Speech recognition method and device, storage medium and electronic device

Abstract

The application provides a speech recognition method and device, a storage medium and an electronic device, and the method comprises the following steps: monitoring the noise state of the environment where the target device is located, wherein the noise state is used to represent the noise intensity of the environment; configuring a smoothing algorithm according to the noise state; performing smoothing processing on the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target speech signal; and identifying the target speech signal by using a speech recognition model. The technical problem of low speech recognition accuracy in the related art is solved, and the user experience is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of speech, in particular to a speech recognition method and device, a storage medium and an electronic device. BACKGROUND

[0002] In related technologies, with the rapid development of Internet of Things technology, more and more devices have voice interaction functions. However, the environment in which the Internet of Things devices are located is complex and diverse. For example, a smart speaker in a smart home may be subjected to cooking noise in the kitchen, television noise in the living room, etc.; sensor devices in the industrial Internet of Things may be subjected to the interference of the roar of machine operation. At the same time, Internet of Things devices often have limited computing resources and limited power consumption.

[0003] The speech recognition noise reduction technology in related technologies has problems such as low efficiency, poor adaptability, and inability to fully utilize device characteristics for optimization when dealing with noise problems of Internet of Things devices, resulting in low speech recognition accuracy and affecting user experience.

[0004] In view of the above problems in related technologies, no efficient and accurate solution has been found. SUMMARY

[0005] The present application provides a speech recognition method and device, a storage medium and an electronic device to solve the technical problem of low speech recognition accuracy in related technologies.

[0006] According to an embodiment of the present application, a speech recognition method is provided, comprising: monitoring the noise state of the environment in which the target device is located, wherein the noise state is used to represent the noise intensity of the environment; configuring a smoothing algorithm according to the noise state; performing smoothing processing on the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target speech signal; and identifying the target speech signal using a speech recognition model.

[0007] Optionally, monitoring the noise state of the environment in which the target device is located comprises: obtaining the environmental sound collected by the target device according to a preset period; calculating the average sound intensity of the environmental sound in the preset period; determining whether the average sound intensity is less than a preset threshold; if the average sound intensity is less than the preset threshold, determining that the noise state of the environment in which the target device is located is a quiet state; and if the average sound intensity is greater than or equal to the preset threshold, analyzing the frequency value of the environmental sound, and if the frequency value is distributed within a preset frequency range, determining that the noise state of the environment in which the target device is located is a noisy state.

[0008] Optionally, configuring the smoothing algorithm according to the noise state comprises: if the noise state is a quiet state, selecting a first smoothing algorithm of a first order, wherein the first order is greater than a preset order; configuring an interpolation weight parameter of the first smoothing algorithm based on the first order, and setting an adjustment step of the first smoothing algorithm, wherein the interpolation weight parameter is used to represent a weight coefficient of each order language model in the first smoothing algorithm, and the adjustment step is used to indicate a probability amplitude allowed to be adjusted by the first smoothing algorithm.

[0009] Optionally, configuring the smoothing algorithm according to the noise state comprises: if the noise state is a noisy state, selecting a second smoothing algorithm of a second order, wherein the second order is less than the preset order; configuring an anti-noise parameter and a smoothing factor of the second smoothing algorithm, wherein the anti-noise parameter is used to indicate the attenuation intensity of the speech signal of different frequency ranges, and the smoothing factor is used to improve the probability of a vocabulary combination whose occurrence frequency is lower than a preset value.

[0010] Optionally, the smoothing processing of the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target speech signal comprises: if the noise state is a quiet state, collecting a first sound signal; enhancing the amplitude of a high-frequency component in the first sound signal to obtain a first intermediate speech; performing frame processing on the first intermediate speech to obtain a second intermediate speech; performing windowing processing on each frame of the second intermediate speech by using a window function to obtain a third intermediate speech, and filtering invalid frames in the third intermediate speech by using an energy value and a zero-crossing rate; performing interpolation smoothing calculation on the third intermediate speech by using the smoothing algorithm to obtain a first target speech signal.

[0011] Optionally, the interpolation smoothing calculation on the third intermediate speech by using the smoothing algorithm to obtain a first target speech signal comprises: constructing a first language model of an n-order sliding window, wherein n is greater than 1; obtaining a statistical result based on the first language model and the occurrence frequency of each order vocabulary combination in a text corpus; determining a first vocabulary combination in the third intermediate speech that does not appear in the text corpus based on the statistical result; calculating the occurrence probability of each word in the first vocabulary combination in the n-order sliding window, and determining an initial probability of the first vocabulary combination; and performing weighted calculation on the final probability of the first vocabulary combination in the third intermediate speech by using the occurrence probability and the initial probability to obtain a first target speech signal.

[0012] Optionally, the smoothing the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target voice signal comprises: collecting a second sound signal if the noise state is a noisy state; converting the second sound signal into a frequency domain signal; analyzing a characteristic parameter of environmental sound in the frequency domain signal; filtering the environmental sound in the second sound signal based on the characteristic parameter to obtain a fourth intermediate voice; and performing anti-noise smoothing calculation on the fourth intermediate voice by using the smoothing algorithm to obtain a second target voice signal.

[0013] Optionally, the anti-noise smoothing calculation on the fourth intermediate voice by using the smoothing algorithm to obtain a second target voice signal comprises: performing attenuation processing on the voice signal in the noise frequency range of the fourth intermediate voice to obtain a fifth intermediate voice; constructing a second language model of an m-order sliding window, where m is greater than 1; obtaining a statistical result by counting the occurrence frequency of each-order word group in a text corpus based on the second language model; screening a second word group with an occurrence frequency lower than a preset frequency in the fifth intermediate voice based on the statistical result; and improving the occurrence probability of the second word group in the fifth intermediate voice to obtain a second target voice signal.

[0014] According to another embodiment of the present application, a voice recognition device is provided, comprising: a monitoring module configured to monitor a noise state of an environment where a target device is located, wherein the noise state is used to represent the noise intensity of the environment; a configuration module configured to configure a smoothing algorithm according to the noise state; a smoothing module configured to smooth a sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target voice signal; and a recognition module configured to recognize the target voice signal by using a voice recognition model.

[0015] Optionally, the monitoring module comprises: an acquisition unit configured to acquire environmental sound collected by the target device at a preset period; a calculation unit configured to calculate the average sound intensity of the environmental sound in the preset period; a judgment unit configured to judge whether the average sound intensity is less than a preset threshold; a determination unit configured to determine that the noise state of the environment where the target device is located is a quiet state if the average sound intensity is less than the preset threshold, and analyze the frequency value of the environmental sound if the average sound intensity is greater than or equal to the preset threshold, and determine that the noise state of the environment where the target device is located is a noisy state if the frequency value is distributed within a preset frequency range.

[0016] Optionally, the configuration module comprises: a first selection unit configured to select a first smoothing algorithm of a first order if the noise state is a quiet state, wherein the first order is greater than a preset order; and a first configuration unit configured to configure an interpolation weight parameter of the first smoothing algorithm based on the first order, and set an adjustment step of the first smoothing algorithm, wherein the interpolation weight parameter is used to represent a weight coefficient of each order language model in the first smoothing algorithm, and the adjustment step is used to indicate a probability amplitude allowed to be adjusted by the first smoothing algorithm.

[0017] Optionally, the configuration module comprises: a second selection unit configured to select a second smoothing algorithm of a second order if the noise state is a noisy state, wherein the second order is less than the preset order; and a second configuration unit configured to configure an anti-noise parameter and a smoothing factor of the second smoothing algorithm, wherein the anti-noise parameter is used to indicate an attenuation intensity of a speech signal of different frequency ranges, and the smoothing factor is used to improve a probability of a vocabulary combination with an occurrence frequency lower than a preset value.

[0018] Optionally, the smoothing module comprises: a first acquisition unit configured to acquire a first sound signal if the noise state is a quiet state; an enhancement unit configured to enhance an amplitude of a high-frequency component in the first sound signal to obtain a first intermediate speech; a frame division unit configured to perform frame division processing on the first intermediate speech to obtain a second intermediate speech; a windowing unit configured to perform windowing processing on each frame in the second intermediate speech by using a window function to obtain a third intermediate speech, and filter invalid frames in the third intermediate speech by using an energy value and a zero-crossing rate; and a first smoothing unit configured to perform interpolation smoothing calculation on the third intermediate speech by using the smoothing algorithm to obtain a first target speech signal.

[0019] Optionally, the first smoothing unit comprises: a construction sub-unit configured to construct a first language model of an n-order sliding window, wherein n is greater than 1; a statistics sub-unit configured to obtain a statistics result by statistically counting occurrence frequencies of each order vocabulary combination in a text corpus based on the first language model; a determination sub-unit configured to determine a first vocabulary combination not appearing in the text corpus in the third intermediate speech based on the statistics result; a first calculation sub-unit configured to calculate occurrence probabilities of words in the first vocabulary combination in the n-order sliding window, respectively, and determine an initial probability of the first vocabulary combination; and a second calculation sub-unit configured to calculate a final probability of the first vocabulary combination in the third intermediate speech by using the occurrence probabilities and the initial probability to obtain the first target speech signal.

[0020] Optionally, the smoothing module comprises: a second acquisition unit, configured to acquire a second sound signal if the noise state is a noisy state; a conversion unit, configured to convert the second sound signal into a frequency domain signal; an analysis unit, configured to analyze a characteristic parameter of environmental sound in the frequency domain signal; a filtering unit, configured to filter the environmental sound in the second sound signal based on the characteristic parameter to obtain a fourth intermediate speech; and a second smoothing unit, configured to perform anti-noise smoothing calculation on the fourth intermediate speech using the smoothing algorithm to obtain a second target speech signal.

[0021] Optionally, the second smoothing unit comprises: an attenuation sub-unit, configured to perform attenuation processing on a speech signal in a noise frequency range in the fourth intermediate speech to obtain a fifth intermediate speech; a construction sub-unit, configured to construct a second language model of an m-order sliding window, where m is greater than 1; a statistics sub-unit, configured to count the occurrence frequency of each-order word group in a text corpus based on the second language model to obtain a statistics result; a screening sub-unit, configured to screen a second word group with an occurrence frequency lower than a preset frequency in the fifth intermediate speech based on the statistics result; and a promotion sub-unit, configured to promote the occurrence probability of the second word group in the fifth intermediate speech to obtain a second target speech signal.

[0022] According to another aspect of the embodiments of the present application, a storage medium is also provided, which comprises a stored program, and the program performs the steps described above when running.

[0023] According to another aspect of the embodiments of the present application, an electronic device is also provided, which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus; the memory is used to store a computer program; and the processor is used to execute the steps in the above method by running the program stored in the memory.

[0024] According to another aspect of the embodiments of the present application, a storage medium is also provided, which comprises a stored computer program, and the computer program is set to execute the steps in any of the above device embodiments when running.

[0025] By the embodiment of the present application, the noise state of the environment where the target device is located is monitored, wherein the noise state is used to represent the noise intensity of the environment; a smoothing algorithm is configured according to the noise state; a voice signal is obtained by smoothing the sound signal collected by the target device according to the noise state and the smoothing algorithm; and the voice recognition model is used to recognize the target voice signal. By monitoring the noise state of the environment where the target device is located, and adaptively configuring the smoothing algorithm based on the noise state, and then smoothing the sound signal according to the noise state and the configured smoothing algorithm, the voice recognition model in the subsequent processing can more fully utilize and recognize the voice feature information carried in the sound signal, thereby solving the technical problem of low voice recognition accuracy in the related art, and improving the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0026] The accompanying drawings, which are included to provide a further understanding of the present application, constitute a part of this application, illustrate the illustrative embodiments of the present application and the description thereof, and do not constitute improper limitations on the present application. In the drawings:

[0027] Figure 1 is a hardware structure block diagram of an intelligent sound box according to an embodiment of the present application;

[0028] Figure 2 is a flowchart of a voice recognition method according to an embodiment of the present application;

[0029] Figure 3 is a flowchart of voice recognition according to an embodiment of the present application;

[0030] Figure 4 is a structure block diagram of a voice recognition device according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0032] It should be noted that the terms "first", "second" and the like in the description and in the claims of the present application and in the above description of the drawings merely mean different instances of similar objects and do not necessarily imply a specific order or chronology. It is to be understood that the data thus used can be interchanged, where appropriate, so that the embodiments of the present application described herein can be carried out in other than the order shown or described herein. Furthermore, the terms "comprise" and "include" and variations thereof as used in the description and in the claims of the present application are intended to cover not exclusively the features explicitly mentioned but also any other steps or units which are in essence equivalent thereto.

[0033] Embodiment 1

[0034] The method provided by the embodiment of the present application can be executed in an Internet of Things device, a voice robot, a smart speaker, a voice controller or a similar device management apparatus. Taking the case of running on a smart speaker, Figure 1 is a hardware structure block diagram of a smart speaker according to an embodiment of the present application. As shown in Figure 1 , the smart speaker can include one or more (only one is shown in Figure 1 ) processors 102 (the processor 102 can include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the above-mentioned smart speaker can further include a transmission device 106 for communication function and an input and output device 108. Those skilled in the art can understand that Figure 1 the structure shown is only schematic, which does not limit the structure of the above-mentioned smart speaker. For example, the smart speaker can further include more or less components than those shown in Figure 1 , or have a different configuration from that shown in Figure 1 .

[0035] The memory 104 can be used to store smart speaker programs, such as software programs and modules of application software, and a smart speaker program corresponding to a voice recognition method according to an embodiment of the present application. The processor 102 executes various function applications and data processing by running the smart speaker program stored in the memory 104, i.e. implements the above-mentioned method. The memory 104 can include a high-speed random access memory, and can further include a non-volatile memory such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to the smart speaker through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0036] The transmission device 106 is configured to receive or send data via a network. The network can include a wireless network provided by a communication provider of the smart speaker. In one example, the transmission device 106 includes a network interface controller (NIC) that can connect to other network devices through a base station to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module configured to communicate with the Internet wirelessly.

[0037] In the embodiment, a speech recognition method is provided, Figure 2 A flowchart of a speech recognition method according to an embodiment of the present application is shown in FIG. 2, which includes the following steps: Figure 2

[0038] In step S202, the noise state of the environment where the target device is located is monitored, wherein the noise state is used to represent the noise intensity of the environment.

[0039] Optionally, the target device can be any device with a voice collection function, such as a voice robot, a voice control screen, etc.

[0040] In step S204, a smoothing algorithm is configured according to the noise state.

[0041] In addition to configuring the smoothing algorithm according to the noise state of the environment, the smoothing algorithm can also be configured according to the running state of the target device, such as adjusting the smoothing parameters of the smoothing algorithm according to the running state of the Internet of Things device (such as the working mode, the power, the network connection state, etc.). When the power of the device is low, a smoothing algorithm with low computational complexity is used to preferentially ensure the basic function of voice recognition and reduce unnecessary calculation consumption. When the device is in a high-load working mode (such as a smart camera simultaneously recording a video and performing voice interaction), more calculation resources are dynamically allocated to the smoothing optimization link of voice processing to improve the voice recognition accuracy.

[0042] In step S206, the voice signal collected by the target device is smoothed according to the noise state and the smoothing algorithm to obtain a target voice signal.

[0043] In step S208, a voice recognition model is used to recognize the target voice signal.

[0044] Optionally, the voice recognition model can be deployed on the target device that collects the voice signal, or deployed on a cloud connected to the target device. When deployed on the cloud, the target voice signal needs to be transmitted to the cloud server.

[0045] ​For example, the smoothed target voice signal is input to a voice recognition model (assuming a voice recognition system combining a deep neural network and a hidden Markov model) for recognition. The recognition result is compared and matched with a locally stored standard voice instruction library (containing preset control instructions, query questions, etc.) to obtain the final recognition result.

[0046] By the above steps, the noise state of the environment where the target device is located is monitored, wherein the noise state is used to represent the noise intensity of the environment; the smoothing algorithm is configured according to the noise state; the voice signal collected by the target device is smoothed according to the noise state and the smoothing algorithm to obtain a target voice signal; and the target voice signal is recognized by using a voice recognition model. By monitoring the noise state of the environment where the target device is located, and adaptively configuring the smoothing algorithm based on the noise state, and finally smoothing the voice signal according to the noise state and the configured smoothing algorithm, the subsequent voice recognition model can more fully utilize and recognize the voice feature information carried in the voice signal, thereby solving the technical problem of low voice recognition accuracy in the related art and improving the user experience.

[0047] In one embodiment of the present embodiment, monitoring the noise state of the environment where the target device is located includes: obtaining the environmental sound collected by the target device at a preset period; calculating the average sound intensity of the environmental sound at the preset period; determining whether the average sound intensity is less than a preset threshold; if the average sound intensity is less than the preset threshold, determining that the noise state of the environment where the target device is located is a quiet state; and if the average sound intensity is greater than or equal to the preset threshold, analyzing the frequency value of the environmental sound, and if the frequency value is distributed within a preset frequency range, determining that the noise state of the environment where the target device is located is a noisy state.

[0048] Optionally, the sound intensity threshold corresponding to the preset threshold is set to 40 decibels. The preset period is 5 seconds.

[0049] In the present embodiment, a sound intensity threshold is preset in the Internet of Things device to distinguish between quiet and noisy environments, and the sound intensity threshold of a quiet environment is set to 40 decibels (dB). The environmental sound intensity is monitored in real time by the microphone built in the device, and the average sound intensity in a period of time (such as 5 seconds) is continuously sampled and calculated. When the average sound intensity is lower than 40 dB, it is determined to be a bedroom type quiet environment; and when it is higher than this threshold, it starts to evaluate whether it is a living room noisy environment.

[0050] In the confirmation of the noisy environment, the frequency range and the fluctuation range of the environmental sound can be determined. When the frequency range of the environmental sound is within the preset frequency range and the fluctuation range is less than the preset range, it is determined that the environment is a noisy environment. The frequency fluctuation threshold is set. Since the sound emitted by the television and other devices has a specific frequency range and fluctuation rule, when a signal with a continuous frequency fluctuation in the range of 100Hz-10000Hz is detected (such as a television audio signal feature), and the sound intensity is higher than the set threshold, it is confirmed that the living room has a noisy environment with background noise such as television and sound.

[0051] The scheme of the embodiment is adopted. The average sound intensity and frequency value of the environmental sound are used to identify the noise state of the environment, and the identification accuracy of the noise state is improved.

[0052] In one embodiment of the present embodiment, configuring a smoothing algorithm according to the noise state includes: if the noise state is a quiet state, selecting a first smoothing algorithm with a first order, wherein the first order is greater than a preset order; configuring an interpolation weight parameter of the first smoothing algorithm based on the first order, and setting an adjustment step of the first smoothing algorithm, wherein the interpolation weight parameter is used to represent the weight coefficient of each language model in the first smoothing algorithm, and the adjustment step is used to indicate the probability amplitude allowed by the first smoothing algorithm.

[0053] The quiet environment can be a bedroom environment, a living room environment with little interference sound, etc. Optionally, the first smoothing algorithm is an interpolation smoothing algorithm, and the first order is 4.

[0054] For more accurate but slightly larger calculation smoothing algorithm used in the bedroom, such as interpolation smoothing algorithm based on high-order N-gram (assuming 4-gram) model. Initialize the interpolation weight parameters, including 1-gram weight, 2-gram weight, 3-gram weight, and 4-gram weight. At the same time, set the probability adjustment step parameter to fine-tune the result when calculating the probability to improve accuracy.

[0055] For example, when using gradient descent and other optimization algorithms to train the language model of the first smoothing algorithm, after calculating the gradient direction of the probability, instead of directly updating the probability according to the gradient size, the step parameter is multiplied to realize small and gradual probability update, avoid missing the optimal probability value or making the model unstable due to the large update amplitude. Assuming that the gradient calculation of the interpolation weight parameter shows that the probability of a certain word group needs to be increased by 0.5, but the adjustment step is set to 0.05, so the actual value of the probability of the word group in this iteration training is 0.5*0.05. After multiple iterations of training, the probabilities of various word groups in the language model are gradually adjusted to make them more accurate.

[0056] The scheme of the embodiment is configured with a smoothing algorithm of a large order for a quiet environment, and is configured with an interpolation weight parameter and an adjustment step of the smoothing algorithm, reasoning is performed through a language model of a large order, and the reasoning result of the model is fine-tuned, so that the accuracy of speech recognition in a quiet environment is improved.

[0057] In another embodiment of the embodiment, configuring the smoothing algorithm according to the noise state includes: if the noise state is a noisy state, selecting a second smoothing algorithm of a second order, wherein the second order is less than a preset order; configuring an anti-noise parameter and a smoothing factor of the second smoothing algorithm, wherein the anti-noise parameter is used to indicate the attenuation intensity of the speech signal of different frequency ranges, and the smoothing factor is used to improve the probability of the word combination whose occurrence frequency is lower than a preset value.

[0058] Optionally, the second smoothing algorithm is an anti-noise fast smoothing algorithm parameter, and the second order can be 2.

[0059] In the embodiment, the noisy state can be the noise state of an environment such as a living room, a public place (such as a shopping mall, a station, etc.), and the like. For example, when used in a user's living room environment, a fast smoothing algorithm focusing on anti-noise can be used, and a method based on frequency shielding and a low-order N-gram (2-gram) combination can be used. The anti-noise parameter includes a frequency shielding threshold range, for example, for frequency components lower than 200 Hz and higher than 8000 Hz, the shielding coefficient is set to 0.8 (that is, the signal strength is attenuated by 20%), and for the part in the range of 200 Hz-8000 Hz and meeting the television audio characteristic frequency interval, the shielding coefficient is set to 0.5. The initial configuration probability smoothing factor of the 2-gram model is set to β = 0.2, which is used to improve the probability of the word combination whose occurrence frequency is lower than a preset value.

[0060] The scheme of the embodiment is configured with a smoothing algorithm of a large order for a quiet environment, and is configured with an interpolation weight parameter and an adjustment step of the smoothing algorithm, reasoning is performed through a language model of a large order, and the reasoning result of the model is fine-tuned, so that the accuracy of speech recognition in a quiet environment is improved.

[0061] In one implementation scenario of the embodiment, the smoothing processing is performed on the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target voice signal, including: collecting a first sound signal if the noise state is a quiet state; enhancing the amplitude of the high-frequency component in the first sound signal to obtain a first intermediate voice; performing frame processing on the first intermediate voice to obtain a second intermediate voice; performing windowing processing on each frame of the second intermediate voice by using a window function to obtain a third intermediate voice, and filtering invalid frames in the third intermediate voice by using an energy value and a zero-crossing rate; performing interpolation smoothing calculation on the third intermediate voice by using the smoothing algorithm to obtain a first target voice signal.

[0062] In this implementation scenario, the high-frequency component in the first sound signal is a component in a frequency band of 2560 to 5120 Hz.

[0063] For example, when the device determines to be in a quiet bedroom environment, the microphone collects voice data at a high sampling frequency (such as 44.1 kHz). The collected data is first pre-emphasized to enhance the signal energy of the high-frequency part, and then frame-processed, with a frame length of 25 ms and a frame shift of 10 ms. A Hamming window function is used for windowing processing on each frame of data to reduce spectral leakage.

[0064] For example, when the pre-emphasis coefficient is set to 0.97, the amplitude of the high-frequency component is enhanced according to a specific mathematical relationship, so that the energy ratio of the high-frequency part in the entire voice signal is relatively more reasonable, and the subsequent processing can more fully utilize the voice feature information carried by the high-frequency.

[0065] The energy value and the zero-crossing rate of each frame of voice data can also be calculated to remove frames with energy lower than a threshold (such as 10% of the average energy) and abnormal zero-crossing rate (such as 5 times higher than the average zero-crossing rate). These frames can be noise or invalid signals.

[0066] By using the scheme of the embodiment, the amplitude enhancement, frame windowing, and invalid frame filtering of the sound signal collected in the quiet state can make the subsequent smoothing algorithm processing more fully utilize and recognize the voice feature information carried in the first sound signal, and improve the voice processing efficiency of the smoothing algorithm.

[0067] In one example, the third intermediate speech is interpolated and smoothed by using the smoothing algorithm to obtain a first target speech signal, including: constructing a first language model of an n-order sliding window, where n is greater than 1; obtaining a statistical result based on the first language model and the occurrence frequency of each order word group in a text corpus; determining a first word group in the third intermediate speech that does not occur in the text corpus based on the statistical result; calculating the occurrence probability of each word in the first word group in the n-order sliding window, and determining the initial probability of the first word group; and calculating the final probability of the first word group in the third intermediate speech by using the occurrence probability and the initial probability to obtain a first target speech signal.

[0068] In this example, taking n=4 as an example, based on the interpolation and smoothing calculation of high-order N-gram, a 4-gram language model is constructed, and the occurrence frequencies of 4-gram, 3-gram, 2-gram and 1-gram are counted from the locally stored text corpus (assuming to contain a large amount of daily conversation, story, news and other texts). For example, for the corpus “I went to the park today”, the occurrence times of different order word groups such as “I went to the park today”, “went to the park today”, “went to the park”, and “park” are counted.

[0069] According to the statistical result, the probability distribution is calculated, and for the 4-gram combination that does not occur in the corpus, the probability is calculated by using the interpolation and smoothing algorithm. Assuming that a 4-gram word group “watching movies and eating popcorn” is encountered, the final probability is calculated by weighting according to the occurrence probability of the words “watching”, “movie”, “popcorn” and “popcorn” in 1-gram, the probability of the combinations “watching movies” and “popcorn” in 2-gram, the probability of the combination “watching movies and eating popcorn” in 3-gram, and the initial probability estimation of 4-gram (set as a very small value, such as 10-6). The final probability P is calculated by weighting, such as P=0.1*p1+0.2*p2...., where P1 is the 1-gram probability, p2 is the 2-gram, and 0.1 and 0.2 are interpolation weight parameters pre-configured by the smoothing algorithm, each corresponding to a language model, and the sum is 1. In the calculation process, the final probability is fine-tuned according to the probability adjustment step to improve the accuracy.

[0070] By using the scheme of the embodiment, the occurrence frequency of the word group is counted by using the first language model of the high-order sliding window, the probability distribution of the unfamiliar first word group is calculated, the recognition rate of the unfamiliar word group is improved, and the accuracy of the speech recognition in the quiet state is improved.

[0071] In another implementation scenario of the embodiment, the smoothing of the sound signal collected by the target device according to the noise state and the smoothing algorithm obtains a target voice signal, including: collecting a second sound signal if the noise state is a noisy state; converting the second sound signal into a frequency domain signal; analyzing a characteristic parameter of environmental sound in the frequency domain signal; filtering the environmental sound in the second sound signal based on the characteristic parameter to obtain a fourth intermediate voice; and performing anti-noise smoothing calculation on the fourth intermediate voice using the smoothing algorithm to obtain a second target voice signal.

[0072] In this implementation scenario, when the device detects that it is in a noisy environment in the living room and there is background noise such as a television, the collected mixed sound signal is first subjected to fast Fourier transform (FFT) to convert the second sound signal into a frequency domain signal, and then the frequency domain signal characteristics are analyzed to determine the main frequency component and energy distribution of the television audio signal (taking the environmental sound as an example). An adaptive filter is used for signal separation, and the estimated component of the television audio signal is subtracted from the mixed signal using the television audio signal characteristics as a reference signal. For example, according to the energy concentration characteristics of the television audio signal in a specific frequency range (such as 500Hz-3000Hz), the filter parameters are set so that the signal in this frequency range is attenuated after passing through the filter.

[0073] Using the scheme of this implementation scenario, the frequency domain conversion, environmental sound filtering and other processing of the sound signal collected in the noisy state can make the subsequent smoothing algorithm processing more fully utilize and recognize the voice feature information carried in the second sound signal, thereby improving the voice processing efficiency of the smoothing algorithm.

[0074] In one example, the smoothing algorithm is used to perform anti-noise smoothing calculation on the fourth intermediate voice to obtain a second target voice signal, including: performing attenuation processing on the voice signal in the noise frequency range of the fourth intermediate voice to obtain a fifth intermediate voice; constructing a second language model of an m-order sliding window, where m is greater than 1; based on the second language model, the frequency of occurrence of each order word group in a text corpus is counted to obtain a statistical result; based on the statistical result, a second word group with a frequency of occurrence lower than a preset frequency in the fifth intermediate voice is screened; and the probability of occurrence of the second word group in the fifth intermediate voice is improved to obtain a second target voice signal.

[0075] Optionally, m = 2.

[0076] In the anti-noise smoothing calculation, the fourth intermediate speech after signal separation is processed by the frequency mask and low-order N-gram fast smoothing. The speech signal is processed according to the set frequency mask threshold range, and the residual television noise frequency components are further removed. At the same time, a 2-gram language model is constructed to calculate the probability of the word combination in the speech signal. For the second word combination with low frequency, the probability estimate value is improved according to the set smoothing factor. For example, when processing the speech signal "turn on the TV", if the combination of "turn on" and "TV" appears less frequently in the training data, the probability in the recognition process is adjusted by the smoothing factor to improve the recognition accuracy.

[0077] The second word group is a low-frequency word group, which is a word with a low probability of appearing in the learning corpus and vocabulary. The size of the learning library can be set, such as setting less than 100 times as low frequency. When improving the probability estimate value of the second word group, λ1=0.5, λ2=0.3, and λ3=0.2 can be set (λ represents the degree of dependence of the smoothing factor on the back-off probability, which is equivalent to giving the back-off probability a certain weight to improve the overall probability estimate). Assuming that the word combination "sunlight shines on the earth" in the corpus has a statistical result of count(n sunlight, shines, earth / ) = 3 and count(n sunlight, shines / ) = 10, P(\ earth / | shines") = 0.2 is calculated according to 2-gram, and the probability of "earth" in the unigram language model P(\ earth / ) = 0.1, then P(earth / sunlight, shines) = 0.5x0.3 + 0.3x0.2 + 0.2x0.1 = 0.15 + 0.06 + 0.02 = 0.23. By reasonably setting these interpolation weights (λ1 is related to the smoothing factor), the probability estimate value of the word combination is improved by the smoothing factor, especially for the three-word combination with low frequency, which can better adjust the probability.

[0078] By using the scheme of the embodiment, the frequency range of the decay noise is first processed, and then the second language model with a low-order sliding window is used to count the frequency of the word group, and the probability of the second word group with low frequency is improved, which improves the recognition rate of the low-frequency word group, and further improves the accuracy of speech recognition in a noisy state.

[0079] The scheme of the embodiment designs a lightweight and efficient smoothing algorithm, which utilizes edge computing technology to perform preliminary smoothing processing at the device end, thereby reducing data transmission amount and cloud computing pressure. For example, the smart home device can adopt a customized smoothing strategy according to different room environment noise characteristics and user daily language habits. In a quiet bedroom environment, a more accurate but slightly larger calculation amount smoothing method is adopted; and in a living room with background noise such as a television, a fast smoothing algorithm focusing on noise reduction is automatically switched to, thereby improving the voice interaction experience of users in different Internet of Things application scenarios. Figure 3 is a flowchart of voice recognition performed by the embodiment of the application, which includes environment monitoring and threshold setting, algorithm parameter initialization (including accurate smoothing algorithm parameters and noise-resistant fast smoothing algorithm parameters), selection of different smoothing algorithm parameters according to user requirements, calculation of probability distribution according to statistical results, voice recognition and model updating.

[0080] When updating the model, if it is a quiet state scenario, the language model parameters of the smoothing algorithm are updated according to the accuracy feedback of the recognition result. If the recognition is incorrect, it is analyzed whether the error reason is that the language model probability estimation is inaccurate or other factors (such as voice signal acquisition problems). If it is a probability estimation problem, the interpolation weight parameter is adjusted or the frequency of related word groups is re-counted to optimize the language model.

[0081] If it is a noisy state scenario, the noise-resistant voice recognition and algorithm adjustment input the processed voice signal into a voice recognition model optimized for noisy environments (the model is trained by a large amount of noisy environment voice data and has strong noise resistance) for recognition. According to the accuracy of the recognition result, the frequency shielding threshold and the smoothing factor of the 2-gram model are dynamically adjusted. If the recognition accuracy continuously falls below a certain threshold (such as 70%), the frequency shielding coefficient is appropriately increased (such as increased by 0.1 based on the original) or the smoothing factor is adjusted (such as increased by 0.05) to improve the noise resistance effect and recognition accuracy; if the accuracy is high and stable, the current algorithm parameters remain unchanged, and voice recognition processing continues. Through the above specific implementation schemes in different environments, the Internet of Things device can automatically switch to a suitable smoothing algorithm according to the environment noise, thereby improving the accuracy and efficiency of voice recognition.

[0082] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be realized by means of software on a necessary general hardware platform, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product in essence or the part that contributes to the prior art, and the computer software product is stored in a storage medium (such as a ROM / RAM, a magnetic disk, an optical disk), and includes a plurality of instructions for causing an end device (which can be a mobile phone, a computer, or a network device, etc.) to execute the method described in each embodiment of the present application.

[0083] Embodiment 2

[0084] In this embodiment, a speech recognition device is also provided, which is used to implement the above embodiments and preferred embodiments, and has been described above and will not be repeated. The term "module" as used below can be a combination of software and hardware that implements the scheduled function. Although the device described in the following embodiments is preferably implemented in software, hardware or a combination of software and hardware can also be conceived.

[0085] Figure 4 is a structural block diagram of a speech recognition device according to an embodiment of the present application, as shown in Figure 4 , comprising:

[0086] a monitoring module 40, configured to monitor a noise state of an environment where a target device is located, wherein the noise state is used to represent a noise intensity of the environment;

[0087] a configuration module 42, configured to configure a smoothing algorithm according to the noise state;

[0088] a smoothing module 44, configured to perform smoothing processing on a sound signal collected by the target device according to the noise state and the smoothing algorithm, to obtain a target speech signal;

[0089] a recognition module 46, configured to recognize the target speech signal by using a speech recognition model.

[0090] Optionally, the monitoring module comprises: an acquisition unit, configured to acquire environmental sound collected by the target device at a preset period; a calculation unit, configured to calculate an average sound intensity of the environmental sound at the preset period; a judgment unit, configured to judge whether the average sound intensity is less than a preset threshold; a determination unit, configured to determine that the noise state of the environment where the target device is located is a quiet state if the average sound intensity is less than the preset threshold, and to analyze a frequency value of the environmental sound if the average sound intensity is greater than or equal to the preset threshold, and to determine that the noise state of the environment where the target device is located is a noisy state if the frequency value is distributed within a preset frequency range.

[0091] Optionally, the configuration module comprises: a first selection unit configured to select a first smoothing algorithm of a first order if the noise state is a quiet state, wherein the first order is greater than a preset order; and a first configuration unit configured to configure an interpolation weight parameter of the first smoothing algorithm based on the first order, and set an adjustment step of the first smoothing algorithm, wherein the interpolation weight parameter is used to represent a weight coefficient of each order language model in the first smoothing algorithm, and the adjustment step is used to indicate a probability amplitude allowed to be adjusted by the first smoothing algorithm.

[0092] Optionally, the configuration module comprises: a second selection unit configured to select a second smoothing algorithm of a second order if the noise state is a noisy state, wherein the second order is less than the preset order; and a second configuration unit configured to configure an anti-noise parameter and a smoothing factor of the second smoothing algorithm, wherein the anti-noise parameter is used to indicate an attenuation intensity of a speech signal in different frequency ranges, and the smoothing factor is used to improve a probability of a vocabulary combination whose occurrence frequency is lower than a preset value.

[0093] Optionally, the smoothing module comprises: a first acquisition unit configured to acquire a first sound signal if the noise state is a quiet state; an enhancement unit configured to enhance an amplitude of a high-frequency component in the first sound signal to obtain a first intermediate speech; a frame division unit configured to perform frame division processing on the first intermediate speech to obtain a second intermediate speech; a windowing unit configured to perform windowing processing on each frame in the second intermediate speech by using a window function to obtain a third intermediate speech, and filter invalid frames in the third intermediate speech by using an energy value and a zero-crossing rate; and a first smoothing unit configured to perform interpolation smoothing calculation on the third intermediate speech by using the smoothing algorithm to obtain a first target speech signal.

[0094] Optionally, the first smoothing unit comprises: a construction sub-unit configured to construct a first language model of an n-order sliding window, wherein n is greater than 1; a statistics sub-unit configured to obtain a statistics result by counting occurrence frequencies of each order vocabulary group in a text corpus based on the first language model; a determination sub-unit configured to determine a first vocabulary group not appearing in the text corpus in the third intermediate speech based on the statistics result; a first calculation sub-unit configured to calculate occurrence probabilities of words in the first vocabulary group in the n-order sliding window respectively, and determine an initial probability of the first vocabulary group; and a second calculation sub-unit configured to calculate a final probability of the first vocabulary group in the third intermediate speech by using the occurrence probabilities and the initial probability to obtain a first target speech signal.

[0095] Optionally, the smoothing module comprises: a second acquisition unit, configured to acquire a second sound signal if the noise state is a noisy state; a conversion unit, configured to convert the second sound signal into a frequency domain signal; an analysis unit, configured to analyze a characteristic parameter of environmental sound in the frequency domain signal; a filtering unit, configured to perform filtering processing on the environmental sound in the second sound signal based on the characteristic parameter to obtain a fourth intermediate speech; and a second smoothing unit, configured to perform anti-noise smoothing calculation on the fourth intermediate speech using the smoothing algorithm to obtain a second target speech signal.

[0096] Optionally, the second smoothing unit comprises: an attenuation sub-unit, configured to perform attenuation processing on a speech signal in a noise frequency range in the fourth intermediate speech to obtain a fifth intermediate speech; a construction sub-unit, configured to construct a second language model of an m-order sliding window, where m is greater than 1; a statistics sub-unit, configured to statistically analyze the occurrence frequency of each-order word group in a text corpus based on the second language model to obtain a statistical result; a screening sub-unit, configured to screen a second word group with an occurrence frequency lower than a preset frequency in the fifth intermediate speech based on the statistical result; and a promotion sub-unit, configured to promote the occurrence probability of the second word group in the fifth intermediate speech to obtain a second target speech signal.

[0097] It should be noted that the above various modules can be implemented by software or hardware, and for the latter, the following implementation manners can be used, but are not limited thereto: the above modules are located in the same processor; or the above various modules are located in different processors in any combination.

[0098] Embodiment 3

[0099] The embodiments of the present application also provide a storage medium in which a computer program is stored, wherein the computer program is configured to execute the steps in any of the method embodiments when running.

[0100] Optionally, in the present embodiment, the above storage medium can be configured to store a computer program for execution.

[0101] S1, monitoring a noise state of an environment where a target device is located, wherein the noise state is used to represent the noise intensity of the environment;

[0102] S2, configuring a smoothing algorithm according to the noise state;

[0103] S3, performing smoothing processing on a sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target speech signal;

[0104] S4, identifying the target speech signal using a speech recognition model.

[0105] Optionally, in the embodiment, the storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various storage media that can store computer programs.

[0106] The embodiment of the application further provides an electronic device including a memory and a processor, the memory stores a computer program, and the processor is configured to execute the computer program to perform the steps in any of the method embodiments.

[0107] Optionally, the electronic device can further include a transmission device and an input / output device, wherein the transmission device is connected with the processor, and the input / output device is connected with the processor.

[0108] Optionally, in the embodiment, the processor can be configured to execute the following steps through the computer program:

[0109] S1, monitoring a noise state of an environment where a target device is located, wherein the noise state is used to represent a noise intensity of the environment;

[0110] S2, configuring a smoothing algorithm according to the noise state;

[0111] S3, performing smoothing processing on a sound signal collected by the target device according to the noise state and the smoothing algorithm, to obtain a target voice signal;

[0112] S4, identifying the target voice signal by using a voice recognition model.

[0113] Optionally, specific examples in the embodiment can refer to the examples described in the above embodiments and optional implementation manners, and the embodiment will not be described here.

[0114] The serial numbers of the above embodiments of the application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0115] In the above embodiments of the application, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can refer to the related description of other embodiments.

[0116] In several embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other manners. For example, the described unit embodiments can be divided into other ways, for example, the division of units is only a logical function division, and there can be another division manner in actual implementation; for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, accessors or modules, and can be electrical, mechanical or in other forms.

[0117] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0118] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0119] 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 program product readable by a computer. Based on such an understanding, the technical solutions of the present application essentially or substantially or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a controller, or a network device, etc.) to perform all or part of the steps of the methods in the embodiments of the present application. The foregoing storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various other media that can store program codes.

[0120] The above is only the preferred embodiment of the present application, and it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, some improvements and refinements can be made, and these improvements and refinements should be regarded as the protection scope of the present application.

Claims

1. A speech recognition method, characterized by, The method comprises the following steps: monitoring the noise state of the environment where the target device is located, wherein the noise state is used to represent the noise intensity of the environment; configuring a smoothing algorithm according to the noise state; smoothing the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target voice signal; using a voice recognition model to recognize the target voice signal; wherein configuring a smoothing algorithm according to the noise state comprises: if the noise state is a quiet state, selecting a first smoothing algorithm with a first order greater than a preset order; configuring the interpolation weight parameters of the first smoothing algorithm based on the first order, and setting the adjustment step of the first smoothing algorithm, wherein the interpolation weight parameters are used to represent the weight coefficients of each order language model in the first smoothing algorithm, and the adjustment step is used to indicate the probability amplitude allowed by the first smoothing algorithm; if the noise state is a noisy state, selecting a second smoothing algorithm with a second order less than a preset order; configuring the noise reduction parameter and the smoothing factor of the second smoothing algorithm, wherein the noise reduction parameter is used to indicate the attenuation intensity of the voice signal in different frequency ranges, and the smoothing factor is used to improve the probability of the word combination with an occurrence frequency lower than a preset value.

2. The method of claim 1, wherein, Monitoring the noise state of the environment where the target device is located comprises: acquiring the environmental sound collected by the target device according to a preset period; calculating the average sound intensity of the environmental sound in the preset period; determining whether the average sound intensity is less than a preset threshold; if the average sound intensity is less than the preset threshold, determining that the noise state of the environment where the target device is located is a quiet state; if the average sound intensity is greater than or equal to the preset threshold, analyzing the frequency value of the environmental sound, and if the frequency value is distributed within a preset frequency range, determining that the noise state of the environment where the target device is located is a noisy state.

3. The method of claim 1, wherein, Smoothing the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target voice signal comprises: if the noise state is a quiet state, collecting a first sound signal; enhancing the amplitude of the high frequency component in the first sound signal to obtain a first intermediate voice; performing frame processing on the first intermediate voice to obtain a second intermediate voice; performing windowing processing on each frame of the second intermediate voice using a window function to obtain a third intermediate voice, and filtering invalid frames in the third intermediate voice through energy value and zero-crossing rate; performing interpolation smoothing calculation on the third intermediate voice using the smoothing algorithm to obtain a first target voice signal.

4. The method of claim 3, wherein, Performing interpolation smoothing calculation on the third intermediate voice using the smoothing algorithm to obtain a first target voice signal comprises: constructing a first language model of an n-order sliding window, wherein n is greater than 1; statistically obtaining the occurrence frequency of each order word group in a text corpus based on the first language model to obtain a statistical result; determining a first word group in the third intermediate voice that does not appear in the text corpus based on the statistical result; Calculate the occurrence probability of each word in the first word group in an n-order sliding window, and determine the initial probability of the first word group; Calculate the final probability of the first word group in the third intermediate speech by weighting the occurrence probability and the initial probability, to obtain a first target speech signal.

5. The method of claim 1, wherein, According to the noise state and the smoothing algorithm, the sound signal collected by the target device is smoothed to obtain a target speech signal, including: If the noise state is a noisy state, a second sound signal is collected; The second sound signal is converted into a frequency domain signal; The characteristic parameters of the environmental sound in the frequency domain signal are analyzed; Based on the characteristic parameters, the environmental sound in the second sound signal is filtered to obtain a fourth intermediate speech; The fourth intermediate speech is calculated by the smoothing algorithm to obtain a second target speech signal.

6. The method of claim 5, wherein, The fourth intermediate speech is calculated by the smoothing algorithm to obtain a second target speech signal, including: The speech signal in the noise frequency range of the fourth intermediate speech is attenuated to obtain a fifth intermediate speech; A second language model of an m-order sliding window is constructed, where m is greater than 1; Based on the second language model, the occurrence frequency of each order word group in the text corpus is counted to obtain a statistical result; Based on the statistical result, a second word group with an occurrence frequency lower than a preset frequency in the fifth intermediate speech is screened; The occurrence probability of the second word group in the fifth intermediate speech is improved to obtain a second target speech signal.

7. A speech recognition apparatus characterized by comprising: Including: A monitoring module for monitoring the noise state of the environment where the target device is located, wherein the noise state is used to represent the noise intensity of the environment; A configuration module for configuring a smoothing algorithm according to the noise state; A smoothing module for smoothing the sound signal collected by the target device according to the noise state and the smoothing algorithm to obtain a target speech signal; An identification module for identifying the target speech signal using a speech recognition model; The configuration module includes: a first selection unit for selecting a first-order first smoothing algorithm if the noise state is a quiet state, wherein the first order is greater than a preset order; a first configuration unit for configuring the interpolation weight parameter of the first smoothing algorithm based on the first order, and setting the adjustment step of the first smoothing algorithm, wherein the interpolation weight parameter is used to represent the weight coefficient of each language model in the first smoothing algorithm, and the adjustment step is used to indicate the probability amplitude allowed by the first smoothing algorithm; a second selection unit for selecting a second-order second smoothing algorithm if the noise state is a noisy state, wherein the second order is less than a preset order; a second configuration unit for configuring the noise reduction parameter and the smoothing factor of the second smoothing algorithm, wherein the noise reduction parameter is used to indicate the attenuation intensity of the speech signal in different frequency ranges, and the smoothing factor is used to improve the probability of the word combination with an occurrence frequency lower than a preset value.

8. A storage medium, characterized by The computer program is stored in the storage medium and is configured to execute the steps of the speech recognition method in any one of claims 1 to 6 when running.

9. An electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein, The processor, the communication interface and the memory complete mutual communication through a communication bus; wherein: The memory is used for storing a computer program. The processor is used for executing the steps of the speech recognition method in any one of claims 1 to 6 by running the program stored in the memory.

Images