A speech recognition method, device and storage medium

Abstract

The present invention discloses a speech recognition method, device and storage medium. The method comprises: receiving a speech signal input by a user; detecting whether the speech signal contains a preset command word, if so, converting the speech signal to obtain a second speech signal, and further detecting whether the second speech signal contains a preset command word, if so, performing an operation corresponding to the preset command word. If the speech signal input by the user is not a preset command word, after the conversion process, its similarity with the preset command word will decrease, and it will be more easily recognized as a non-preset command word, thereby greatly reducing the error rate of preset command word recognition.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition, and in particular to a speech recognition method, device and storage medium. Background Art

[0002] With the continuous promotion and application of artificial intelligence technology, voice interaction products are increasingly entering people's work and life. Among them, most of the voice interaction processes are initiated by the user saying the wake-up command word. However, people who have used voice interaction devices often have such an experience: when no one says the wake-up command word to it, the voice device is awakened, which is also called false awakening.

[0003] This situation mainly occurs because the voice interaction device periodically collects voice and detects wake-up command words on the collected voice. Some voice signals are very similar to the wake-up command word voice signals and are thus mistakenly identified as wake-up command words.

[0004] The current method to reduce false awakenings is to increase the amount of speech data and the number of parameters and robustness of the detection algorithm. Unfortunately, practice has shown that the effects of these solutions are not satisfactory. Some solutions can only improve the false awakening rate in certain scenarios, but in most cases they still cannot significantly reduce the false awakening rate; while some solutions, although they can reduce the false awakening rate, will also have a negative impact on the awakening rate.

[0005] Therefore, how to reduce the false awakening rate and reduce the impact on the awakening rate is still a technical problem to be solved. Summary of the invention

[0006] In view of the above problems, embodiments of the present invention provide a speech recognition method, device and storage medium.

[0007] According to the first aspect of an embodiment of the present invention, a speech recognition method includes: receiving a first voice signal; detecting whether the first voice signal contains a preset command word, and if so, performing at least one conversion process on the first voice signal to obtain a second voice signal, and further detecting whether the second voice signal contains a preset command word, and if so, executing an operation corresponding to the preset command word.

[0008] According to an implementation of an embodiment of the present invention, performing at least one conversion process on a first voice signal to obtain a second voice signal includes: performing a first conversion process on the first voice signal to obtain the second voice signal.

[0009] According to an implementation of an embodiment of the present invention, the first conversion process includes a speed change process.

[0010] According to an implementation of an embodiment of the present invention, the first conversion process includes a timbre conversion process.

[0011] According to an implementation of an embodiment of the present invention, a first voice signal is converted at least once to obtain a second voice signal, including: performing a first conversion on the first voice signal to obtain a third voice signal; detecting whether the third voice signal contains a preset command word, and if so, performing a second conversion on the third voice signal to obtain a second voice signal.

[0012] According to an implementation of an embodiment of the present invention, the first conversion process includes a speed change process, and the second conversion process includes a timbre conversion process.

[0013] According to an implementation manner of an embodiment of the present invention, when detecting whether the first voice signal includes a preset command word and further detecting whether the second voice signal is a wake-up command, the detection method adopted is the same.

[0014] According to an implementation manner of an embodiment of the present invention, the detection methods are all executed on the client.

[0015] According to the second aspect of an embodiment of the present invention, a speech recognition device includes: a speech receiving module for receiving a first speech signal; a first preset command word detection module for detecting whether the first speech signal contains a preset command word; a speech signal conversion module for performing at least one conversion process on the first speech signal to obtain a second speech signal; a second preset command word detection module for further detecting whether the second speech signal contains a preset command word; and an operation execution module for executing an operation corresponding to the preset command word.

[0016] According to an implementation manner of an embodiment of the present invention, the voice signal conversion module is specifically configured to perform a first conversion process on a first voice signal to obtain a second voice signal.

[0017] According to an embodiment of the present invention, the speech signal conversion module includes: a first conversion processing submodule, used to perform a first conversion processing on the first speech signal to obtain a third speech signal; a second conversion processing submodule, used to perform a second conversion processing on the third speech signal to obtain a second speech signal.

[0018] According to an implementation of an embodiment of the present invention, the first preset command word detection module and the second preset command word detection module use the same detection method.

[0019] According to an implementation manner of an embodiment of the present invention, the first preset command word detection module and the second preset command word detection module are both arranged on the client.

[0020] According to a third aspect of an embodiment of the present invention, a computer storage medium stores program instructions on the storage medium, wherein the program instructions are used to execute any of the above-mentioned speech recognition methods when running.

[0021] The embodiment of the present invention provides a speech recognition method, device and storage medium, the method comprising: receiving a speech signal input by a user; detecting whether the speech signal contains a preset command word, if so, converting the speech signal to obtain a second speech signal, and further detecting whether the second speech signal contains a preset command word, if so, performing an operation corresponding to the preset command word. If the speech signal input by the user does not contain a preset command word, after the conversion process, its similarity with the preset command word will decrease, and it will be more easily recognized as a non-preset command word; and because the dialogue system often includes more features of the preset command word, the fault tolerance for the correct preset command word is also higher, therefore, if the speech signal input by the user really contains a preset command word, after the conversion process, it will still be recognized as a preset command word. In this way, the error rate of the speech signal after the conversion process being misidentified as a preset command word is greatly reduced, and the impact on the recognition rate of the correct preset command word is also very small.

[0022] It should be understood that the teachings of the present invention are not required to achieve all of the beneficial effects described above, but specific technical solutions can achieve specific technical effects, and other embodiments of the present invention can also achieve beneficial effects not mentioned above. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which:

[0024] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0025] Figure 1 A schematic diagram of the implementation flow of the speech recognition method according to an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of a specific implementation process of the speech recognition method applied in the first embodiment of the present invention;

[0027] Figure 3 The figure is a schematic diagram of the structure of a speech recognition device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0029] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0030] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0031] Figure 1 The implementation process of the speech recognition method according to the embodiment of the present invention is shown. Figure 1 The method includes: operation 110, receiving a first voice signal; operation 120, detecting whether the first voice signal contains a preset command word, if so, performing at least one conversion process on the first voice signal to obtain a second voice signal, and further detecting whether the second voice signal contains a preset command word, if so, executing an operation corresponding to the preset command word.

[0032] In operation 110, the first voice signal mainly refers to an audio signal similar to human voice received by a voice receiving device. The voice receiving device includes various types of microphones, headsets, and audio signal receivers. The first voice signal is usually an audio signal obtained after pre-processing such as denoising, filtering, and enhancement. The pre-processed audio signal is easier to perform voice recognition.

[0033] In operation 120, the preset command word mainly refers to a specific command word preset by the system and associated with a specific operation. When the user speaks the preset specific command word, the corresponding operation will be performed. For example, when the user speaks the wake-up command word, the wake-up operation will be performed and the conversation program will be started. The preset here mainly refers to the preset before the conversation, and it is not unchangeable. The preset command word can also be customized by the user according to personal preferences.

[0034] The detection method used when detecting whether the first voice signal contains a preset command word and the detection method used when further detecting whether the second voice signal is a wake-up command can be different or the same. The implementer can select any applicable implementation method according to the implementation conditions and implementation intentions.

[0035] The detection of whether the first voice signal contains a preset command word and the further detection of whether the second voice signal is a wake-up command can be performed both in the client, or in the client and the server respectively. The implementer can select any applicable implementation method according to the implementation conditions and implementation intentions.

[0036] For example, the process of detecting whether the first voice signal or the second voice signal contains a preset command word may include: sampling the first voice signal to obtain waveform data; then, inputting these waveform data into a feature extraction module to extract appropriate acoustic feature parameters; then, inputting the extracted acoustic feature parameters into an acoustic model (e.g., a hidden Markov model) to perform speech recognition to obtain a classification result; then, judging whether the first voice signal contains the preset command word based on the classification result.

[0037] Among them, when the acoustic model is performing speech recognition, it mainly compares the spatial distribution of acoustic feature parameters and judges whether the speech signal contains preset command words based on the degree of similarity. The reason why a speech signal that is not a preset command word is mistakenly recognized as a preset command word is mainly because the spatial distribution of acoustic feature parameters obtained after the speech signal is processed by the feature extraction module is similar to the spatial distribution of acoustic feature parameters corresponding to the preset command word, so it is mistakenly recognized as a preset command word. However, the similarity between non-preset command words and preset command words is often not too high, with poor stability and a certain degree of randomness. It is possible that it will not be recognized as a preset command word if it is said again.

[0038] Therefore, when the first voice signal is identified as a preset command word, the first voice signal is converted once or multiple times to change the spatial distribution of its acoustic feature parameters to produce a lifting, extension or compression effect. For voice signals that are not preset command words, the spatial distribution of acoustic feature parameters after lifting, extension or compression will increase the difference between it and the spatial distribution of acoustic feature parameters of preset command words, making it easier to be identified as a non-preset command word; and for voice signals that are preset command words, since the acoustic model is often obtained through training of preset command word voice signals with various speech speeds, various accents, various tones and various timbres, it includes various possible features of various preset command word voice signals, including the most basic fundamental frequency features. Therefore, even if the spatial distribution of acoustic feature parameters changes after one or more conversions, it can still be identified as a preset command word with a high probability. In this way, the error recognition rate of preset command words will be greatly reduced, and it will not affect the recognition rate of correct preset command words.

[0039] Among them, conversion processing mainly refers to conversion processing that can produce a certain change in the spatial distribution of acoustic feature parameters to produce an effect of lifting, extension or compression. The embodiment of the present invention does not limit the specific method of conversion processing. As long as the conversion processing method can cause a certain change in the spatial distribution of acoustic feature parameters to produce an effect of lifting, extension or compression, it can be adopted. If the effect is not ideal after one conversion, the implementer can also adjust the conversion amplitude by modifying the parameters, or perform multiple conversions, and even use different methods in each conversion to ultimately achieve the desired effect.

[0040] According to an implementation of an embodiment of the present invention, performing at least one conversion process on a first voice signal to obtain a second voice signal includes: performing a first conversion process on the first voice signal to obtain the second voice signal.

[0041] Generally speaking, if an appropriate method is used when converting the first voice signal, the misrecognition rate of the preset command words can be greatly reduced without converting the preset command words multiple times.

[0042] Therefore, in the embodiment of the present invention, the first speech signal is converted only once. However, it should be noted that the selection of the specific method of the first conversion process is very important. If it is selected properly, it can not only greatly reduce the error recognition rate, but also effectively save computing resources; on the contrary, if it is selected improperly, it cannot achieve the expectation of reducing the error recognition rate, and it may be necessary to perform multiple conversions, which consumes more computing resources and slows down the processing.

[0043] Therefore, on the one hand, it is recommended to use a conversion processing method that has a larger change range of the spatial distribution of acoustic feature parameters obtained after the conversion processing when performing the first processing conversion; on the other hand, it is recommended to use more training data with various acoustic features and the recognition result of "is a preset command word" to train the acoustic model so that the acoustic model can maintain the robustness of the recognition of the preset command word. In this way, a smaller error recognition rate can be achieved.

[0044] According to an implementation of an embodiment of the present invention, the first conversion process includes a speed change process.

[0045] Speed-changing processing of speech signals, for example, speeding up or slowing down the speech rate, can extend or compress the spatial distribution of acoustic feature parameters, thereby increasing the difference in the spatial distribution of acoustic feature parameters corresponding to preset command words, making it easier to identify as an expectation of non-preset command words.

[0046] In addition, speed change processing usually only changes the speaking speed but not the fundamental frequency value, and most acoustic models include the fundamental frequency features of preset command words. This means that the correct preset command words can still maintain the fundamental frequency features that can be recognized by the acoustic model after speed change processing, and can still be recognized as preset command words, which will not affect the recognition rate of the correct preset command words.

[0047] Commonly used speed change processing methods include time domain method, frequency domain method, and parametric method, such as SOLA, LSEE-MSTFTM, phase vocoder, etc. In the implementation of the present invention, the specific speed change processing method and implementation details are not limited, and the implementer can independently determine according to the implementation conditions and implementation effects.

[0048] According to an implementation of an embodiment of the present invention, the first conversion process includes a timbre conversion process.

[0049] Voice conversion itself is a technology that changes the characteristics of a speaker's voice. When performing voice conversion, the source features can be mapped to the target features through pre-established rules, such as Gauss mixture model, linear multivariate regression, and StarGAN. Therefore, voice conversion processing of speech signals can change the spatial distribution of acoustic feature parameters, thereby increasing the difference between the spatial distribution of acoustic feature parameters corresponding to preset command words, making it easier to identify as a non-preset command word.

[0050] However, it should be noted that the voice-color conversion will also change the acoustic characteristics of the correct preset command words. Therefore, when using this implementation, it is often necessary to use more correct preset command word speech data with different voices to fully train the acoustic model so that the acoustic model continues to maintain the robustness of preset command word recognition.

[0051] According to an implementation of an embodiment of the present invention, a first voice signal is converted at least once to obtain a second voice signal, including: performing a first conversion on the first voice signal to obtain a third voice signal; detecting whether the third voice signal contains a preset command word, and if so, performing a second conversion on the third voice signal to obtain a second voice signal.

[0052] As mentioned above, when the error recognition rate is still not satisfactory after one conversion process using the first conversion method or when the error recognition rate requirement is relatively strict, multiple conversion processes may be considered.

[0053] In the embodiment of the present invention, after the first conversion process, if it is still recognized as a preset command word, the second conversion process is further performed. The possibility that the non-preset command word speech is still mistakenly recognized as a preset command word after two conversion processes is smaller, so the misrecognition rate will be lower.

[0054] According to an implementation of an embodiment of the present invention, the first conversion process includes a speed change process, and the second conversion process includes a timbre conversion process.

[0055] When the conversion principles of the first conversion process and the second conversion process are different and the effects are not offset, the effect of reducing the false recognition rate will be more obvious. And taking the conversion process with a more obvious effect of reducing the false recognition rate as the first conversion process can further save computing resources and shorten processing time.

[0056] In this embodiment, the speed change process is used as the first conversion process, and the tone conversion process is used as the second process, so as to further reduce the error recognition rate without consuming too much computing resources.

[0057] According to an implementation manner of an embodiment of the present invention, when detecting whether the first voice signal includes a preset command word and further detecting whether the second voice signal is a wake-up command, the detection method adopted is the same.

[0058] As mentioned above, the detection method used when detecting whether the first voice signal contains a preset command word and the detection method used when further detecting whether the second voice signal is a wake-up command may be different or the same.

[0059] In this embodiment, when detecting whether the first voice signal contains a preset command word and further detecting whether the second voice signal is a wake-up command, the detection method used is the same, so that the voice detection module can be reused, and when the detection methods are the same, the interference caused by different detection methods can be eliminated, further highlighting the differences caused by changes in acoustic features, thereby reducing the false recognition rate. The effect is better.

[0060] According to an implementation manner of an embodiment of the present invention, the detection methods are all executed on the client.

[0061] The client here mainly refers to the voice interaction device, as opposed to the cloud server that performs semantic analysis, natural language understanding, and determines the response sentence.

[0062] As mentioned above, the detection of whether the first voice signal contains a preset command word and the further detection of whether the second voice signal is a wake-up command can be performed both in the client, or in the client and the server respectively.

[0063] In this embodiment, the detection of whether the first voice signal contains a preset command word and the further detection of whether the second voice signal is a wake-up command can both be performed in the client. Because the speech recognition method of the embodiment of the present invention has no special requirements on the detection accuracy of the second voice signal, and does not need to be higher than the detection accuracy of the first voice signal, therefore, a detection method with detection accuracy suitable for execution in the client can be applied.

[0064] The detection of whether the first voice signal contains a preset command word and the further detection of whether the second voice signal is a wake-up command can both be performed in the client, which reduces the network transmission process between the server and the response speed is faster.

[0065] Figure 2 The specific implementation flow diagram of the speech recognition method of the first embodiment of the present invention is shown. In the application, the speech recognition method of the embodiment of the present invention recognizes the wake-up command word, and the two speech feature detections use the same wake-up command word detection method, and the wake-up command word detection method is executed on the client of the speech recognition.

[0066] Specifically, the specific implementation process of the application speech recognition method includes:

[0067] Step 2010, receiving a voice signal;

[0068] Step 2020, periodically buffering the voice signal;

[0069] Step 2030, extracting features from the speech signal and buffering the extracted features;

[0070] Step 2040, determining whether the amount of buffered data reaches a certain amount of data, if so, proceeding to step 2050, if not, continuing back to step 2030;

[0071] Step 2050, executing a detection algorithm on the buffered features to determine whether the voice signal contains a wake-up command word, if so, proceeding to step 2060, if not, proceeding to step 2070 without waking up the dialogue program;

[0072] Step 2060, performing speed change processing on the buffered features;

[0073] Step 2080, executing a detection algorithm on the features after speed change processing to determine whether the voice signal contains a wake-up command word, if so, proceeding to step 2090, if not, proceeding to step 2070 without waking up the dialogue program;

[0074] Step 2090, wake up the dialogue program.

[0075] In order to further verify the degree of reduction of the false awakening rate by the speech recognition method of the embodiment of the present invention, the inventors used the above application to respectively execute the speech recognition method without secondary detection and the speech recognition method of the embodiment of the present invention to recognize the wake-up command words in the same execution environment, and respectively performed statistics on the wake-up rate and false awakening rate generated by executing the two methods, and the results are shown in Table 1:

[0076]

[0077] Table 1

[0078] Among them, the positive test set is a test set whose speech signal contains wake-up command words; and the negative test set is a test set whose speech signal does not contain wake-up command words; serial number 1 is the statistical result of the speech recognition method without secondary detection; serial number 2 is the statistical result of the speech recognition method of an embodiment of the present invention.

[0079] It can be seen that, compared with the speech recognition method without secondary detection, the wake-up rate of the speech recognition method according to the embodiment of the present invention is not affected, while the false wake-up rate is reduced by about 82%.

[0080] It should be noted that the specific application of the above-mentioned embodiments of the present invention is only an illustrative description, and is not a limitation on the implementation methods and application scenarios of the embodiments of the present invention. Implementers can adopt any applicable implementation methods and apply them to any applicable application scenarios according to the implementation conditions.

[0081] Furthermore, an embodiment of the present invention also provides a speech recognition device, such as Figure 3 As shown, the device 30 includes: a voice receiving module 301, used to receive a first voice signal; a first preset command word detection module 302, used to detect whether the first voice signal contains a preset command word; a voice signal conversion module 303, used to convert the first voice signal at least once to obtain a second voice signal; a second preset command word detection module 304, used to further detect whether the second voice signal contains a preset command word; an operation execution module 305, used to execute an operation corresponding to the preset command word.

[0082] According to an implementation of an embodiment of the present invention, the speech signal conversion module 303 is specifically configured to perform a first conversion process on the first speech signal to obtain a second speech signal.

[0083] According to an embodiment of the present invention, the speech signal conversion module 303 includes: a first conversion processing submodule, used to perform a first conversion processing on the first speech signal to obtain a third speech signal; and a second conversion processing submodule, used to perform a second conversion processing on the third speech signal to obtain a second speech signal.

[0084] According to an implementation of an embodiment of the present invention, the first preset command word detection module 302 and the second preset command word detection module 304 use the same detection method.

[0085] According to an implementation of an embodiment of the present invention, the first preset command word detection module 302 and the second preset command word detection module 304 are both arranged in the client.

[0086] According to a third aspect of an embodiment of the present invention, a computer storage medium stores program instructions on the storage medium, wherein the program instructions are used to execute any of the above-mentioned speech recognition methods when running.

[0087] It should be pointed out here that the above description of the speech recognition configuration embodiment and the above description of the computer storage medium embodiment are similar to the description of the aforementioned method embodiment, and have similar beneficial effects as the aforementioned method embodiment, so they are not described in detail. For the technical details that have not been disclosed in the description of the speech recognition configuration embodiment and the description of the computer storage medium embodiment in the embodiment of the present invention, please refer to the description of the aforementioned method embodiment of the present invention for understanding, and in order to save space, they are not described in detail.

[0088] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0089] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another device, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0090] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0091] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0092] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage medium, read-only memory (ROM), disk or optical disk, etc. Various media that can store program codes.

[0093] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage media, ROM, magnetic disks or optical disks.

[0094] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A speech recognition method, the method comprising: receiving a first voice signal; Detecting whether the first voice signal contains a preset command word, and if so, performing at least one conversion process on the first voice signal to obtain a second voice signal, and further detecting whether the second voice signal contains the preset command word, and if so, performing an operation corresponding to the preset command word, wherein the conversion process is used to reduce the similarity between a voice signal other than the preset command word and the preset command word; The performing at least one conversion process on the first voice signal to obtain the second voice signal comprises: performing a first conversion process on the first voice signal to obtain the second voice signal; The first conversion process includes at least one of the following: speed change process and timbre conversion process.

2. The method according to claim 1, wherein the converting process of the first speech signal at least once to obtain the second speech signal comprises: Performing a first conversion process on the first speech signal to obtain a third speech signal; It is detected whether the third voice signal contains the preset command word. If so, a second conversion process is performed on the third voice signal to obtain a second voice signal.

3. The method according to claim 2, wherein the first conversion process comprises a speed change process, and the second conversion process comprises a timbre conversion process.

4. According to the method of claim 1, the detection method used when detecting whether the first voice signal includes a preset command word and the detection method used when further detecting whether the second voice signal includes the preset command word are the same.

5. According to the method of claim 4, the detection methods are all executed on the client.

6. A speech recognition device, comprising: A voice receiving module, used for receiving a first voice signal; A first preset command word detection module, used to detect whether the first voice signal contains a preset command word; A voice signal conversion module, configured to perform at least one conversion process on the first voice signal to obtain a second voice signal, wherein the conversion process is configured to reduce the similarity between a voice signal other than the preset command word and the preset command word; A second preset command word detection module, used to further detect whether the second voice signal contains the preset command word; An operation execution module, used for executing the operation corresponding to the preset command word; The voice signal conversion module is specifically used to perform a first conversion process on the first voice signal to obtain a second voice signal; The first conversion process includes at least one of the following: speed change process and timbre conversion process.

7. A computer storage medium having program instructions stored thereon, wherein: The program instructions are used to execute the speech recognition method according to any one of claims 1 to 5 when running.

Images