CN111489740A - Voice processing method and device and elevator control method and device - Google Patents

Abstract

The present invention provides a voice processing method and device, and an elevator control method and device. The voice processing method includes: extracting a first voice feature of the to-be-processed voice; Two voice features; obtaining the processing result of the voice to be processed based on the second voice feature. The embodiment of the present invention can improve the performance of the network model in the process of speech processing.

Description

Voice processing method and device, elevator control method and device

technical field

The present invention relates to the technical field of natural language processing, and in particular, to a method and device for speech processing, and a method and device for controlling an elevator.

Background technique

Natural language refers to the language used for communication between human beings produced through natural evolution. Natural Language Processing (NLP) is a field of computer science, artificial intelligence, and linguistics concerned with the interaction between computers and human (natural) language. Natural language processing technology can process speech based on network models to meet the needs of various usage scenarios. For example, in speech recognition usage scenarios, speech can be processed into text based on speech recognition network models.

In the process of voice processing, the same voice content and different volume will lead to different voice amplitudes, so the voice characteristics also show great differences, but in fact, the results of voice processing are the same, which will make the network The performance of the model is poor.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a voice processing method and device, and an elevator control method and device, so as to solve the problem that in the prior art, the amplitude of the voice is different due to the different volume, so that the voice features show a large difference, so that the performance of the network model is better than that of the network model. Bad question.

In order to solve the above-mentioned technical problems, the present invention is achieved in this way:

In a first aspect, an embodiment of the present invention provides a speech processing method, the method comprising:

extracting the first speech feature of the speech to be processed;

Performing voice amplitude feature separation processing on the first voice feature to obtain a second voice feature;

The processing result of the to-be-processed speech is acquired based on the second speech feature.

In a second aspect, an embodiment of the present invention provides an elevator control method, the method comprising:

Receive the target voice input by the user in the elevator scenario;

Use the voice processing method described in the embodiment of the present invention to perform off-line intention recognition on the target voice to obtain first control information;

The elevator is controlled to perform the first operation corresponding to the first control information.

In a third aspect, an embodiment of the present invention provides a voice processing apparatus, where the voice processing apparatus includes:

an extraction module for extracting the first speech feature of the speech to be processed;

A separation module, configured to perform speech amplitude feature separation processing on the first speech feature to obtain a second speech feature;

An obtaining module, configured to obtain the processing result of the voice to be processed based on the second voice feature.

In a fourth aspect, an embodiment of the present invention provides an elevator control device, and the elevator control device includes:

The first receiving module is used to receive the target voice input by the user under the scene of using the elevator;

a recognition module, configured to perform off-line intention recognition on the target voice by using the voice processing method described in the embodiment of the present invention to obtain first control information;

The first control module is used to control the elevator to perform the first operation corresponding to the first control information.

In a fifth aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a program stored in the memory and executable on the processor, when the program is executed by the processor The steps in the voice processing method according to the first aspect are implemented, or, when the program is executed by the processor, the steps in the elevator control method according to the second aspect are implemented.

In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the voice processing described in the first aspect is implemented The steps in the method, or, when the computer program is executed by the processor, implements the steps in the elevator control method according to the second aspect.

In the embodiment of the present invention, a first voice feature of the to-be-processed voice is extracted; a voice amplitude feature separation process is performed on the first voice feature to obtain a second voice feature; and a second voice feature is obtained based on the second voice feature process result. In this way, the voice signals with the same voice content and different volumes can be characterized as having a magnification of the voice amplitude. Thereby, the performance of the network model can be improved.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative labor.

1 is a flowchart of a speech processing method provided by an embodiment of the present invention;

2 is a schematic diagram of a network model learning provided by an embodiment of the present invention;

3 is a schematic structural diagram of a voice processing apparatus provided by an embodiment of the present invention;

Fig. 4 is one of the structural schematic diagrams of an elevator control device provided by an embodiment of the present invention;

Fig. 5 is the second structural schematic diagram of an elevator control device provided by an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Referring to FIG. 1, FIG. 1 is a flowchart of a speech processing method provided by an embodiment of the present invention, as shown in FIG. 1, including the following steps:

Step 101: Extract the first speech feature of the speech to be processed.

Wherein, the first speech feature may include a plurality of first feature values. The plurality of first eigenvalues may be obtained based on a logarithmic operation, and may take a logarithm with a constant e as a base, or may also take a logarithm with other numbers as a base, which is not limited in this embodiment of the present invention. The first voice feature can be filter banks (filter bank) feature.

Step 102: Perform a voice amplitude feature separation process on the first voice feature to obtain a second voice feature.

Wherein, the second speech feature may include a plurality of second feature values. The performing voice amplitude feature separation processing on the first voice feature to obtain a second voice feature may include: performing feature averaging processing on the plurality of first feature values to obtain a voice amplitude feature value, the voice amplitude feature The value is used to characterize the speech amplitude feature of the to-be-processed speech, and based on the speech amplitude feature value, each of the first feature values is subjected to speech amplitude feature separation processing to obtain the second speech features corresponding to the second eigenvalue of each of the first eigenvalues.

Step 103: Acquire a processing result of the to-be-processed speech based on the second speech feature.

Wherein, the obtaining the processing result of the speech to be processed based on the second speech feature may be, performing speech recognition based on the second speech feature to obtain a speech recognition result; or may also be, based on the first speech recognition The second voice feature is used for voice translation, and the result of the voice translation is obtained; or, the second voice feature is used in other usage scenarios to obtain the processing result of the to-be-processed voice, which is not limited in this embodiment of the present invention. . In the process of training the network model, the second speech feature may be input into the network model for training; or, in the process of using the network model for prediction, the second speech feature may be input into the network model for prediction.

In practical applications, taking the voice processing of the first voice and the second voice as an example, the second voice can be the same as the voice content of the first voice, and the second voice can be obtained by amplifying the volume of the first voice n times, The first voice feature of the first voice and the first voice feature of the second voice may both be filter banks features. A plurality of first feature values of the first voice can be (a ₁ , a ₂ , a ₃ , ..., a _i ), since the second voice is obtained by amplifying the volume of the first voice n times, and the filter banks feature is based on taking pairs Obtained by arithmetic operation, the plurality of first eigenvalues of the second speech are (a ₁ +ln(n), a ₂ +ln(n), a ₃ +ln(n), ..., a _i +ln(n)) . The speech amplitude feature value may be the average value of the plurality of first feature values, and the average value of the plurality of first feature values of the first speech is: a _avg =(a ₁ +a ₂ +a ₃ +...+ a _i )/i, the average value of multiple first feature values of the second speech is: a _avg +ln(n), and the relationship between each first feature value in the multiple first feature values and the The difference value of the average value is obtained to obtain the second eigenvalue corresponding to each of the first eigenvalues, and the plurality of second eigenvalues of the first speech obtained by calculation may be (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , ..., a _i -a _avg ), the calculated second eigenvalues of the second speech may be (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , ... , a _i -a _avg ).

In practical applications, one-dimensional eigenvalues can be added to the improved filter banks feature to represent the speech amplitude feature of the speech to be processed. The speech amplitude feature can represent the volume, so that the eigenvalues of other dimensions are independent of the volume. For example, you can use (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , ..., a _i -a _avg , a _avg ) as the improved filterbanks features of the first speech, (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , ..., a _i -a _avg , a _avg +ln(n)) as an improved filter banks feature for the second voice; alternatively, (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , ..., a _i -a _avg ) as the improved filter banks feature of the first speech, which can be (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg ) , ..., a _i -a _avg ) as an improved filter banks feature for the second speech. For the same speech content, changing the speech amplitude, the first i dimension of the improved filter banks feature is a fixed value and will not change, only the eigenvalue representing the average value of the last dimension will change, the first i dimension is a pure feature, and the amplitude The size has nothing to do with the network model learning. Using the same network model, the performance of the improved filter banks feature can be improved by 3% to 10% over the filter banks feature during speech processing.

It should be noted that when the volume of the voice is within a certain range, for example, the volume of the voice is greater than a preset value, and the same voice content results in the same voice recognition result. The volume has the same effect on each dimension of the feature vector of the speech feature, which will reduce the efficiency of network model learning in the process of speech processing. Taking the filter banks feature as an example, the filterbanks feature can be obtained by Fourier transform, Mel filtering and logarithmic operations in turn. If the volume of the voice is amplified by n times, and n is a positive integer, the amplitude of the voice is amplified. n times, through the Fourier transform, in the spectrum, the corresponding amplification is n times; through mel filtering, the amplification is maintained n times; by taking the logarithmic operation, the volume of the voice amplified by n times is larger than that of the unamplified voice in the filter banks feature , each dimension increases by ln(n). As shown in Figure 2, for the network model, when the input voice content is A, but the filter banks feature of multiple voices with different volumes, the same voice content, because the voice amplitude is different, A varies with the amplitude. The value affects the translation on a straight line, but when performing speech processing such as speech recognition, the results obtained are all A. Since the volume is different, each dimension of the filter banks feature is doing the same translation, and the speech amplitude has the same effect on each dimension, which will increase the burden of the network model.

In the embodiment of the present invention, a first voice feature of the to-be-processed voice is extracted; a voice amplitude feature separation process is performed on the first voice feature to obtain a second voice feature; and a second voice feature is obtained based on the second voice feature process result. In this way, the voice signals with the same voice content and different volumes can be characterized as having a magnification of the voice amplitude. Thereby, the performance of the network model can be improved.

Optionally, the first voice feature includes multiple first feature values, and the second voice feature includes multiple second feature values;

The said first voice feature is subjected to voice amplitude feature separation processing to obtain a second voice feature, including:

Performing feature averaging processing on the plurality of first feature values to obtain a voice amplitude feature value, where the voice amplitude feature value is used to characterize the voice amplitude feature of the to-be-processed voice;

Based on the speech amplitude feature values, the speech amplitude feature separation processing is performed on each of the first feature values respectively, and the second features corresponding to each of the first feature values in the second speech features are obtained. value.

The plurality of first eigenvalues may be obtained based on a logarithmic operation, the logarithm may be obtained by taking the constant e as the base, or the logarithm may be obtained by taking other numbers as the base, which is not performed in this embodiment of the present invention. limited.

In addition, the feature averaging process performed on the plurality of first feature values may be to calculate an average value of the plurality of first feature values based on the plurality of first feature values, and the speech amplitude feature value may be is the average value of the plurality of first feature values. The voice amplitude feature separation processing is performed on each of the first feature values based on the feature value of the voice amplitude, to obtain the first feature value corresponding to each of the first feature values in the second voice feature. Two eigenvalues, which can be obtained from the second speech feature corresponding to each the second eigenvalue of the first eigenvalue.

In addition, obtaining the second speech feature corresponding to each of the The second eigenvalue of an eigenvalue may include, respectively calculating the difference between each of the first eigenvalues in the plurality of first eigenvalues and the average value to obtain each of the first eigenvalues The corresponding second feature value; or, it may include calculating the difference between each first feature value in the plurality of first feature values and the average value, respectively, and comparing the difference with the first preset value Multiplying to obtain a second eigenvalue corresponding to each of the first eigenvalues; or, it may include separately calculating the difference between each of the first eigenvalues in the plurality of first eigenvalues and the average value, and A second feature value corresponding to each of the first feature values is obtained by subtracting the difference value from the second preset value, and so on, which is not limited in this embodiment of the present invention.

Preferably, according to the average value of each of the first feature values in the plurality of first feature values and the plurality of first feature values, the second speech features corresponding to each of the The second eigenvalue of the first eigenvalue may include: separately calculating the difference between each of the first eigenvalues in the plurality of first eigenvalues and the average value to obtain each of the first eigenvalues the corresponding second eigenvalue. It can make at least two voice signals with the same voice content and different volume, the second feature value of each voice signal is the same value, so that the difference of voice features caused by different volume can be further reduced, and then the network model can be improved. performance.

Taking the first voice and the second voice as an example, the second voice can be the same as the voice content of the first voice, and the second voice can be obtained by amplifying the volume of the first voice n times. The first feature value can be obtained based on the logarithmic operation, and the first feature value in the first voice feature of the first voice can be obtained based on the logarithmic operation. Therefore, each first feature in the first voice feature of the second voice The value is increased by ln(n) or log(n) from the corresponding first feature value in the first voice feature of the first voice.

With the first voice feature of the first voice and the first voice feature of the second voice are filter banks feature as example, each first feature value in the first voice feature of the second voice is more than the first voice feature of the first voice _ln ( _{n )} is added to the corresponding first _eigenvalues in The values can be (a ₁ +ln(n), a ₂ +ln(n), a ₃ +ln(n), ..., a _i +ln(n)), where i is a positive integer. The average value of the plurality of first feature values of the first speech may be: a _avg =(a ₁ +a ₂ +a ₃ +...+a _i )/i, the average value of the plurality of first feature values of the second speech It can be: a _avg +ln(n), the difference between each first eigenvalue in the plurality of first eigenvalues and the average value can be calculated separately, and the first eigenvalue corresponding to each first eigenvalue can be obtained. Two eigenvalues, the calculated second eigenvalues of the first speech may be (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , ..., a _i -a _avg ), and the calculated The plurality of second feature values of the second speech may be (a ₁ -a _avg , a ₂ -a _avg , a ₃ -a _avg , . . . , a _i -a _avg ).

In this embodiment, by performing feature averaging processing on the plurality of first feature values, the voice amplitude feature separation processing can be performed on the first voice feature quickly and accurately.

Optionally, the second speech feature further includes the speech amplitude feature value.

Wherein, the speech amplitude feature value may be an average value of the plurality of first feature values. Taking multiple first eigenvalues (x ₁ , x ₂ , x ₃ , . . . , x _k ) and the average value of multiple first eigenvalues as x _avg as an example, corresponding to each of the first eigenvalues The second eigenvalues may be: (x ₁ -x _avg , x ₂ -x _avg , x ₃ -x _avg , . . . , x _k -x _avg , x _avg ). (x ₁ -x _avg , x ₂ -x _avg , x ₃ -x _avg , . . . , x _k -x _avg , x _avg ) can be fed into the network model for further speech processing.

In this embodiment, the second voice feature further includes the voice amplitude feature value, and the volume-related features in the voice signal can be separately extracted as a part of the voice features, so that the voice features caused by different volumes can be reduced. The difference of the speech amplitude can improve the performance of the network model; in addition, the speech amplitude characteristic value can also be used to distinguish noise, and the speech amplitude characteristic value can be used to distinguish noise and speech, and the noise needs to be used in the process of speech processing. In this case, the effect of voice processing on the second voice feature including the voice amplitude feature value is better.

Optionally, the dimension of the second speech feature is greater than or equal to the dimension of the first speech feature.

Wherein, the first voice feature may include multiple first feature values, and the second voice feature may include multiple second feature values. The first feature value may correspond to the second feature value one-to-one, so that the dimension of the second speech feature may be equal to the dimension of the first speech feature. The second voice feature may further include a voice amplitude feature value, and the voice amplitude feature value is used to represent the voice amplitude feature of the to-be-processed voice, so that the dimension of the second voice feature is larger than the dimension of the first voice feature. Further, the second speech feature may further include a feature value used to represent other features of the speech to be processed, which is not limited in this embodiment of the present invention.

In this embodiment, the dimension of the second speech feature is greater than or equal to the dimension of the first speech feature, so that more features of the speech to be processed can be acquired, thereby improving the effect of speech processing.

Optionally, the first voice feature includes filter banks filter bank feature.

Among them, the filter banks feature, that is, the Fbank feature, is a commonly used speech feature at present. The response of the human ear to the sound spectrum is non-linear. The Fbank feature can imitate the human ear to process speech. In the speech recognition process, using Fbank features can improve the performance of speech recognition. The Fbank feature can be obtained by performing Fourier transform and Mel filtering on the speech frame by frame, and then taking the logarithm. In practical applications, Fourier transform can be performed on the speech to be processed to obtain the frequency domain features of the to-be-processed speech, and mel filtering is performed on the frequency-domain features of the to-be-processed speech to obtain a filtering result, and the filtering result is taken as Logarithm to get the Fbank feature of the speech to be processed.

Taking the first voice and the second voice as an example, the second voice can be the same as the voice content of the first voice, and the second voice can be obtained by amplifying the volume of the first voice n times. The first voice features of the two voices may both be filter banks features. For example, after performing Fourier transform and Mel filtering on the first speech, (b ₁ , b ₂ , b ₃ , . . . , b _i ) can be obtained, and by taking the logarithm, the first speech of the first speech can be obtained features(ln b ₁ , ln b ₂ , lnb ₃ , . . . , ln b _i ). After performing Fourier transform and Mel filtering on the second speech, (nb ₁ , nb ₂ , nb ₃ , ..., nb _i ) can be obtained, and by taking the logarithm, the first speech feature of the second speech can be obtained ( ln b ₁ +ln(n), ln b ₂ +ln(n), ln b ₃ +ln(n),  , ln b _i +ln(n)).

In this embodiment, the described first voice feature includes filter banks feature, filter banks feature is more sensitive to the volume of sound, and volume has the same influence to each dimension of the feature vector of filter banks feature, by separating the described waiting By processing the voice amplitude feature of the voice, the features related to the volume in the voice signal can be extracted, and the second voice feature can be obtained by calculation, which can realize the improvement of the filter banks feature, and can reduce the difference of the voice feature caused by the different volume. The improved filter banks feature for speech processing can improve the performance of the network model.

An embodiment of the present invention also provides an elevator control method, the method comprising:

Receive the target voice input by the user in the elevator scenario;

Use the voice processing method described in the embodiment of the present invention to perform off-line intention recognition on the target voice to obtain first control information;

The elevator is controlled to perform the first operation corresponding to the first control information.

Among them, the elevator control method can be applied to the elevator control device in the elevator to control the elevator to go to a certain floor or cancel going to a certain floor, or to control the elevator to open or close the door. The elevator control device can also control the elevator to perform other operations. This embodiment of the invention does not limit this. The implementation manner of the elevator control device receiving the target voice input by the user may be as follows: the elevator control device receives the input voice as the target voice in response to receiving the wake-up word input by the user. The wake-up word can be set as required, for example, the wake-up word can be "Hello, elevator". Alternatively, the elevator control device may directly receive the input voice as the target voice.

In addition, the use of the voice processing method according to the embodiment of the present invention to perform off-line intent recognition on the target voice to obtain the first control information may be: extracting the first voice feature of the target voice; Performing voice amplitude feature separation processing to obtain a second voice feature; and acquiring the first control information based on the second voice feature.

It should be noted that acquiring the first control information based on the second voice feature may be performing offline intent recognition based on the second voice feature to obtain the first control information. Specifically, the second The voice feature is input to a network model for offline intent recognition to obtain first control information. There are two ways to obtain the first control information by performing offline intent recognition based on the second voice feature. The first way is to obtain the first control information directly according to the second voice feature, and the second way is to obtain the first control information based on the second voice feature. The feature converts the target speech into text, and obtains the first control information according to the text.

For example, in a scenario where elevator control is performed on an elevator, the first control information may include a control instruction for controlling the elevator and a floor corresponding to the control instruction, wherein the control instruction may include a confirmation instruction for controlling the elevator to go to a certain floor, It can also include a cancel command for canceling the operation of the elevator going to a certain floor. It should be noted that the foregoing first control information is an exemplary description, and the first control information may be changed according to an application scenario, which is not limited in this embodiment of the present invention.

In the first mode, the elevator control device can pre-store a voice command vocabulary, and the voice command vocabulary is used to store a plurality of voice command words, and can store the voice characteristics of each voice command word, and one voice command word corresponds to an intent message. The implementation of the offline intention recognition based on the second voice feature to obtain the first control information may be: selecting a voice command word corresponding to the voice feature with the highest similarity of the second voice feature from the voice command vocabulary, The intent information corresponding to the command word is used as the first control information.

In the second manner, the elevator control device may store a text command thesaurus, the text command thesaurus is used to store a plurality of text command words, and one text command word corresponds to one piece of intention information. The offline intention recognition based on the second voice feature to obtain the first control information may be implemented by: obtaining the first text corresponding to the second voice feature, and selecting the text with the highest similarity to the first text from the text command vocabulary For the command word, the intention information corresponding to the text command word is used as the first control information.

In the prior art, online speech recognition needs to be transmitted through the network, the response speed is slow, and the online speech recognition is easily affected by the quality of the network, and in the case of poor network quality, the delay of the response is relatively large, thereby causing the efficiency of elevator control to be low. .

In the embodiment of the present invention, the target voice input by the user in the scenario of using the elevator is received; the off-line intention recognition of the target voice is performed by using the voice processing method described in the embodiment of the present invention, and the first control information is obtained; the elevator is controlled to execute the The first operation corresponding to the first control information. The offline intent recognition is performed by the speech processing method of the embodiment of the present invention, which can improve the recognition efficiency. Since the offline recognition generally has a faster response speed than the online recognition, the first control information is obtained by performing the offline intent recognition, and the corresponding first control information is executed. The first operation ensures the response speed of the elevator control, thereby improving the efficiency of the elevator control. Applying this method to the scene of controlling an elevator can greatly improve the efficiency of elevator startup and operation, and improve user stickiness.

Optionally, the method further includes:

sending the target speech to a server, so that the server performs online intent recognition on the target speech;

receiving the second control information sent by the server;

If the second control information is inconsistent with the first control information, control the elevator to cancel the execution of the first operation, and execute the second operation corresponding to the second control information.

The second control information is the same as the first control information, and details are not described here. The server performs online intention recognition on the target speech, and can obtain the second control information. The implementation manner of obtaining the second control information may be performed with the elevator control device for offline intention identification, and the implementation manner of obtaining the first control information is the same, and will not be repeated here. It should be noted that since the speech command thesaurus and the text command thesaurus for online intent recognition are stored in the cloud, the sample data therein is more abundant, and the success rate and accuracy of speech recognition are high. For example, the voice command thesaurus can be used to control the elevator, the voice command thesaurus for online intent recognition can include the voice command word "go to the restaurant", and the intent information corresponding to "go to the restaurant" can be "confirm instruction - 3rd floor", so the elevator Controls are smarter.

If it is determined that the second control information is consistent with the first control information, the elevator control device may ignore the second control information and continue to control the elevator to perform the first operation. If the elevator control device determines that the second control information is inconsistent with the first control information, it can directly control the elevator to cancel the execution of the first operation and execute the second operation corresponding to the second control information. The method is simple and efficient.

Taking the first control information as "confirm command-3rd floor" as an example, the corresponding first operation is to go to the 3rd floor, the second control information can be "confirm command-5th floor", and the corresponding second operation is to go to the 5th floor, Then the elevator can directly cancel the operation to go to the 3rd floor and go to the 5th floor.

In this embodiment, online intention identification is performed through the server. Since the accuracy of online identification is generally higher than that of offline identification, if the second control information is inconsistent with the first control information, the elevator is controlled to cancel the first operation and execute the The second operation corresponding to the second control information ensures the accuracy of elevator control.

Referring to FIG. 3, FIG. 3 is a schematic structural diagram of a speech processing apparatus provided by an embodiment of the present invention. As shown in FIG. 3, the speech processing apparatus 200 includes:

The extraction module 201 is used for extracting the first speech feature of the speech to be processed;

A separation module 202, configured to perform speech amplitude feature separation processing on the first speech feature to obtain a second speech feature;

The obtaining module 203 is configured to obtain the processing result of the voice to be processed based on the second voice feature.

Optionally, the first voice feature includes multiple first feature values, and the second voice feature includes multiple second feature values;

The separation module 202 is specifically used for:

Performing feature averaging processing on the plurality of first feature values to obtain a voice amplitude feature value, where the voice amplitude feature value is used to characterize the voice amplitude feature of the to-be-processed voice;

Based on the speech amplitude feature values, the speech amplitude feature separation processing is performed on each of the first feature values respectively, and the second features corresponding to each of the first feature values in the second speech features are obtained. value.

Optionally, the second speech feature further includes the speech amplitude feature value.

Optionally, the dimension of the second speech feature is greater than or equal to the dimension of the first speech feature.

Optionally, the first voice feature includes filter banks filter bank feature.

The speech processing apparatus can implement each process implemented in the method embodiment of FIG. 1 , and to avoid repetition, details are not repeated here.

Referring to FIG. 4, FIG. 4 is one of the structural schematic diagrams of an elevator control device provided by an embodiment of the present invention. As shown in FIG. 4, the elevator control device 300 includes:

The first receiving module 301 is used to receive the target voice input by the user in the scenario of using the elevator;

A recognition module 302, configured to perform off-line intention recognition on the target voice by using the voice processing method described in the embodiment of the present invention, to obtain first control information;

The first control module 303 is configured to control the elevator to perform the first operation corresponding to the first control information.

Optionally, as shown in FIG. 5 , the elevator control device 300 further includes:

A sending module 304, configured to send the target voice to a server, so that the server performs online intention recognition on the target voice;

A second receiving module 305, configured to receive the second control information sent by the server;

The second control module 306 is configured to control the elevator to cancel the execution of the first operation and execute the second operation corresponding to the second control information if the second control information is inconsistent with the first control information .

The elevator control device can implement each process implemented in the elevator control method in the embodiment of the present invention, and in order to avoid repetition, details are not repeated here.

In this embodiment of the present invention, electronic devices include but are not limited to mobile phones, tablet computers, notebook computers, palmtop computers, vehicle-mounted mobile terminals, wearable devices, elevators, and the like.

Please refer to FIG. 6. FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. As shown in FIG. 6, the electronic device 400 includes: a memory 402, a processor 401, and a memory 402 and a processor 401 stored on the memory 402 and available in the The program running on the processor 401, wherein:

As an implementation manner, the processor 401 reads the program in the memory 402 to execute:

extracting the first speech feature of the speech to be processed;

Performing voice amplitude feature separation processing on the first voice feature to obtain a second voice feature;

The processing result of the to-be-processed speech is acquired based on the second speech feature.

Optionally, the first voice feature includes multiple first feature values, and the second voice feature includes multiple second feature values;

The processor 401 is configured to perform speech amplitude feature separation processing on the first speech feature to obtain a second speech feature, including:

Performing feature averaging processing on the plurality of first feature values to obtain a voice amplitude feature value, where the voice amplitude feature value is used to characterize the voice amplitude feature of the to-be-processed voice;

Based on the speech amplitude feature values, the speech amplitude feature separation processing is performed on each of the first feature values respectively, and the second features corresponding to each of the first feature values in the second speech features are obtained. value.

Optionally, the second speech feature further includes the speech amplitude feature value.

Optionally, the dimension of the second speech feature is greater than or equal to the dimension of the first speech feature.

Optionally, the first voice feature includes filter banks filter bank feature.

As another implementation manner, the processor 401 reads the program in the memory 402 to execute:

Receive the target voice input by the user in the elevator scenario;

Use the voice processing method described in the embodiment of the present invention to perform off-line intention recognition on the target voice to obtain first control information;

The elevator is controlled to perform the first operation corresponding to the first control information.

Optionally, the processor 401 is further configured to execute:

sending the target speech to a server, so that the server performs online intent recognition on the target speech;

receiving the second control information sent by the server;

If the second control information is inconsistent with the first control information, control the elevator to cancel the execution of the first operation, and execute the second operation corresponding to the second control information.

In FIG. 6, the bus architecture may include any number of interconnected buses and bridges, in particular one or more processors represented by processor 401 and various circuits of memory represented by memory 402 linked together. The bus architecture may also link together various other circuits, such as peripherals, voltage regulators, and power management circuits, which are well known in the art and, therefore, will not be described further herein. The bus interface provides the interface.

The processor 401 is responsible for managing the bus architecture and general processing, and the memory 402 may store data used by the processor 401 in performing operations.

It should be noted that any implementation manner in the method embodiment of the embodiment of the present invention can be implemented by the electronic device in this embodiment, and achieve the same beneficial effects, which will not be repeated here.

Embodiments of the present invention further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, each process of the foregoing speech processing method embodiment is implemented, or the computer program is When the processor executes, each process of the above-mentioned embodiments of the elevator control method is implemented, and the same technical effect can be achieved. In order to avoid repetition, details are not repeated here. The computer-readable storage medium is, for example, a read-only memory (Read-Only Memory, ROM for short), a random access memory (Random Access Memory, RAM for short), a magnetic disk, or an optical disk.

It should be noted that, herein, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements, It also includes other elements not expressly listed or inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

From the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is better implementation. Based on this understanding, the technical solutions of the present invention can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products are stored in a storage medium (such as ROM/RAM, magnetic disk, CD-ROM), including several instructions to make a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present invention.

The embodiments of the present invention have been described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments, which are merely illustrative rather than restrictive. Under the inspiration of the present invention, without departing from the spirit of the present invention and the scope protected by the claims, many forms can be made, which all belong to the protection of the present invention.

Claims (10)

1. A method of speech processing, the method comprising:

extracting a first voice feature of the voice to be processed;

carrying out voice amplitude characteristic separation processing on the first voice characteristic to obtain a second voice characteristic;

and acquiring a processing result of the voice to be processed based on the second voice characteristic.

2. The method of claim 1, wherein the first speech feature comprises a plurality of first feature values and the second speech feature comprises a plurality of second feature values;

the performing speech amplitude feature separation processing on the first speech feature to obtain a second speech feature includes:

carrying out feature average processing on the plurality of first feature values to obtain a voice amplitude feature value, wherein the voice amplitude feature value is used for representing the voice amplitude feature of the voice to be processed;

and respectively carrying out voice amplitude characteristic separation processing on each first characteristic value based on the voice amplitude characteristic value to obtain second characteristic values corresponding to each first characteristic value in the second voice characteristic.

3. The method of claim 2, wherein the second speech feature further comprises the speech magnitude feature value.

4. The method of claim 1, wherein the dimension of the second speech feature is greater than or equal to the dimension of the first speech feature.

5. The method of claim 1, wherein the first speech features comprise filter banks features.

6. An elevator control method, characterized in that the method comprises:

receiving target voice input by a user in a scene of using an elevator;

performing offline intention recognition on the target voice by adopting the voice processing method of any one of claims 1 to 5 to obtain first control information;

and controlling the elevator to execute a first operation corresponding to the first control information.

7. The method of claim 6, further comprising:

sending the target voice to a server so that the server performs online intention recognition on the target voice;

receiving second control information sent by the server;

and if the second control information is inconsistent with the first control information, controlling the elevator to cancel the execution of the first operation and executing a second operation corresponding to the second control information.

8. A speech processing apparatus, characterized in that the speech processing apparatus comprises:

the extraction module is used for extracting a first voice feature of the voice to be processed;

the separation module is used for carrying out voice amplitude characteristic separation processing on the first voice characteristic to obtain a second voice characteristic;

and the acquisition module is used for acquiring the processing result of the voice to be processed based on the second voice characteristic.

9. An elevator control device, characterized by comprising:

the first receiving module is used for receiving target voice input by a user in an elevator using scene;

the recognition module is used for performing offline intention recognition on the target voice by adopting the voice processing method of any one of claims 1 to 5 to obtain first control information;

and the first control module is used for controlling the elevator to execute the first operation corresponding to the first control information.

10. An electronic device, comprising: a memory, a processor and a program stored on the memory and executable on the processor, the program, when executed by the processor, implementing the steps in the speech processing method according to any of claims 1 to 5; alternatively, the program realizes the steps in the elevator control method according to any one of claims 6 to 7 when executed by the processor.

Images