Noise recognition method and device, equipment and storage medium

Abstract

The invention relates to the field of voice recognition, in particular to a noise recognition method and device, equipment and a storage medium, and the method comprises the steps: obtaining voice data; inputting the time domain feature and the frequency domain feature corresponding to the voice data into a first model to obtain a noise recognition result output by the first model; the noise recognition result indicates frequency distribution of noise in the voice data; and processing the voice data according to the frequency distribution of the noise in the voice data. According to the method, the noise recognition precision can be improved, and the voice recognition result is improved.

Description

Technical Field

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

[0002] With the development of artificial intelligence, speech recognition technology has been widely used in many fields. However, in practical applications, speech data is often interfered by various background noises, such as traffic noise, wind noise, etc. These noises will significantly reduce the clarity of speech data and increase the difficulty of speech recognition.

[0003] At present, most existing noise recognition methods have low noise recognition accuracy when processing speech data, which affects the speech recognition results. Summary of the invention

[0004] The present application provides a noise recognition method, apparatus, device and storage medium for improving noise recognition accuracy.

[0005] In a first aspect, an embodiment of the present application provides a noise recognition method, the method comprising:

[0006] Get voice data;

[0007] Inputting the time domain features and frequency domain features corresponding to the speech data into a first model to obtain a noise recognition result output by the first model; the noise recognition result indicates the frequency distribution of noise in the speech data;

[0008] The speech data is processed according to the frequency distribution of noise in the speech data.

[0009] In an embodiment of the present invention, the time domain features and frequency domain features in the speech data are input into the first model, and the first model is used to perform in-depth analysis on the time domain features and frequency domain features, which can more accurately distinguish and identify different noise types in the speech data, provide the frequency distribution of the noise, and significantly improve the effect of noise recognition.

[0010] In a possible embodiment, a spectrum analysis result corresponding to the speech data is determined based on the speech data; time domain features and frequency domain features of multiple random windows are acquired in chronological order and at a preset step size based on the spectrum analysis result; wherein the time domain features corresponding to the speech data include statistical results of the time domain features of the multiple random windows; and the frequency domain features corresponding to the speech data include statistical results of the frequency domain features under the multiple random windows.

[0011] In this embodiment, by performing spectrum analysis on the speech data, time domain features can be obtained; by Fourier transform, the time domain is converted to the frequency domain to obtain frequency domain features; and these two features are combined to form a more comprehensive feature description of the speech data. In addition, by obtaining multiple random windows in time sequence and preset step size, the features of the speech data in different time periods can be analyzed in more detail.

[0012] In a possible embodiment, the time domain characteristics of the multiple random windows include variance and / or peak value; the frequency domain characteristics of the multiple random windows include frequency spectrum and / or power spectrum.

[0013] In this embodiment, the variance and / or peak value are selected as time domain features, which can reflect the volatility and extreme value of the speech signal, and help to identify the sudden changes and unstable parts in the speech signal; the frequency spectrum and / or power spectrum are selected as frequency domain features, which can reflect the energy distribution of the signal at different frequencies and improve the accuracy of noise recognition.

[0014] In a possible embodiment, historical call data is acquired; noise data in the historical call data is determined based on the historical call data; and the first model is trained based on time domain features and frequency domain features corresponding to the noise data.

[0015] In this embodiment, the DNN model is trained by using real historical call records, so that the DNN model can learn the real characteristics of noise in the actual call environment, which helps to train a first model with higher recognition accuracy.

[0016] In a possible embodiment, the noise recognition result further indicates the type of noise included in the voice data; the type of noise includes one or more of wind noise, whistle noise or electric current noise.

[0017] In this embodiment, the noise identification result is a frequency distribution, wherein different types of noise have unique frequency distribution characteristics, and these frequency distribution characteristics can help effectively distinguish the noise types.

[0018] In a second aspect, the present application further provides a noise identification device, the device comprising: a communication unit and a processing unit;

[0019] The communication unit is used to obtain voice data;

[0020] The processing unit is used to input the time domain features and frequency domain features corresponding to the speech data into a first model to obtain a noise recognition result output by the first model; the noise recognition result indicates the frequency distribution of the noise in the speech data; and is also used to process the speech data according to the frequency distribution of the noise in the speech data.

[0021] In a possible embodiment, the processing unit is used to:

[0022] When determining the spectrum analysis result corresponding to the speech data according to the speech data, obtaining time domain features and frequency domain features of multiple random windows according to the spectrum analysis result in chronological order and a preset step size;

[0023] The time domain features corresponding to the speech data include statistical results of the time domain features of the multiple random windows;

[0024] The frequency domain features corresponding to the speech data include statistical results of the frequency domain features under the multiple random windows.

[0025] In a possible embodiment, the time domain characteristics of the multiple random windows include variance and / or peak value; the frequency domain characteristics of the multiple random windows include frequency spectrum and / or power spectrum.

[0026] In a possible embodiment, the communication unit is used to:

[0027] Get historical call data;

[0028] The processing unit is further used for:

[0029] Determine noise data in the historical call data according to the historical call data;

[0030] The first model is trained according to the time domain features and frequency domain features corresponding to the noise data.

[0031] In a possible embodiment, the noise recognition result further indicates the type of noise included in the voice data; the type of noise includes one or more of wind noise, whistle noise or electric current noise.

[0032] In a third aspect, the present application provides a noise identification device, including:

[0033] A memory for storing program instructions;

[0034] The processor is used to call the program instructions stored in the memory, and execute the steps included in any one of the methods in the first aspect according to the obtained program instructions.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a computer, the computer executes any one of the methods described in the first aspect.

[0036] In a fifth aspect, the present application provides a computer program product, comprising: a computer program code, when the computer program code is run on a computer, the computer executes any one of the methods described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flowchart of a noise identification method provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of a first model for obtaining a noise recognition result provided by an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of the structure of the noise recognition device provided in the embodiment of the present application Figure 1 ;

[0041] Figure 4 A schematic diagram of the structure of the noise recognition device provided in the embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0042] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] In order to improve the accuracy of noise recognition, Figure 1 As shown, an embodiment of the present invention provides a method for noise recognition, the execution subject of the method may be a server or a processor, and the following description is made taking a server as an example. The method specifically includes:

[0044] Step 100: The server obtains voice data.

[0045] In a specific implementation, voice data is collected over a period of time and cut into multiple segments. For example, the length of each segment can be set between 5 and 10 seconds.

[0046] Step 110: The server inputs the time domain features and frequency domain features corresponding to the speech data into the first model to obtain the noise recognition result output by the first model.

[0047] The noise recognition result indicates the frequency distribution of the noise in the speech data.

[0048] Exemplarily, before the time domain features and frequency domain features corresponding to the speech data are input into the first model, the first model is first trained.

[0049] Exemplarily, the first model can be trained in the following manner but not limited to:

[0050] 1. Obtain historical call data. For example, historical call data over a long period of time is collected.

[0051] 2. Determine the noise data in the historical call data based on the historical call data. For example, the technician can retain the data with a small proportion of human voice but a significant proportion of noise based on the historical call data as the noise data. For example, the noise data is saved in a waveform audio file format (WAV).

[0052] 3. Train the first model based on the time domain features and frequency domain features corresponding to the noise data.

[0053] The time domain features and frequency domain features of the noise data are calculated. The calculation process is consistent with the process of calculating the time domain features and frequency domain features of the speech data described below, and will not be described in detail here.

[0054] The calculated time domain features and frequency domain features are input into a deep neural network (DNN) for training, wherein DNN is a basic deep learning model composed of multiple layers of neurons and connection weights, including multiple hidden layers, activation functions, pooling layers, etc.; during the training process, the mean absolute error (MAE) is used as the loss function to improve the performance of the model; at the same time, parameters such as the learning rate, batch size, and number of training rounds are continuously adjusted to optimize the model, and through multiple iterative training, a first model with higher noise recognition accuracy can be obtained, wherein the output result of the first model can include the frequency distribution of various noises in the noise data.

[0055] After obtaining the trained first model, the time domain features and frequency domain features corresponding to the speech data may be input into the first model.

[0056] In a specific implementation, the time domain features and frequency domain features corresponding to the speech data are first calculated.

[0057] First, the spectrum analysis result corresponding to the speech data is determined according to the speech data.

[0058] Second, according to the spectrum analysis results, the time domain features and frequency domain features of multiple random windows are obtained in chronological order and at a preset step size. The time domain features corresponding to the speech data include the statistical results of the time domain features of multiple random windows; the frequency domain features corresponding to the speech data include the statistical results of the frequency domain features under multiple random windows.

[0059] During the specific implementation process, in order to more comprehensively analyze the speech data, for the spectrum analysis results corresponding to the speech data, multiple random windows are extracted according to the time sequence and the preset step size. For example, the preset step size is 40 milliseconds and the window size is 40 milliseconds. Time domain features and frequency domain features are extracted for each window respectively.

[0060] In one possible design, in order to improve the temporal resolution, there can be a certain overlap between windows. For example, taking the preset step size of 20 milliseconds and the window size of 40 milliseconds as an example, window 1 starts at time 0 and lasts for 40 milliseconds; window 2 starts at time 10 milliseconds and lasts for 40 milliseconds, wherein window 2 overlaps with window 1 for 30 seconds; window 3 starts at time 20 milliseconds and lasts for 40 milliseconds, and window 3 overlaps with window 2 for 30 milliseconds, and so on. By overlapping, rapidly changing features can be better captured in time, wherein the present application does not limit the specific preset step size and window size.

[0061] In a specific implementation, the time domain characteristics of the multiple random windows include variance and / or peak value; the frequency domain characteristics of the multiple random windows include frequency spectrum and / or power spectrum.

[0062] The time domain features are obtained according to the spectrum analysis results of the speech data, and the time domain features include variances and / or peaks of multiple random windows.

[0063] Among them, variance is a measure of the dispersion of the signal waveform around its average value, which can reflect the dynamic range and stability of the speech signal. Specifically, the variance is calculated under each random window of the speech data, and the variances of multiple random windows are combined to obtain the statistical results of the variances under multiple random windows. Furthermore, the variances under each random window can be added and divided by the number of windows to obtain the average variance, which can better reflect the degree of fluctuation of the entire speech signal in the time domain.

[0064] The peak value is used to represent the local maximum value of the signal in the time domain, which usually corresponds to an important part of the speech data, such as sudden noise. The peak characteristics of noise are often different from the peak characteristics of human voice. Specifically, the peak value is calculated under each random window of the speech data, and the peak values ​​of each random window are combined to obtain the statistical results of the peak values ​​under multiple random windows.

[0065] Thus, all the statistical results of variance and / or peak value are combined to obtain the statistical results of time domain features under multiple random windows, thereby forming the time domain features of the speech data. The extraction of time domain features is usually implemented using the relevant processing library in Python.

[0066] Furthermore, the frequency domain features of the speech data are calculated.

[0067] In a specific implementation, the spectrum analysis results corresponding to the speech data are Fourier transformed from the time domain to the frequency domain to display the energy distribution of the speech data at different frequencies. The frequency domain features include frequency spectra and / or power spectra of multiple random windows.

[0068] Among them, the frequency spectrum is the amplitude of the Fourier transform result, which is used to represent the intensity of the signal at each frequency component, wherein the frequency spectrum includes positive frequencies and negative frequencies. Specifically, before calculating the frequency spectrum under each random window of the speech data, a window screening mechanism will be set to screen out windows whose frequency spectrum intensity exceeds the threshold. For example, the threshold is set to 0.2, and the speech data is divided into 10 windows. If the intensity of the frequency spectrum calculated by 6 windows is higher than 0.2, and the intensity of the frequency spectrum calculated by 4 windows is lower than 0.2, then the frequency spectra of these 6 windows are selected and combined to obtain the statistical results of all frequency spectra. This application does not limit the specific threshold parameters.

[0069] The power spectrum is the square of the frequency spectrum, which is used to represent the distribution of signal power as the frequency changes in a certain area, and the power spectrum is a non-negative value. Specifically, before calculating the power spectrum under each random window of the speech data, a window screening mechanism is also set to screen out windows whose power spectrum intensity exceeds the threshold, and the power spectra of the screened multiple windows are combined to obtain the statistical results of all power spectra. The specific screening process is the same as the above screening and will not be repeated in detail.

[0070] Thus, the statistical results of all the selected frequency spectra and / or power spectra are combined to obtain the statistical results of frequency domain features under multiple random windows, thereby forming the frequency domain features of the speech data. The extraction of frequency domain features is usually implemented using a related processing library in Python.

[0071] Third, if Figure 2 As shown, the time domain features and frequency domain features corresponding to the speech data are input into the first model to obtain a noise recognition result. The noise recognition result corresponds to the frequency distribution of the noise in the speech data. Optionally, the noise recognition result may also indicate the type of noise included in the speech data, and the type of noise may include one or more of wind noise, whistle or current noise.

[0072] In a specific implementation, since the DNN model has learned how to identify different types of noise based on input features during the training process, after the training is completed, when the time domain features and frequency domain features corresponding to the input voice data are input, the first model can output a frequency distribution, and the corresponding type of noise can be identified based on the frequency distribution. Among them, different types of noise have unique frequency distribution characteristics, and these frequency distribution characteristics can effectively distinguish the types of noise. For example, if the frequency distribution of the output result of the first model is relatively flat and distributed in a wider frequency range, the corresponding noise type is wind noise; if the frequency distribution of the output result of the first model has an obvious peak frequency and is transient, mainly concentrated in a certain frequency, then the corresponding noise type is whistle; if the frequency distribution of the output result of the first model shows a certain periodicity, then the corresponding noise type is current sound.

[0073] Step 120: The server processes the speech data according to the frequency distribution of noise in the speech data.

[0074] In a specific implementation, the noise position in the speech data is determined according to the frequency distribution of the noise in the speech data output by the first model, and the frequency distribution of the corresponding noise position is removed using a python tool, thereby clearly retaining the human voice data in the speech data.

[0075] Furthermore, the speech data with noise removed is converted into a WAV format with a sampling rate of 16kHz and input into the paraformer model. The transmitted speech data is recognized through real-time reasoning, and high-quality text data can be output, that is, high-quality speech recognition results can be obtained.

[0076] In summary, this application provides a noise recognition method to achieve effective recognition of noise in speech data. The time domain features and frequency domain features in the speech data are input into the first model, and the first model is used to perform in-depth analysis of the time domain features and frequency domain features, which can more accurately distinguish and identify different types of noise in the speech data, provide the frequency distribution of noise, and significantly improve the effect of noise recognition. By eliminating or reducing the interference of noise, the accuracy of speech-to-text conversion is improved.

[0077] Figure 3 and Figure 4 A schematic diagram of the structure of a possible noise recognition device provided in an embodiment of the present application. These noise recognition devices can be used to implement the functions of the server in the above method embodiment.

[0078] like Figure 3 As shown, the noise identification device 300 includes a communication unit 310 and a processing unit 320 .

[0079] When the noise identification device 300 is used to implement Figure 1 The functions of the server in the method embodiment shown are:

[0080] The communication unit 310 is used to obtain voice data;

[0081] The processing unit 320 is used to input the time domain features and frequency domain features corresponding to the speech data into the first model to obtain the noise recognition result output by the first model; the noise recognition result indicates the frequency distribution of the noise in the speech data; the processing unit 320 is also used to process the speech data according to the frequency distribution of the noise in the speech data.

[0082] In one possible design, the processing unit 320 is configured to:

[0083] When determining the spectrum analysis result corresponding to the speech data according to the speech data, obtaining time domain features and frequency domain features of multiple random windows according to the spectrum analysis result in chronological order and a preset step size;

[0084] The time domain features corresponding to the speech data include statistical results of the time domain features of the multiple random windows;

[0085] The frequency domain features corresponding to the speech data include statistical results of the frequency domain features under the multiple random windows.

[0086] In a possible design, the time domain characteristics of the multiple random windows include variance and / or peak value; the frequency domain characteristics of the multiple random windows include frequency spectrum and / or power spectrum.

[0087] In one possible design, the communication unit 310 is configured to:

[0088] Get historical call data;

[0089] The processing unit 320 is further configured to:

[0090] Determine noise data in the historical call data according to the historical call data;

[0091] The first model is trained according to the time domain features and frequency domain features corresponding to the noise data.

[0092] In one possible design, the noise recognition result also indicates the type of noise included in the voice data; the type of noise includes one or more of wind noise, whistle or current sound.

[0093] For a more detailed description of the communication unit 310 and the processing unit 320, please refer to Figure 1 The relevant description in the method embodiment shown is directly obtained and will not be repeated here.

[0094] like Figure 4 As shown, the noise recognition device 400 includes a processor 410 and a communication interface 420. The processor 410 and the communication interface 420 are coupled to each other. It is understandable that the communication interface 420 can be a transceiver or an input-output interface. Optionally, the noise recognition device 400 may also include a memory 430 for storing instructions executed by the processor 410 or storing input data required by the processor 410 to execute instructions or storing data generated after the processor 410 executes instructions.

[0095] When the noise identification device 400 is used to implement Figure 1 When the method is shown, the processor 410 is used to implement the functions of the above-mentioned processing unit 410, and the communication interface 420 is used to implement the functions of the above-mentioned communication unit 310.

[0096] The division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0097] Based on the same inventive concept, the embodiment of the present application provides a computer-readable storage medium, and the computer program product includes: computer program code, when the computer program code is executed on a computer, the computer executes any of the noise identification methods discussed above. Since the principle of solving the problem by the above-mentioned computer-readable storage medium is similar to that of the noise identification method, the implementation of the above-mentioned computer-readable storage medium can refer to the implementation of the method, and the repeated parts will not be repeated.

[0098] Based on the same inventive concept, the embodiment of the present application also provides a computer program product, which includes: computer program code, when the computer program code is run on a computer, the computer executes any of the noise identification methods discussed above. Since the principle of solving the problem by the above computer program product is similar to that of the noise identification method, the implementation of the above computer program product can refer to the implementation of the method, and the repeated parts will not be repeated.

[0099] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0100] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0101] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of user-operated steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0103] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0104] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A noise identification method, characterized in that: The method includes: Get voice data; Inputting the time domain features and frequency domain features corresponding to the speech data into a first model to obtain a noise recognition result output by the first model; the noise recognition result indicates the frequency distribution of noise in the speech data; The speech data is processed according to the frequency distribution of noise in the speech data.

2. The method according to claim 1, characterized in that Also includes: Determine, according to the voice data, a spectrum analysis result corresponding to the voice data; According to the spectrum analysis results, the time domain features and frequency domain features of multiple random windows are obtained in chronological order and according to a preset step size; The time domain features corresponding to the speech data include statistical results of the time domain features of the multiple random windows; The frequency domain features corresponding to the speech data include statistical results of the frequency domain features under the multiple random windows.

3. The method according to claim 2, characterized in that The time domain characteristics of the multiple random windows include variance and / or peak value; The frequency domain characteristics of the multiple random windows include frequency spectrum and / or power spectrum.

4. The method according to claim 1, characterized in that Also includes: Get historical call data; Determine noise data in the historical call data according to the historical call data; The first model is trained according to the time domain features and frequency domain features corresponding to the noise data.

5. The method according to claim 1, characterized in that The noise recognition result also indicates the type of noise included in the voice data; the type of noise includes one or more of wind noise, whistle noise or current noise.

6. A noise identification device, characterized in that: The device comprises: a communication unit and a processing unit; The communication unit is used to obtain voice data; The processing unit is used to input the time domain features and frequency domain features corresponding to the speech data into a first model to obtain a noise recognition result output by the first model; the noise recognition result indicates the frequency distribution of the noise in the speech data; and is also used to process the speech data according to the frequency distribution of the noise in the speech data.

7. The device according to claim 6, characterized in that The processing unit is used to: when determining the spectrum analysis result corresponding to the speech data according to the speech data, obtain the time domain features and frequency domain features of multiple random windows according to the spectrum analysis result in a chronological order and a preset step size; The time domain features corresponding to the speech data include statistical results of the time domain features of the multiple random windows; The frequency domain features corresponding to the speech data include statistical results of the frequency domain features under the multiple random windows.

8. The device according to claim 7, characterized in that The time domain characteristics of the multiple random windows include variance and / or peak value; the frequency domain characteristics of the multiple random windows include frequency spectrum and / or power spectrum.

9. The device according to claim 6, characterized in that The communication unit is further used to: obtain historical call data; The processing unit is further used to: determine the noise data in the historical call data according to the historical call data; and train the first model according to the time domain features and frequency domain features corresponding to the noise data.

10. The device according to claim 6, characterized in that The noise recognition result also indicates the type of noise included in the voice data; the type of noise includes one or more of wind noise, whistle noise or current noise.

11. A noise identification device, characterized in that: include: A memory for storing program instructions; A processor is used to call the program instructions stored in the memory, and execute the steps included in any one of the methods of claims 1-5 according to the obtained program instructions.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a computer, the computer executes the method according to any one of claims 1 to 5.

13. A computer program product, characterized in that The computer program product comprises: a computer program code, and when the computer program code is run on a computer, the computer is enabled to execute the method according to any one of claims 1 to 5.

Images