Absorption network system and construction method, device, equipment and medium thereof

Abstract

The present invention provides an absorption network system and its construction method, apparatus, equipment, and medium. The method comprises: obtaining non-keyword phonemes and the probabilities of the non-keyword phonemes; setting N first phonemes in at least one non-keyword transmission channel; sequentially combining M second phonemes with the first phonemes, and setting the second phoneme with the highest probability after the combination in the corresponding non-keyword transmission channel; and so on, until all the non-keyword phonemes are set in the non-keyword transmission channels, and the phoneme combination on the corresponding non-keyword transmission channel has the highest probability of appearing in a speech recognition application scenario; and connecting the non-keyword transmission channels in parallel to construct the absorption network system. The present invention improves the absorption rate of non-keywords in speech recognition application scenarios, thereby improving the accuracy of keyword recognition.

Description

Technical Field

[0001] The present invention relates to the field of voice processing technology, and in particular to an absorption network system and a construction method, device, equipment and medium thereof. Background Art

[0002] In keyword speech recognition, the false positive rate (ERR) is as important as the recognition rate. The ERR refers to the probability of misidentifying non-keywords as keywords. Therefore, a complete decoding network not only includes decoding for keywords but also a network for misidentifying non-keywords as keywords, known as the absorption network. The absorption network determines both the ERR and the probability of misidentifying keywords as non-keywords. The design of this network, along with the design of the decoding network, determines the performance of the entire keyword speech recognition system.

[0003] In continuous speech automatic speech recognition (ASR), there's no absorption network, so all potentially recognizable words and phrases are covered, eliminating the possibility of misidentification. However, in keyword spotting, not all recognized speech is necessarily keywords, so misidentification can occur.

[0004] Currently, there are several main approaches to addressing misidentification: 1) Using an ASR-based decoding network to absorb non-keyword phrases; 2) Using a threshold for keyword recognition probability to determine whether a misidentification is a mistake. The first approach requires a large decoding network, making it difficult to implement in practice; the second approach is significantly affected by noise, resulting in unstable absorption results.

[0005] Therefore, the present invention proposes an absorption network system and its construction method, device, equipment and medium to improve the absorption rate of non-keywords in speech recognition application scenarios, thereby improving the accuracy of keyword recognition. Summary of the Invention

[0006] The present invention proposes an absorption network system and its construction method, device, equipment and medium to improve the absorption rate of non-keywords in speech recognition application scenarios, thereby improving the accuracy of keyword recognition.

[0007] In a first aspect, the present invention provides a method for constructing an absorption network system, comprising: obtaining non-keyword phonemes and the probability of the non-keyword phonemes, the probability being obtained by counting the frequency of occurrence of the non-keyword phonemes in a speech recognition application scenario; setting N first phonemes in at least one non-keyword transmission channel, the first phoneme being a phoneme among the non-keyword phonemes whose probability exceeds a first preset threshold, and N being a positive integer; combining M second phonemes with the first phonemes in sequence, and setting the second phoneme with the highest probability after the combination in the corresponding non-keyword transmission channel, the second phoneme being a phoneme among the non-keyword phonemes whose probability exceeds a second preset threshold but does not exceed the first preset threshold, and M being a positive integer; and so on, until all the non-keyword phonemes are set in the non-keyword transmission channel, and the probability of the phoneme combination on the corresponding non-keyword transmission channel occurring in the speech recognition application scenario is maximized; and connecting the non-keyword transmission channels in parallel to construct the absorption network system.

[0008] Its beneficial effect is that: the present invention has the highest probability of appearing in the speech recognition application scenario through the phoneme combination on the non-keyword transmission channel, and can achieve the strongest absorption ability of non-keyword phonemes in the speech recognition application scenario, further improving the recognition efficiency of keywords in the speech recognition application scenario.

[0009] Optionally, the method for constructing the absorption network system further includes: obtaining non-speech phonemes and placing the non-speech phonemes in the non-keyword transmission channel. This advantageously allows for the absorption of non-speech phonemes present in speech recognition application scenarios, as they may be present in speech recognition application scenarios. By placing the non-speech phonemes in the non-keyword transmission channel, the non-speech phonemes present in speech recognition application scenarios can be absorbed, further improving keyword recognition efficiency.

[0010] Further optionally, obtaining non-speech phonemes and setting them in the non-keyword transmission channel includes obtaining the non-speech phonemes and, based on the number of the non-keyword phonemes set in the non-keyword transmission channel exceeding the number of the non-speech phonemes, setting the non-speech phonemes in the non-keyword transmission channel. This advantageously enables the number of the non-keyword phonemes set in the non-keyword transmission channel to exceed the number of the non-speech phonemes, thereby avoiding weakening the non-keyword transmission channel's ability to absorb the non-keyword phonemes.

[0011] Furthermore, the method for constructing the absorption network system optionally further includes obtaining keyword phonemes and placing the keyword phonemes in the non-keyword transmission channel. This advantageously improves the efficiency of absorbing non-keywords, further improving the efficiency of recognizing keywords, as non-keywords often contain keyword phonemes in speech recognition applications.

[0012] Further optionally, obtaining the keyword phonemes and setting the keyword phonemes in the non-keyword transmission channel includes obtaining the keyword phonemes and, based on the number of the non-keyword phonemes set in the non-keyword transmission channel exceeding the number of the keyword phonemes, setting the keyword phonemes in the non-keyword transmission channel. This advantageously allows the number of the non-keyword phonemes set in the non-keyword transmission channel to exceed the number of the keyword phonemes, thereby avoiding weakening the non-keyword transmission channel's ability to absorb the non-keyword phonemes.

[0013] Further optionally, the ratio of the number of non-keyword phonemes, the number of keyword phonemes, and the number of non-speech phonemes in the non-keyword transmission channel is 3:1:1. This advantageously provides the following advantages: in some cases, the non-keyword transmission channel configured in this manner has a higher absorption capacity for non-keyword phonemes and a higher efficiency in keyword recognition than other embodiments.

[0014] In a second aspect, the present invention provides a device for constructing an absorption network system, the device comprising modules and units for executing the construction method described in any one of the first aspects. These modules and units can be implemented in hardware, or in hardware executing corresponding software.

[0015] In a third aspect, the present invention provides an absorption network system obtained by the construction method described in any one of the first aspects, comprising: the non-keyword transmission channels connected in parallel, wherein the non-keyword transmission channels are provided with a phoneme combination including non-keyword phonemes.

[0016] In a fourth aspect, an embodiment of the present application provides an electronic device comprising a processor and a memory. The memory is configured to store one or more computer programs; when the one or more computer programs stored in the memory are executed by the processor, the electronic device is capable of implementing any possible design method of the first aspect.

[0017] In a fifth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method as described in any one of the above embodiments is implemented.

[0018] For the beneficial effects of the second to fifth aspects, please refer to the description of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of an embodiment of a method for constructing an absorption network system provided in this application;

[0020] Figure 2 A schematic diagram of an embodiment of a construction device for an absorption net system provided in this application;

[0021] Figure 3 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application. Among them, in the description of the embodiments of the present application, the terms used in the following embodiments are only for the purpose of describing specific embodiments, and are not intended to be used as limitations on the present application. As used in the specification and claims of the present application, the singular expressions "a", "the", "above", "the" and "this" are intended to also include expressions such as "one or more", unless there is a clear contrary indication in the context. It should also be understood that in the following embodiments of the present application, "at least one", "one or more" refer to one or more (including two). The term "and / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist; for example, A and / or B can represent: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0023] References to "one embodiment" or "some embodiments" etc. described in this specification mean that the specific features, structures or characteristics described in conjunction with the embodiment are included in one or more embodiments of the present application. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. appearing in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in another way. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in another way. The term "connected" includes direct and indirect connections, unless otherwise stated. "First" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated.

[0024] In the embodiments of this application, words such as "exemplarily" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplarily" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplarily" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] Generally, voice recognition is mostly used to control smart devices equipped with voice recognition functions based on the text commands contained in the voice uttered by the user. For example: "Play a song", "Set an alarm for 8:30". Then, the "play" and "song" in "play a song" are both keywords, that is, they are both instructions for smart devices and need to be executed. Among them, the instruction executed by "play" is to control the smart device to open and play audio, video and other files containing sound information; the instruction executed by "song" is to control the smart device to open a file designated as an audio type, and it may be the song track currently playing in a certain music software. So if the user issues a voice command such as "play a song", there may be noise in the environment at the same time, including the creaking sound of the decoration next door and the sound of cars speeding on the road. Then these non-speech words will be input into the voice recognition system (or voice recognition chip) of the smart device together with the speech part. When these non-speech phonemes are recognized as containing phoneme states, it will affect the subsequent execution of commands by the smart device; or, when the user issues a command, the words spoken due to hesitation or stuttering contain non-keywords, such as "play, um, song", then "um" is a non-keyword at this time and needs to be processed, otherwise it may affect the smart device's recognition of keywords.

[0026] To address the above-mentioned issues, embodiments of the present invention provide an absorption network system and its construction method, apparatus, device, and medium to improve the absorption rate of non-keywords, thereby improving the accuracy of keyword recognition in speech recognition. The absorption network system provided by the present invention can be applied to various speech recognition chips, speech recognition systems, and devices with speech recognition capabilities, such as smartphones, tablet electronic devices, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, toys, smart home control devices, and assembly line equipment controllers. The smart wearable devices may include smart glasses, smart watches, and smart bracelets.

[0027] The present invention provides a method for constructing an absorption network system, the process of which is as follows: Figure 1 Shown, including:

[0028] S101 : Acquire non-keyword phonemes and the probability of the non-keyword phonemes, wherein the probability is obtained by counting the frequency of occurrence of the non-keyword phonemes in a speech recognition application scenario.

[0029] S102: Set N first phonemes in at least one non-keyword transmission channel, where the first phonemes are phonemes whose probabilities exceed a first preset threshold among the non-keyword phonemes, and N is a positive integer.

[0030] In this step, the first preset threshold can be set based on the number of first phonemes that are actually required to be screened. For example, if 90% or more of the phonemes are 8, and 85% or more of the phonemes are 10, and the number of first phonemes to be screened is 10, then the first preset threshold is 85%. Furthermore, these phonemes can be set in the same non-keyword transmission channel or in different non-keyword transmission channels.

[0031] S103, combining M second phonemes with the first phoneme in sequence, and setting the second phoneme with the highest probability after combination in the corresponding non-keyword transmission channel, wherein the second phoneme is a phoneme among the non-keyword phonemes whose probability exceeds a second preset threshold but does not exceed the first preset threshold, and M is a positive integer.

[0032] In this step, the second preset threshold can be set according to the number of the second phonemes that are actually required to be screened out.

[0033] S104, and so on, until all the non-keyword phonemes are set in the non-keyword transmission channel, and the phoneme combination on the corresponding non-keyword transmission channel has the highest probability of appearing in the speech recognition application scenario, and the non-keyword transmission channels are connected in parallel to construct the absorption network system.

[0034] For example, assume that the non-keyword phonemes contain 30 phonemes. According to statistics, the probability of 10 phonemes is 80% or more, 10 phonemes are between 60% and 80% (including 60%), and the remaining 10 phonemes are between 30% and 60% (including 30%). There are exactly 10 non-keyword transmission channels in the absorption network system. Then, the 10 phonemes with a probability of more than 80% can be set in these 10 non-keyword transmission channels respectively, and then the 10 phonemes with a probability between 60% and 80% (including 60%) need to be combined with the phonemes in each of the non-keyword transmission channels, and the phoneme combination that maximizes the probability of the non-keyword phonemes on the corresponding non-keyword transmission channel is retained. Finally, the remaining 10 phonemes are combined again with the phonemes in each of the non-keyword transmission channels, and the phoneme combination that maximizes the probability of the non-keyword phonemes on the corresponding non-keyword transmission channel is retained.

[0035] Of course, the above example is only an ideal case. In another possible example, it is also possible that the non-keyword phonemes include 30 phonemes, 8 of which have a probability of 80% or more, 10 phonemes between 60% and 80% (including 60%), and the remaining 12 phonemes between 30% and 60% (including 30%). There are exactly 10 non-keyword transmission channels in the absorption network system. Then, the 8 phonemes with a probability of more than 80% can be first set in 8 of the 10 non-keyword transmission channels, and then each of the 10 phonemes with a probability between 60% and 80% (including 60%) needs to be combined with the phonemes in each of the non-keyword transmission channels, and the phoneme combination that maximizes the probability of the non-keyword phonemes on the corresponding non-keyword transmission channel is retained, and the remaining 2 phonemes are placed in the non-keyword transmission channel without non-keyword phonemes. Finally, the remaining 12 phonemes are combined again with the phonemes in each of the non-keyword transmission channels, and the phoneme combination that maximizes the probability of occurrence of the non-keyword phonemes on the corresponding non-keyword transmission channel is retained. Then, the remaining two phonemes are combined again with the phonemes in each of the non-keyword transmission channels, and these two phonemes are set in the non-keyword transmission channel where the phoneme combination with the highest probability is located.

[0036] To address the issues mentioned in the above embodiment, the non-keyword portion "hmm" can be absorbed through the non-keyword transmission channel to ensure that the portion used for speech recognition is the keyword "play song." In this embodiment of the present application, the present invention maximizes the probability of the phoneme combination on the non-keyword transmission channel appearing in the speech recognition application scenario, thereby achieving the strongest absorption capability for non-keyword phonemes in the speech recognition application scenario, further improving the efficiency of keyword recognition in the speech recognition application scenario.

[0037] In a possible embodiment, the construction method of the absorption network system further includes: obtaining non-speech phonemes, and setting the non-speech phonemes in the non-keyword transmission channel. In this embodiment, because non-speech phonemes may exist in the speech recognition application scenario, by setting the non-speech phonemes in the non-keyword transmission channel, the non-speech phonemes existing in the speech recognition application scenario can be absorbed, further improving the efficiency of keyword recognition. For example, with respect to the squeaking sound of the decoration next door and the sound of cars speeding on the road mentioned in the above embodiment, these non-speech phonemes that exist in real life and non-speech phonemes that people can expect can be absorbed through the non-keyword transmission channel to further ensure that what is recognized in the speech recognition application scenario is the keyword part "playing songs".

[0038] In another possible embodiment, the obtaining of non-speech phonemes and setting the non-speech phonemes in the non-keyword transmission channel includes: obtaining the non-speech phonemes and setting the non-speech phonemes in the non-keyword transmission channel based on the number of the non-keyword phonemes set in the non-keyword transmission channel exceeding the number of the non-speech phonemes.

[0039] In this embodiment, the number of the non-keyword phonemes set in the non-keyword transmission channel exceeds the number of the non-speech phonemes, so as to avoid weakening the absorption capacity of the non-keyword transmission channel for the non-keyword phonemes.

[0040] In another possible embodiment, the construction method of the absorption network system further includes: obtaining keyword phonemes, and setting the keyword phonemes in the non-keyword transmission channel. In this embodiment, because non-keywords will contain certain keyword phonemes in the speech recognition application scenario, the keyword phonemes are set in the non-keyword transmission channel to improve the absorption efficiency of the non-keywords, and further improve the recognition efficiency of keywords. Exemplarily, non-keywords may carry some words in some keywords, for example: the "close" in the keyword "turn off the air conditioner" may be the "close" in the non-keyword "close your eyes and relax"; the "open" in the keyword "turn on the air conditioner" may be the "open" in the non-keyword "dinner is ready".

[0041] In yet other possible embodiments, obtaining keyword phonemes and setting the keyword phonemes in the non-keyword transmission channel includes obtaining the keyword phonemes and, based on the number of the non-keyword phonemes set in the non-keyword transmission channel exceeding the number of the keyword phonemes, setting the keyword phonemes in the non-keyword transmission channel. In this embodiment, the number of the non-keyword phonemes set in the non-keyword transmission channel exceeds the number of the keyword phonemes to avoid weakening the non-keyword transmission channel's ability to absorb the non-keyword phonemes.

[0042] In some other possible embodiments, the ratio of the number of the non-keyword phonemes, the number of the keyword phonemes, and the number of the non-speech phonemes in the non-keyword transmission channel is 3:1:1. In this embodiment, through experiments conducted by the researchers of this patent, it was found that by setting the ratio of the number of the non-keyword phonemes, the number of the keyword phonemes, and the number of the non-speech phonemes in the non-keyword transmission channel to 3:1:1, in some cases, the absorption capacity of non-keyword phonemes is better than that of other embodiments, and the efficiency of keyword recognition is better than that of other embodiments. Optionally, the ratio of the number of the non-keyword phonemes, the number of the keyword phonemes, and the number of the non-speech phonemes in the non-keyword transmission channel is 3:0:1. In another optional embodiment, the ratio of the number of the non-keyword phonemes, the number of the keyword phonemes, and the number of the non-speech phonemes in the non-keyword transmission channel is 3:1:0. In practical applications, the ratio of the number of the non-keyword phonemes, the number of the keyword phonemes and the number of the non-speech phonemes in the non-keyword transmission channel can be flexibly designed and is not limited to the design method mentioned in this application document.

[0043] In order to implement the construction method as described in any of the above embodiments, the present invention provides a construction device for an absorption network system, such as Figure 2As shown, the construction device includes: an acquisition module 201, a classification module 202 and a processing module 203. These modules and units can be implemented by hardware, or by hardware executing corresponding software.

[0044] The acquisition module 201 is used to acquire non-keyword phonemes and the probability of the non-keyword phonemes, where the probability is obtained by counting the frequency of occurrence of the non-keyword phonemes in speech recognition application scenarios.

[0045] The classification module 202 includes at least L classification units, namely a first classification unit 2021, a second classification unit 2022, ..., an Lth classification unit 202L, where L is a positive integer greater than or equal to 2. The first classification unit 2021 is used to set N first phonemes in at least one non-keyword transmission channel, where the first phoneme is a phoneme among the non-keyword phonemes whose probability exceeds a first preset threshold, and N is a positive integer; the second classification unit 2022 is used to combine M second phonemes with the first phoneme in sequence, and set the second phoneme with the highest probability after the combination in the corresponding non-keyword transmission channel, where the second phoneme is a phoneme among the non-keyword phonemes whose probability exceeds a second preset threshold but does not exceed the first preset threshold, and M is a positive integer; and so on, until the Lth classification unit 202L sets all the non-keyword phonemes in the non-keyword transmission channel, and makes the phoneme combination on the corresponding non-keyword transmission channel have the highest probability of appearing in the speech recognition application scenario.

[0046] The processing module 203 is used to connect the non-keyword transmission channels in parallel to construct the absorption network system.

[0047] As described in any of the above embodiments, the present invention obtains an absorption network system, which is obtained by the construction method described in any of the first aspects, including: the non-keyword transmission channels in parallel, and the non-keyword transmission channels are provided with a phoneme combination including non-keyword phonemes.

[0048] In this embodiment, the line structure of the non-keyword transmission channel can be a path or a tree structure, and when the line structure of the non-keyword transmission channel is a tree structure, one or more phonemes can be set on each node of the tree, and the phonemes set on each node can be phonemes with a higher frequency of occurrence.

[0049] Because the vast majority of commands issued by ordinary people can be preset according to the functions of smart devices, the phoneme states of the possible appearance of keywords and non-keywords in the voice can be set in advance, and through different combinations, the absorption network system can correspond to various occasions where each factor may appear, thereby improving the accuracy of keyword recognition in voice recognition application scenarios. For example, when a smart device includes a device that listens to music software, the phoneme states of the voice corresponding to the keywords that may appear in the smart device may include "play something", "I want to listen to something", "play the next song", "loop something", where "something" can refer to the name of the song. Then the phonemes corresponding to the remaining words can be defined as non-keyword phonemes.

[0050] Moreover, because some non-keyword parts in the commands issued by ordinary people can be judged based on the coherence of language expression, including whether the semantics are coherent, whether the sentences are coherent, and finding words that are not comprehensible from the speech, the phoneme states belonging to the non-keywords can be identified to further improve the accuracy of the absorption network system in absorbing non-keyword phonemes.

[0051] All relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding unit module and will not be repeated here.

[0052] In other embodiments of the present application, the present application discloses an electronic device, such as Figure 3 As shown, the electronic device may include: one or more processors 301; a memory 302; a display 303; one or more application programs (not shown); and one or more computer programs 304. The above components may be connected via one or more communication buses 305. The one or more computer programs 304 are stored in the memory 302 and configured to be executed by the one or more processors 301. The one or more computer programs 304 include instructions, which may be used to execute the various steps in the above corresponding embodiments.

[0053] Through the description of the above embodiments, those skilled in the art will clearly understand that for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0054] The functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0055] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method of each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as flash memory, mobile hard disk, read-only memory, random access memory, magnetic disk or optical disk.

[0056] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for constructing an absorption network system, characterized in that: include: Obtaining non-keyword phonemes and probabilities of the non-keyword phonemes, wherein the probabilities are obtained by counting the frequencies of the non-keyword phonemes appearing in speech recognition application scenarios; Setting N first phonemes in a plurality of non-keyword transmission channels, wherein the first phonemes are phonemes whose probabilities among the non-keyword phonemes exceed a first preset threshold, and N is a positive integer; Combine M second phonemes with the first phoneme in each non-keyword transmission channel in sequence, and set the second phoneme with the highest probability after combination to the corresponding non-keyword transmission channel, wherein the second phoneme is a phoneme among the non-keyword phonemes whose probability exceeds a second preset threshold but does not exceed the first preset threshold, and M is a positive integer; The same process is repeated until all the non-keyword phonemes are placed in the non-keyword transmission channel, and the probability of the corresponding phoneme combination on the non-keyword transmission channel appearing in the speech recognition application scenario is maximized; The non-keyword transmission channels are connected in parallel to construct the absorption network system.

2. The method for constructing an absorption network system according to claim 1, characterized in that: Also includes: Acquire non-speech phonemes, and set the non-speech phonemes in the non-keyword transmission channel.

3. The method for constructing an absorption network system according to claim 2, characterized in that: The acquiring of non-speech phonemes and setting the non-speech phonemes in the non-keyword transmission channel includes: The non-speech phonemes are acquired, and according to the number of the non-keyword phonemes set in the non-keyword transmission channel exceeding the number of the non-speech phonemes, the non-speech phonemes are set in the non-keyword transmission channel.

4. The method for constructing an absorption network system according to claim 3, characterized in that: Also includes: Obtain keyword phonemes, and set the keyword phonemes in the non-keyword transmission channel.

5. The method for constructing an absorption network system according to claim 4, characterized in that: The acquiring of keyword phonemes and setting the keyword phonemes in the non-keyword transmission channel includes: The keyword phoneme is acquired, and according to the number of the non-keyword phonemes set in the non-keyword transmission channel exceeding the number of the keyword phonemes, the keyword phoneme is set in the non-keyword transmission channel.

6. The method for constructing an absorption network system according to claim 5, characterized in that: The ratio of the number of the non-keyword phonemes, the number of the keyword phonemes, and the number of the non-speech phonemes in the non-keyword transmission channel is 3:1:

1.

7. A device for constructing an absorption network system, characterized in that: Used to execute the construction method according to any one of claims 1 to 6, comprising: an acquisition module, a classification module and a processing module; The acquisition module is used to acquire non-keyword phonemes and the probability of the non-keyword phonemes, wherein the probability is obtained by counting the frequency of occurrence of the non-keyword phonemes in the speech recognition application scenario; The classification module is configured to set N first phonemes in at least one non-keyword transmission channel, wherein the first phonemes are phonemes whose probabilities among the non-keyword phonemes exceed a first preset threshold, and N is a positive integer; The classification module is further configured to sequentially combine M second phonemes with the first phoneme in each non-keyword transmission channel, and set the second phoneme with the highest probability after the combination in the corresponding non-keyword transmission channel, wherein the second phoneme is a phoneme among the non-keyword phonemes whose probability exceeds a second preset threshold but does not exceed the first preset threshold, and M is a positive integer; The same process is repeated until the classification module places all the non-keyword phonemes in the non-keyword transmission channel, and the probability of the corresponding phoneme combination on the non-keyword transmission channel appearing in the speech recognition application scenario is maximized; The processing module is used to connect the non-keyword transmission channels in parallel to construct the absorption network system.

8. An absorption net system, characterized in that: The method according to any one of claims 1 to 6 is used to obtain the non-keyword transmission channel, comprising: the non-keyword transmission channel being connected in parallel, wherein a phoneme combination including non-keyword phonemes is provided on the non-keyword transmission channel.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the computer program is executed by the processor, the processor implements the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images