Speech recognition method and device, equipment, storage medium and computer program product
By using the preset neural network model in the speech recognition system to de-noise, classify and filter the original speech signal, the problem of low speech recognition accuracy in high wind noise environments during outdoor sports is solved and higher speech recognition accuracy is achieved. .
Patent Information
- Application Number
- CN202411532281.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-30
- Publication Date
- 2025-01-24
AI Technical Summary
The accuracy of speech recognition in high-wind noise environments in outdoor sports is low, and the prior art is difficult to effectively reduce wind noise interference.
The preset neural network model is used for noise reduction, classification and screening of the original voice signal, obtaining effective voice characteristics and identification. The neural network model includes noise reduction modules, classification modules, and processing modules. By training models based on windless and pure wind noise sample features, the accuracy of voice recognition is improved.
In the high wind noise environment of outdoor sports, the accuracy of speech recognition is significantly improved, effectively reducing wind noise interference and improving the performance of the speech recognition system.
Smart Images
Figure CN119360840A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of speech recognition, and in particular to a speech recognition method, apparatus, device, storage medium and computer program product. Background Art
[0002] In outdoor sports, especially in activities such as cycling and running, wind noise becomes a significant problem that affects the accuracy of speech recognition. Existing technologies usually use bone voiceprint signal collection to reduce noise interference, but during exercise, changes in wind speed and noise generated by air flow will be transmitted to the bone voiceprint sensor through the bones and mixed with useful voice signals, thereby greatly reducing the accuracy of the speech recognition system. Therefore, how to effectively improve the accuracy of speech recognition in the high wind noise environment of outdoor sports has become an important challenge facing current technology. Summary of the invention
[0003] The main purpose of the present application is to provide a speech recognition method, apparatus, device, storage medium and computer program product, aiming to solve the technical problem of how to improve the accuracy of speech recognition in a high wind noise environment during outdoor sports.
[0004] To achieve the above object, the present application provides a speech recognition method, the method comprising:
[0005] Obtaining the original speech signal;
[0006] Recognize the original speech signal by using a preset neural network model;
[0007] The step of recognizing the original speech signal by using a preset neural network model comprises:
[0008] De-noising the original speech signal to obtain speech features after de-noising;
[0009] Classifying and processing the original speech signal to obtain the effective time point of the original speech signal;
[0010] Screening the noise-reduced speech features according to the effective time points to obtain effective speech features;
[0011] The effective speech features are identified.
[0012] In one embodiment, the preset neural network model at least includes:
[0013] A noise reduction module, used to reduce noise on the original speech signal to obtain the speech features after noise reduction;
[0014] A classification module, used to classify the original speech signal to obtain the effective time point of the original speech signal;
[0015] The processing module is used to screen the noise-reduced speech features according to the effective time points, obtain effective speech features, and perform speech recognition on the effective speech features.
[0016] In one embodiment, the preset neural network model preprocesses the original speech signal to obtain high-level speech features of the original speech signal;
[0017] The preset neural network model performs noise reduction on the high-level speech features to obtain the noise-reduced speech features;
[0018] The preset neural network model classifies the high-level speech features to obtain the effective time points of the original speech signal.
[0019] In one embodiment, the preset neural network model includes:
[0020] A preprocessing module, used for preprocessing the original speech signal to obtain high-level speech features of the original speech signal;
[0021] The preprocessing module comprises:
[0022] An fbank feature extraction unit is used to extract features from the original speech signal to obtain fbank speech features;
[0023] The neural network feature extraction unit is used to extract the fbank speech features to obtain the high-level speech features of the original speech signal.
[0024] In one embodiment, the step of classifying the high-level speech features by the preset neural network model to obtain the effective time point of the original speech signal includes:
[0025] The preset neural network model classifies the high-level speech features, and obtains the effective speech starting point and the effective speech ending point corresponding to the high-level speech features as the effective time points of the original speech signal.
[0026] In one embodiment, the preset neural network model is obtained by training based on the characteristics of bone voiceprint speech samples without wind noise and the characteristics of pure wind noise samples.
[0027] In addition, to achieve the above-mentioned purpose, the present application also proposes a speech recognition device, the device comprising:
[0028] A voice acquisition module, used to acquire original voice signals;
[0029] A speech processing module, used for recognizing the original speech signal through a preset neural network model;
[0030] The speech processing module is also used to reduce noise on the original speech signal, obtain the speech features after noise reduction, classify the original speech signal, obtain the effective time point of the original speech signal, screen the speech features after noise reduction according to the effective time point, obtain the effective speech features, and identify the effective speech features.
[0031] In addition, to achieve the above-mentioned purpose, the present application also proposes a speech recognition device, wherein the wearable device includes: a memory, a processor, and a speech recognition program stored in the memory and executable on the processor, wherein the speech recognition program implements the steps of the speech recognition method described above when executed by the processor.
[0032] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the speech recognition method described above are implemented.
[0033] In addition, to achieve the above-mentioned purpose, the present application also proposes a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the speech recognition method described above are implemented.
[0034] The present application provides a speech recognition method, apparatus, device, storage medium and computer program product, the method comprising: obtaining an original speech signal, identifying the original speech signal through a preset neural network model, wherein the step of identifying the original speech signal through the preset neural network model comprises: denoising the original speech signal to obtain speech features after denoising; classifying the original speech signal to obtain the effective time point of the original speech signal; screening the speech features after denoising according to the effective time point to obtain effective speech features; identifying the effective speech features. The present application can input the obtained original speech signal into a preset neural network model to obtain speech features after denoising and effective time points, and screen the speech features after denoising according to the effective time points, and identify the screening results to improve the accuracy of speech recognition, thereby improving the accuracy of speech recognition in a high wind noise environment of outdoor sports. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0037] Figure 1 This is a flow chart of the first embodiment of the speech recognition method of the present application;
[0038] Figure 2 This is a schematic diagram of the structure of the preset neural network model in the speech recognition method of this application;
[0039] Figure 3 This is a flow chart of the second embodiment of the speech recognition method of the present application;
[0040] Figure 4 A diagram of a speech recognition device provided in an embodiment of the present application;
[0041] Figure 5 The figure is a schematic diagram of the structure of a speech recognition device suitable for implementing the embodiment of the present application.
[0042] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0043] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.
[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0045] It should be noted that all directional indications in the embodiments of the present application (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.
[0046] It should be noted that in outdoor sports, especially in activities such as cycling and running, wind noise becomes a significant problem affecting the accuracy of speech recognition. Existing technologies usually use bone voiceprint signal collection to reduce noise interference, but during exercise, changes in wind speed and noise generated by air flow will be transmitted to the bone voiceprint sensor through the bones and mixed with useful voice signals, thereby greatly reducing the accuracy of the speech recognition system. Therefore, how to effectively improve the accuracy of speech recognition in the high wind noise environment of outdoor sports has become an important challenge facing current technology.
[0047] Therefore, in order to solve the above challenges, this embodiment provides a speech recognition method, which includes: obtaining an original speech signal; processing the original speech signal through a preset neural network model to obtain effective speech features, and performing speech recognition on the effective speech features, wherein the preset neural network model at least includes: a noise reduction module, a classification module and a processing module, the processing module is connected to the noise reduction module and the classification module respectively, the noise reduction module is used to reduce the noise of the original speech signal to obtain pure speech features; the classification module is used to classify and process the original speech signal to obtain effective time points; the processing module is used to screen the pure speech features according to the effective time points to obtain effective speech features, and perform speech recognition on the effective speech features. This shows that this embodiment is also provided with a noise reduction module, a classification module and a processing module in the preset neural network model, and the original speech signal is denoised by using the noise reduction module to obtain pure speech features, and the original speech signal is classified and processed by the classification module to obtain effective time points, and finally the processing module is used to screen the pure speech features according to the effective time points to obtain effective speech features, and speech recognition is performed on the effective speech features. In this way, the accuracy of speech recognition in the high wind noise environment of outdoor sports is improved.
[0048] For ease of understanding, the following combination Figures 1 to 5 The speech recognition method provided in the embodiment of the present application is specifically introduced.
[0049] The present application embodiment provides a speech recognition method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the speech recognition method of the present application.
[0050] In this embodiment, the speech recognition method includes the following steps:
[0051] Step S10: Acquire the original speech signal.
[0052] It is understandable that the method of this embodiment can be applied to a scenario where a wearable device performs voice recognition, wherein the wearable device can be sports glasses or sports headphones with voice recognition function, etc., and of course, it can also be a scenario where other devices perform voice recognition, and this embodiment does not limit this. The executor of this embodiment can be a voice recognition device with noise reduction, classification and voice recognition, such as sports glasses, or other devices that can achieve the same or similar functions. The following embodiment and the following embodiments are specifically described using the above-mentioned sports glasses (hereinafter referred to as the device).
[0053] It should be understood that the original voice signal may be a voice signal collected by the above device, and the signal may be a bone voiceprint signal, etc., which is not specifically limited in this embodiment. In a specific implementation, the above device collects the bone voiceprint signal emitted by the user through a sound receiving module such as a bone voiceprint sensor to obtain the original voice signal.
[0054] Step S20: recognizing the original speech signal through a preset neural network model.
[0055] In some implementations, the step of recognizing an initial speech signal using a preset neural network model includes:
[0056] De-noise the original speech signal to obtain the speech features after de-noising. Classify the original speech signal to obtain the effective time points of the original speech signal. Filter the speech features after de-noising according to the effective time points to obtain effective speech features. Identify effective speech features.
[0057] It should be emphasized that the above-mentioned preset neural network model can be a model for processing the original voice signal to obtain the voice recognition result. However, since the traditional model generally does not pay attention to the noise information, the above-mentioned preset neural network model in this embodiment at least includes: a noise reduction module, a classification module and a processing module, and the processing module is connected to the noise reduction module and the classification module respectively. Among them, the noise reduction module is used to reduce the noise of the original voice signal and obtain the voice features after noise reduction. The classification module is used to classify the original voice signal and obtain the effective time point of the original voice signal. The processing module is used to screen the voice features after noise reduction according to the effective time point, obtain the effective voice features, and perform voice recognition on the effective voice features.
[0058] In addition, it should be understood that the above-mentioned preset neural network model is obtained by training based on the characteristics of bone voiceprint speech samples without wind noise and the characteristics of pure wind noise samples, that is, before using the preset neural network model, the above-mentioned preset neural network model needs to be trained. Furthermore, before the step of obtaining the original voice signal, it also includes:
[0059] Obtain pure sample signals and noisy sample signals;
[0060] The pure sample features are mixed with the noise sample features, and the initial neural network model is trained through the mixed results to obtain a preset neural network model.
[0061] It should be noted that the above-mentioned pure sample signal can be a voice signal collected in a quiet environment (for example, a bone voiceprint signal collected separately in a quiet environment), the above-mentioned noise sample signal can be a wind noise signal collected by the user in a moving state (for example, a wind noise signal during riding), and the above-mentioned initial neural network model can be an untrained neural network (for example, a long short-term memory (LSTM) model. In a specific implementation, the user can train the above-mentioned neural network model by inputting a preset sample signal through a set loss function to obtain a preset neural network model.
[0062] In one example, the user can separately collect the bone voiceprint signal in a quiet environment and the wind noise signal during riding. During the training process, the bone voiceprint signal with noise is obtained by mixing the bone voiceprint signal in a quiet environment and the wind noise signal during riding. The loss function of the training is: (1) the mean square error (MSE) difference between the fbank feature of the bone voiceprint in a quiet environment and the fbank feature after noise reduction; (2) the VAD binary classification loss function corresponding to each frame of fbank. The deep neural network is trained through the loss function to obtain a preset neural network model that can be deployed to the above-mentioned device.
[0063] For ease of understanding, refer to Figure 2 To explain, Figure 2 Schematic diagram of the structure of the preset neural network model in the speech recognition method of the present application. In a specific implementation, the above-mentioned device uses a noise reduction module (such as Figure 2 The noise reduction module in the original speech signal is used to reduce the noise of the original speech signal to obtain pure speech features, and the classification module (i.e. Figure 2 The classification module in the classification module classifies the original speech signal to obtain the effective time point, and finally uses the processing module to filter the pure speech features according to the effective time point to obtain the effective speech features, and performs speech recognition on the effective speech features. In this way, the accuracy of speech recognition in the high wind noise environment of outdoor sports is improved.
[0064] It should be understood that when performing recognition and other processing on a speech signal, the Mel-frequency cepstral coefficient (fbank) feature of the speech signal can generally be obtained through feature extraction, and the recognition result of the speech signal can be obtained by processing the fbank feature.
[0065] Therefore, in this embodiment, the preset neural network model further includes: a preprocessing module, which is connected to the noise reduction module and the classification module respectively. The preprocessing module is used to preprocess the original speech signal to obtain high-level speech features of the original speech signal.
[0066] It should be understood that the above-mentioned preprocessing module (i.e. Figure 2 The preprocessing module in the example may be a module for extracting features from the original speech signal to obtain speech features, and the high-level speech features may be speech features (such as fbank features) obtained by extracting features from the original speech.
[0067] The above-mentioned preset neural network model includes: a noise reduction module, which is used to reduce the noise of the original speech signal and obtain the speech features after noise reduction.
[0068] It can be understood that the above-mentioned noise reduction module can be a recurrent neural network module obtained through training (for example, in the above example, the MSE difference between the fbank feature of the bone voiceprint in a quiet environment and the fbank feature after noise reduction is used as the loss function, and the above-mentioned bone voiceprint signal with noise is used for training), and the above-mentioned pure speech feature can be a speech feature obtained by inputting the above-mentioned high-level speech feature into the above-mentioned trained noise reduction module, wherein the noise reduction can be an overall noise reduction of the above-mentioned high-level speech feature (for example, the speech feature is input into a filter to filter and obtain a filtered speech feature after filtering a certain frequency). In a specific implementation, after the above-mentioned device uses the above-mentioned preprocessing module to preprocess the original speech signal to obtain the high-level speech feature, it uses the above-mentioned noise reduction module to perform noise reduction on the high-level speech feature to obtain a pure speech feature. It can be seen that the above-mentioned preset neural network model can perform noise reduction on the high-level speech feature to obtain the noise-reduced speech feature.
[0069] Since the above-mentioned noise reduction module performs noise reduction on the above-mentioned high-level speech features as a whole, if pure noise, that is, a signal without a speech signal, appears in the above-mentioned original speech signal, noise reduction of the high-level speech features corresponding to the pure noise signal still cannot obtain accurate speech recognition content through speech recognition of the pure speech feature fragment corresponding to the pure noise signal. Therefore, in this embodiment, the above-mentioned device can also classify the high-level speech features through the classification module to obtain the effective time points corresponding to the high-level speech features.
[0070] It should be noted that the above-mentioned classification module can be a module composed of a binary classification model obtained by training the above-mentioned pure sample features and the above-mentioned noise sample features (for example, in the above-mentioned example, the VAD binary classification corresponding to each frame of fbank is used as the loss function, and the above-mentioned bone voiceprint signal with noise is used for training), and the above-mentioned effective time point can be the time point where the effective speech exists, which is marked by the above-mentioned trained classification module according to the input high-level speech feature. In a specific implementation, after the above-mentioned device uses the above-mentioned preprocessing module to preprocess the original speech signal to obtain the high-level speech feature, it uses the above-mentioned classification module to classify the high-level speech feature and obtains the effective time point corresponding to the above-mentioned high-level speech feature. It can be seen that the above-mentioned preset neural network model can classify the high-level speech features and obtain the effective time point of the original speech signal.
[0071] The above-mentioned preset neural network model includes: a processing module, which is used to screen the noise-reduced speech features according to the effective time points, obtain the effective speech features, and perform speech recognition on the effective speech features.
[0072] It should be understood that the above processing module (i.e. Figure 2 The processing module in the above may be a module that screens the above clean module according to the above effective time point to obtain effective speech features and performs speech recognition on the effective speech features. In a specific implementation, when the above device obtains the clean speech features corresponding to the above high-level speech features and the effective time points corresponding to the above high-level speech features, the above device uses the above processing module to segment the above clean speech according to the above effective time points to select effective speech features, and performs speech recognition on the selected effective speech to obtain a speech recognition result.
[0073] Furthermore, in order to facilitate the screening of effective speech features through the above effective time points, the high-level speech features are classified and processed by the classification module to obtain the effective time points corresponding to the high-level speech features, including:
[0074] The high-level speech features are classified by a classification module to obtain the effective speech starting point and the effective speech ending point corresponding to the high-level speech features;
[0075] The step of using the processing module to screen the pure speech features according to the effective time points to obtain the effective speech features includes:
[0076] The processing module is used to take the pure speech features between the effective speech starting point and the effective speech ending point as the effective speech features.
[0077] It should be noted that the above-mentioned effective speech starting point can be the starting time point of the speech segment with effective speech signals such as bone voiceprint obtained by the above-mentioned classification module when the above-mentioned high-level speech module is binary classified, and the above-mentioned effective speech signal can refer to a speech signal that is not a pure noise signal. The above-mentioned effective speech ending point can be the starting time point of the speech segment with effective speech signals such as bone voiceprint obtained by the above-mentioned classification module when the above-mentioned high-level speech module is binary classified.
[0078] In a specific implementation, since the noise reduction module does not change the time series of the high-level speech features when performing noise reduction on the high-level speech features, the high-level speech features are consistent with the pure speech features in time. Therefore, after the device obtains the effective speech starting point and effective speech ending point of the high-level speech features using the classification module, the processing module can segment the pure speech features in the time series according to the effective speech starting point and the effective speech ending point and filter out the effective speech features. And the device can simultaneously input the high-level speech features into the classification module and the noise reduction module, so that the device can quickly respond to the speech recognition method triggered by the user to obtain accurate speech recognition results.
[0079] The present embodiment discloses a method for speech recognition, which includes: obtaining an original speech signal, and identifying the original speech signal through a preset neural network model, wherein the step of identifying the original speech signal through the preset neural network model includes: denoising the original speech signal to obtain speech features after denoising; classifying the original speech signal to obtain the effective time point of the original speech signal; screening the speech features after denoising according to the effective time point to obtain effective speech features; and identifying effective speech features. This shows that the present embodiment can input the obtained original speech signal into the preset neural network model to obtain the speech features after denoising and the effective time point, and screen the speech features after denoising according to the effective time point, and identify the screening results to improve the accuracy of speech recognition, thereby improving the accuracy of speech recognition in the high wind noise environment of outdoor sports.
[0080] refer to Figure 3 , Figure 3 This is a flow chart of the second embodiment of the speech recognition method of the present application.
[0081] In order to further preprocess the above speech signal to obtain high-level speech features, so as to ensure more accurate speech recognition results, such as Figure 3 As shown, in this embodiment, the preprocessing module includes: an fbank feature extraction unit and a neural network feature extraction unit, the fbank feature extraction unit is connected to the neural network feature extraction unit, and the neural network feature extraction unit is respectively connected to the denoising module and the classification module;
[0082] The step of preprocessing the original speech signal through the preprocessing module to obtain high-level speech features of the original speech signal includes:
[0083] Step S21: extracting features from the original speech signal through the fbank feature extraction unit to obtain fbank speech features.
[0084] It should be noted that the fbank feature extraction unit may be a unit that extracts features from a speech signal through feature extraction processing (such as pre-emphasis, framing, windowing, fast Fourier transform, filter bank, logarithmic operation, and discrete cosine transform, etc.) to obtain fbank features, and the fbank speech features may be speech features obtained through feature extraction. In a specific implementation, the device extracts fbank speech features from the original speech signal through feature extraction processing (such as pre-emphasis, framing, windowing, fast Fourier transform, filter bank, logarithmic operation, and discrete cosine transform, etc.) through the feature extraction subunit.
[0085] Step S22: extracting the fbank speech features using the neural network feature extraction unit to obtain high-level speech features of the original speech signal.
[0086] It should be understood that the neural network feature extraction unit can be a unit composed of neural network (such as LSTM) subunits for performing high-level feature extraction on speech features. In a specific implementation, after the device extracts features from the original speech signal through the fbank feature extraction unit to obtain fbank speech features, it uses the neural network feature extraction unit to perform high-level feature extraction on the fbank speech features to obtain high-level speech features.
[0087] It can be obtained that the preprocessing module includes:
[0088] The fbank feature extraction unit is used to extract features from the original speech signal to obtain fbank speech features;
[0089] The neural network feature extraction unit is used to extract the fbank speech features and obtain the high-level speech features of the original speech signal.
[0090] This embodiment also provides a voice recognition device, please refer to Figure 4 , Figure 4 A diagram of a speech recognition device provided in an embodiment of the present application, wherein the speech recognition device comprises:
[0091] A voice acquisition module, used to acquire original voice signals;
[0092] A speech processing module, used for recognizing original speech signals through a preset neural network model;
[0093] The speech processing module is also used to reduce the noise of the original speech signal, obtain the speech features after noise reduction, classify the original speech signal, obtain the effective time point of the original speech signal, screen the speech features after noise reduction according to the effective time point, obtain the effective speech features, and identify the effective speech features.
[0094] The preset neural network model further includes: a preprocessing module, which is connected to the noise reduction module and the classification module respectively;
[0095] The speech processing module is further used to pre-process the original speech signal through the pre-processing module to obtain high-level speech features of the original speech signal;
[0096] The speech processing module is also used to reduce the noise of the high-level speech features by using the noise reduction module to obtain the pure speech features corresponding to the high-level speech features;
[0097] The speech processing module is also used to classify the high-level speech features through the classification module to obtain the effective time points corresponding to the high-level speech features;
[0098] The speech processing module is also used to use the processing module to screen the pure speech features according to the effective time point, obtain the effective speech features, and perform speech recognition on the effective speech features.
[0099] The speech recognition device provided in this embodiment adopts the speech recognition method in the above embodiment, which can solve the technical problem of how to improve the accuracy of speech recognition in a high wind noise environment during outdoor sports. Compared with the prior art, the beneficial effects of the speech recognition device provided in this embodiment are the same as the beneficial effects of the speech recognition method provided in the above embodiment, and other technical features in the speech recognition device are the same as the features disclosed in the above embodiment method, which will not be repeated here.
[0100] This embodiment provides a speech recognition device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the speech recognition method in the above-mentioned embodiment 1.
[0101] Reference below Figure 5 , Figure 5Schematic diagram of the structure of a speech recognition device suitable for implementing the embodiment of the present application. The speech recognition device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The voice recognition device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0102] like Figure 5 As shown, the speech recognition device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the speech recognition device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a bone voiceprint sensor, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the speech recognition device to communicate wirelessly or wired with other devices to exchange data. Although the speech recognition device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.
[0103] In particular, according to the present embodiment, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present embodiment includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present embodiment are executed.
[0104] The speech recognition device provided in this embodiment adopts the speech recognition method in the above embodiment, which can solve the technical problem of how to improve the accuracy of speech recognition in a high wind noise environment during outdoor sports. Compared with the prior art, the beneficial effects of the speech recognition device provided in this embodiment are the same as the beneficial effects of the speech recognition method provided in the above embodiment, and other technical features in the speech recognition device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.
[0105] It should be understood that the various parts disclosed in this embodiment can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0106] The above is only a specific implementation of this embodiment, but the protection scope of this embodiment is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in this embodiment, which should be included in the protection scope of this embodiment. Therefore, the protection scope of this embodiment should be based on the protection scope of the claims.
[0107] This embodiment provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the speech recognition method in the above embodiment.
[0108] The computer-readable storage medium provided in this embodiment may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: ReadOnlyMemory), an erasable programmable read-only memory (EPROM: Erasable ProgrammableReadOnlyMemory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-ReadOnlyMemory), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination thereof.
[0109] The computer-readable storage medium may be included in the speech recognition device; or may exist independently without being assembled into the speech recognition device.
[0110] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the speech recognition device, the speech recognition device performs speech recognition.
[0111] The computer program code for performing the operations of the present embodiment may be written in one or more programming languages or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0112] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present embodiment. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0113] The modules involved in the description of this embodiment may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.
[0114] The readable storage medium provided in this embodiment is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned speech recognition method, and can solve the technical problem of how to improve the accuracy of speech recognition in a high wind noise environment of outdoor sports. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this embodiment are the same as the beneficial effects of the speech recognition method provided in the above-mentioned embodiment, and will not be repeated here.
[0115] This embodiment also provides a computer program product, including a computer program, which implements the steps of the above-mentioned speech recognition method when executed by a processor.
[0116] The computer program product provided in this embodiment can solve the technical problem of how to improve the accuracy of speech recognition in a high wind noise environment during outdoor sports. Compared with the prior art, the beneficial effects of the computer program product provided in this embodiment are the same as the beneficial effects of the speech recognition method provided in the above embodiment, which will not be described in detail here.
[0117] The above are only some embodiments, and the patent scope of this embodiment is not limited thereto. All equivalent structural changes made by using the contents of the description and drawings of this application under the technical concept of this application, or directly / indirectly applied in other related technical fields are included in the patent protection scope of this application.
Claims
1. A speech recognition method, characterized in that: The method comprises: Obtaining the original speech signal; Recognize the original speech signal by using a preset neural network model; The step of recognizing the original speech signal by using a preset neural network model comprises: De-noising the original speech signal to obtain speech features after de-noising; Classifying and processing the original speech signal to obtain the effective time point of the original speech signal; Screening the noise-reduced speech features according to the effective time points to obtain effective speech features; The effective speech features are identified.
2. The method according to claim 1, characterized in that The preset neural network model at least includes: A noise reduction module, used to reduce noise on the original speech signal to obtain the speech features after noise reduction; A classification module, used to classify the original speech signal to obtain the effective time point of the original speech signal; The processing module is used to screen the noise-reduced speech features according to the effective time points, obtain effective speech features, and perform speech recognition on the effective speech features.
3. The method according to claim 1, characterized in that Also includes: The preset neural network model pre-processes the original speech signal to obtain high-level speech features of the original speech signal; The preset neural network model performs noise reduction on the high-level speech features to obtain the noise-reduced speech features; The preset neural network model classifies the high-level speech features to obtain the effective time points of the original speech signal.
4. The method according to claim 3, characterized in that The preset neural network model includes: A preprocessing module, used for preprocessing the original speech signal to obtain high-level speech features of the original speech signal; The preprocessing module comprises: An fbank feature extraction unit is used to extract features from the original speech signal to obtain fbank speech features; The neural network feature extraction unit is used to extract the fbank speech features to obtain the high-level speech features of the original speech signal.
5. The method according to claim 3, characterized in that The preset neural network model classifies the high-level speech features to obtain the effective time point of the original speech signal, including: The preset neural network model classifies the high-level speech features, and obtains the effective speech starting point and the effective speech ending point corresponding to the high-level speech features as the effective time points of the original speech signal.
6. The method according to claim 1, characterized in that The preset neural network model is obtained by training based on the bone voiceprint speech sample features without wind noise and the pure wind noise sample features.
7. A speech recognition device, characterized in that: The device comprises: A voice acquisition module, used to acquire original voice signals; A speech processing module, used for recognizing the original speech signal through a preset neural network model; The speech processing module is also used to reduce noise on the original speech signal, obtain the speech features after noise reduction, classify the original speech signal, obtain the effective time point of the original speech signal, screen the speech features after noise reduction according to the effective time point, obtain the effective speech features, and identify the effective speech features.
8. A speech recognition device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the speech recognition method according to any one of claims 1 to 6.
9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the speech recognition method according to any one of claims 1 to 6 are implemented.
10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the speech recognition method according to any one of claims 1 to 6 are implemented.