the second keyword transmission channel take any one phoneme as a starting recognition point in the process of voice recognition. For example, in the above embodiment, Figure 2 The decoding network system as shown in

[0043] In still another possible embodiment, the sum of the first phoneme sequence and the second phoneme sequence includes all phoneme sequences of the keyword identified on the bridge. Further, the sum of the first phoneme sequence and the second phoneme sequence is all phoneme sequences of the keyword identified on the bridge. For example, the sum of the first phoneme sequence and the second phoneme sequence corresponding to the first keyword transmission path and the second keyword transmission path connected by the same bridge is all phoneme sequences of the keyword identified on the bridge. In this way, all phoneme sequences of the same keyword can be identified separately, which ensures the recognition efficiency and reduces the specification of the decoding network system. The first keyword transmission path in the first direction network unit does not need to carry a keyword information token at all, and each second zero-in-degree node in the second direction network unit stores a keyword information.

[0044] In still another possible embodiment, the decoding network system is connected with a phoneme sequence module, and the phoneme sequence module includes an acoustic model composed of a hidden Markov model, a Gaussian mixture model or a deep neural network model to obtain phoneme sequences to be recognized by the decoding network system.

[0045] In a possible embodiment, the all phoneme sequences of the keyword include at least one of a bi-phoneme sequence or a tri-factor sequence of the keyword.

[0046] Based on the decoding network system provided in the above embodiments, an embodiment of the present application provides a speech recognition method, and a flowchart of the method is shown in Figure 3 The specific steps are as follows.

[0047] S301, obtaining a speech and a decoding network system as described in any one of the above embodiments.

[0048] In this step, the speech can be real-time speech information issued by a person, or recorded information obtained by a voice recording device. The voice recording device includes software for recording in a terminal device or software containing a recording function, such as a Recorder application in a mobile phone.

[0049] S302, the decoding network system obtains all phoneme sequences of the speech. In this step, the decoding network system is connected with a phoneme sequence module, and the phoneme sequence module includes an acoustic model composed of a hidden Markov model, a Gaussian mixture model or a deep neural network model, and all phoneme sequences in the speech can be obtained through the acoustic model.

[0050] S303, the decoding network system matches the all phoneme sequences with the first phoneme sequence and the second phoneme sequence to realize speech recognition.

[0051] In this step, the keyword all phoneme sequence can be matched with the first phoneme sequence and the second phoneme sequence through the identification of the bridge, and the sum of the first phoneme sequence and the second phoneme sequence corresponding to the first keyword transmission channel and the second keyword transmission channel connected by the same bridge is the all phoneme sequence of the keyword identified on the bridge. For example, the all phoneme sequence in the same keyword can be recognized separately, and the first keyword transmission channel in the first direction network unit does not need to carry the keyword information token at all, and a keyword information is stored on each second zero-in-degree node in the second direction network unit. In this way, the accuracy of keyword recognition in speech recognition can be improved.

[0052] In a possible embodiment, the all phoneme sequence of the keyword includes at least one of a bigram sequence or a triphone sequence of the keyword.

[0053] It should be noted that the speech recognition method provided by the embodiments of the present application can be applied to various speech recognition chips, speech recognition systems and devices with speech recognition functions, such as smart phones, tablet electronic devices, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDA), vehicle-mounted devices, smart wearable devices, toys, smart home control devices, and pipeline device controllers. The smart wearable device can include smart glasses, smart watches, and smart bracelets.

[0054] The speech recognition method provided by the above embodiments can be implemented by the speech recognition apparatus provided by the embodiments of the present application, as shown in Figure 4 The apparatus includes a speech acquisition unit 401 and a decoding network system 402 as described in any of the above embodiments. The speech acquisition unit 401 is configured to acquire speech. The speech acquired by the speech acquisition unit 401 is from various scenarios using speech recognition functions, for example, smart home scenarios, speech input scenarios, vehicle-mounted navigation scenarios, simultaneous interpretation scenarios, and the like. The decoding network system 402 includes an extraction module 4021 and an identification module 4022. The extraction module 4021 is configured to acquire all phoneme sequences of the speech. The extraction module 4021 can acquire all phoneme sequences of the speech processed by an acoustic model. The acoustic model can be modeled by a hidden Markov model, a Gaussian mixture model, or a deep neural network model. The identification module 4022 includes a first identification unit and a second identification unit. Figure 4The first recognition unit is configured to match the entire phoneme sequence with the first phoneme sequence in the decoding network system 4022, and the second recognition unit is configured to match the phoneme sequence contained in the entire phoneme sequence and the first phoneme sequence with the second phoneme sequence in the decoding network system 4022, so as to realize voice recognition.

[0055] In a possible embodiment, the entire phoneme sequence of the keyword includes at least one of a bigram sequence or a triphone sequence of the keyword.

[0056] By using the technical solution, the phoneme sequence stored in the first recognition unit and the second recognition unit is output during the operation of the device, and the entire phoneme sequence in the voice is matched in real time, so that the recognition result can be given without backtracking the cached path history information at the end of recording, and the real-time performance of keyword recognition is improved.

[0057] The voice recognition device can be integrated in an electronic device, which can be a terminal, a server, or the like. The terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a notebook computer, or a personal computer (PC). The server can be a single server or a server cluster composed of multiple servers.

[0058] In some embodiments, the voice recognition device can also be integrated in multiple electronic devices. For example, the voice recognition device can be integrated in multiple servers, and the servers can implement the voice recognition method of the present application. In some embodiments, the server can also be implemented in the form of a terminal. For example, the electronic device can be a terminal, and the voice recognition device can be integrated in the terminal. Figure 5 As shown in the figure, the electronic device 500 includes one or more processors 501, a memory 502, a display 503, one or more application programs (not shown), and one or more computer programs 504. The above devices can be connected through one or more communication buses 505. The one or more computer programs 504 are stored in the memory 502 and configured to be executed by the one or more processors 501. The one or more computer programs 504 include instructions that can be used to perform the steps in the above embodiments.

[0059] Those skilled in the art can clearly understand the technical solutions of the present application according to the above description of the embodiments, and the division of the functional modules is only used for description and simplification, and in actual application, the functions can be completed by different functional modules according to the needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0060] The functional units in the various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0061] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application essentially or the part that makes contributions to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device) or a processor to execute all or part of the steps of the methods of the various embodiments of the present application. The foregoing storage medium is a medium that can store program codes, including: a flash memory, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.

[0062] The above description is only a specific implementation of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto. Any change or replacement within the technical scope disclosed in the embodiments of the present application should be covered in the protection scope of the embodiments of the present application. Therefore, the protection scope of the embodiments of the present application should be subject to the protection scope of the claims.

Claims

1. A decoding mesh system, characterized by, The decoding network system comprises: a first direction network unit and a second direction network unit; the first direction network unit comprises N first keyword transmission channels, the phoneme sequence on the first keyword transmission channel comprises a first phoneme sequence of a keyword, and the N first keyword transmission channels have the same first zero-in-degree node and different first zero-out-degree nodes, and N is a positive integer; the second direction network unit comprises M second keyword transmission channels, the phoneme sequence on the second keyword transmission channel comprises a second phoneme sequence of the keyword, and the M second keyword transmission channels have different second zero-in-degree nodes and the same second zero-out-degree node, and M is a positive integer; the second zero-in-degree node is connected to at least one first zero-out-degree node; the line structure of the first direction network unit is a directed acyclic graph, the line structure of the second direction network unit is a directed acyclic graph, and the directions of the first direction network unit and the second direction network unit are opposite; none of the first keyword transmission channels in the first direction network unit carries a keyword information token, and each second zero-in-degree node in the second direction network unit stores a keyword information.

2. The decoding web system of claim 1, wherein, The decoding network system further comprises L bridges, the bridges are provided with keyword identifiers, the second zero-in-degree node is connected to the first zero-in-degree node through the bridges, and L is a positive integer.

3. The decoding web system of claim 2, wherein, The line structure of the first keyword transmission channel is a straight line path or a tree path; and the second keyword transmission channel is a straight line path or a tree path.

4. The decoding web system of claim 3, wherein, The sum of the first phoneme sequence and the second phoneme sequence comprises all phoneme sequences of the keyword identified on the bridges.

5. The decoding web system of claim 4, wherein, The decoding network system is connected to a phoneme sequence module, and the phoneme sequence module comprises an acoustic model composed of a hidden Markov model, a Gaussian mixture model or a deep neural network model to obtain a phoneme sequence to be recognized by the decoding network system.

6. The decoding web system of claim 4, wherein, All phoneme sequences of the keyword comprise at least one of a triphone sequence or a triphone sequence of the keyword.

7. A method of speech recognition, characterized by The decoding network system comprises: obtaining a voice and the decoding network system according to any one of claims 1 to 6; the decoding network system obtains all phoneme sequences of the voice; the decoding network system matches the all phoneme sequences with the first phoneme sequence and the second phoneme sequence to realize voice recognition.

8. A speech recognition apparatus characterized by comprising: The decoding network system comprises: a voice obtaining unit and the decoding network system according to any one of claims 1 to 6, the voice obtaining unit is used to obtain a voice, and the decoding network system comprises an extraction module and an identification module; the extraction module is used to obtain all phoneme sequences of the voice; the identification module comprises a first identification unit and a second identification unit, the first identification unit is used to match the all phoneme sequences with the first phoneme sequence, and the second identification unit is used to match the phoneme sequence containing the first phoneme sequence in the all phoneme sequences with the second phoneme sequence to realize voice recognition.

9. An electronic device, comprising: The computer program is executed by a processor to implement the method for voice recognition according to claim 7.

10. A computer-readable storage medium having stored therein a computer program, characterized in that, The computer program is executed by a processor to implement the method for voice recognition according to claim 7.

Images