An audio processing method, apparatus, device and medium

Abstract

The application provides an audio processing method, device and equipment and medium. In the method, for the audio to be processed, the target audio category of the audio is determined, and then the audio to be confused is selected in the audio. The target confusion data corresponding to the audio to be confused is used to process the audio in the time interval corresponding to the audio to be confused. Because only part of the phonemes is processed, the audio can sound similar to the original audio, but the voiceprint feature contained in the audio has changed, so the confusion of the user's voiceprint feature is realized, and the user's voiceprint is protected.

Description

Technical Field

[0001] This application belongs to the field of audio, and in particular relates to an audio processing method, apparatus, device and medium. Background Technology

[0002] The audio functions of smart home voice interaction devices and real-time chat software have made life more convenient for users. However, as voiceprint recognition algorithms become increasingly sophisticated, there is a risk that user interaction audio may expose their voiceprints. Currently, it is possible to collect more than five audio clips from a user and, by recognizing the voiceprint of that audio, identify the user among tens of thousands of users, thus compromising user privacy.

[0003] Regarding voiceprint protection, existing solutions involve using a voice changer to convert the frequency of the user's audio, or converting the user's audio into text and then using a voice synthesis algorithm to convert the text back into audio. However, both of these methods require converting the entire audio recording after the user has recorded it, making real-time protection impossible. Furthermore, in scenarios like intercom or sending recordings, the receiving user hears only the content of the converted audio; the timbre has changed drastically. This can make the user feel like they are hearing a completely different person, failing to achieve a friendly interaction and resulting in a mechanical and impersonal interaction.

[0004] Therefore, there is an urgent need for a voiceprint protection method that makes the audio sound similar to the original audio, but the voiceprint has been altered. Summary of the Invention

[0005] This application provides an audio processing method, apparatus, device, and medium to solve the problem that existing audio processing methods cause significant changes in audio timbre, making it impossible to achieve user-friendly interaction.

[0006] In a first aspect, this application provides an audio processing method, the method comprising:

[0007] For the audio to be processed, a pre-trained recognition model is used to determine the target audio category of the audio.

[0008] Identify the starting position of each phoneme in the audio and select the phonemes to be confused;

[0009] In the pre-saved obfuscation data corresponding to each phoneme of each audio type, determine the target obfuscation data corresponding to the phoneme to be obfuscated of the target audio type;

[0010] The time interval in which the phoneme to be confused is located is determined in the audio, and the audio in the time interval is processed according to the target confusion data.

[0011] In one possible implementation, the obfuscated data is the frequency and energy values ​​of the fundamental frequency and overtones, and the processing of the audio within the time interval based on the target obfuscated data includes:

[0012] Filter the original fundamental frequency and overtones of the phonemes to be confused within the time interval of the audio;

[0013] The frequency and energy values ​​of the target fundamental frequency and target overtone are added to the time interval of the audio.

[0014] In one possible implementation, after determining the time interval where the phoneme to be confused is located in the audio, and before filtering the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval of the audio, the method further includes:

[0015] Determine the original energy of the fundamental frequency and overtones of the phoneme to be confused within the time interval;

[0016] Determine the target energy of the target fundamental frequency and the target overtone;

[0017] The gain attenuation coefficient is determined based on the ratio of the original energy to the target energy.

[0018] Adding the frequency and energy values ​​of the target fundamental frequency and target overtone to the time interval of the audio includes:

[0019] The energy values ​​of the target fundamental frequency and the target overtone are processed using the gain attenuation coefficient, and the frequencies of the target fundamental frequency and the target overtone, along with the processed energy values, are added to the time interval of the audio.

[0020] In one possible implementation, the obfuscated data is obfuscated audio, and after determining the time interval where the phoneme to be obfuscated is located in the audio, the processing of the audio within the time interval according to the target obfuscated data includes:

[0021] The target obfuscated audio is sampled and synthesized with the audio within the time interval.

[0022] In one possible implementation, determining the target audio category of the audio using a pre-trained recognition model includes:

[0023] Using a pre-trained first model, the age group and gender of the target user to whom the audio belongs are determined;

[0024] Using a pre-trained second model, the similarity between the audio and audios of various preset sound styles is determined; based on the similarity between the audio and audios of various preset sound styles, the target sound style of the audio is determined.

[0025] The age group, gender, and target voice style are determined as the target audio category of the audio.

[0026] In one possible implementation, determining the target sound style of the audio based on the similarity includes:

[0027] If there is a similarity between the audio and the preset audio of each sound style that falls within the range of a first similarity threshold and a second similarity threshold, then the sound style of the audio with the highest similarity within the range is determined as the target sound style; wherein the first similarity threshold is greater than the second similarity threshold;

[0028] If there is no similarity within the range, and all similarities are greater than the first similarity threshold, then the audio style corresponding to the smallest similarity is determined as the target audio style.

[0029] If there is no similarity within the specified range, and all similarities are less than the second similarity threshold, then the audio style corresponding to the highest similarity is determined as the target audio style.

[0030] In one possible implementation, selecting the phonemes to be confused includes:

[0031] The number of phonemes to be confused is determined based on the preset confusion ratio and the total number of phonemes appearing in the audio.

[0032] Based on the stated quantity and the total quantity, determine the number of phonemes contained in each group;

[0033] Based on the order in which the phonemes appear in the audio, the phonemes are grouped according to their number, and one phoneme from each group is randomly selected as the phoneme to be confused.

[0034] Secondly, this application provides an audio processing apparatus, the apparatus comprising:

[0035] The determination module is used to determine the target audio category of the audio to be processed using a pre-trained recognition model; identify the starting position of each phoneme in the audio and select the phoneme to be confused; and determine the target confusion data corresponding to the phoneme to be confused in the target audio category from the pre-saved confusion data corresponding to each phoneme of each audio category.

[0036] The processing module is used to determine the time interval in the audio where the phoneme to be confused is located, and to process the audio in the time interval according to the target confusion data.

[0037] In one possible implementation, the processing module is specifically used to filter the original fundamental frequency and overtones corresponding to the phonemes to be confused within the time interval of the audio; and to add the frequency and energy values ​​of the target fundamental frequency and target overtones to the time interval of the audio.

[0038] In one possible implementation, the determining module is specifically used to determine the original energy sum of the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval; determine the target energy sum of the target fundamental frequency and target overtones; determine a gain attenuation coefficient based on the ratio of the original energy sum to the target energy sum; and add the frequency and energy value of the target fundamental frequency and target overtones to the time interval of the audio, which includes: processing the energy value of the target fundamental frequency and target overtones using the gain attenuation coefficient, and adding the frequency and processed energy value of the target fundamental frequency and target overtones to the time interval of the audio.

[0039] In one possible implementation, the processing module is specifically configured to sample and synthesize the target obfuscated audio with audio within the time interval.

[0040] In one possible implementation, the determining module is specifically used to: use a pre-trained first model to determine the age group and gender of the target user to whom the audio belongs; use a pre-trained second model to determine the similarity between the audio and audios of various preset voice styles; determine the target voice style of the audio based on the similarity with audios of various preset voice styles; and determine the age group, gender, and target voice style as the target audio category of the audio.

[0041] In one possible implementation, the determining module is specifically configured to: if there is a similarity between the audio samples with preset similarity to each audio sample of a given sound style and a first similarity threshold and a second similarity threshold, then determine the sound style of the audio sample with the highest similarity within the range as the target sound style; wherein the first similarity threshold is greater than the second similarity threshold; if there is no similarity within the range and all similarities are greater than the first similarity threshold, then determine the sound style of the audio sample with the lowest similarity as the target sound style; if there is no similarity within the range and all similarities are less than the second similarity threshold, then determine the sound style of the audio sample with the highest similarity as the target sound style.

[0042] In one possible implementation, the determining module is specifically used to determine the number of selected phonemes to be confused based on a preset confusion ratio and the total number of phonemes appearing in the audio; determine the number of phonemes included in each group based on the number and the total number; group the phonemes according to the number of phonemes based on the order of appearance of the phonemes in the audio, and randomly select one phoneme from each group as the phoneme to be confused.

[0043] Thirdly, this application also provides an electronic device, which includes at least a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the steps of any of the audio processing methods described above.

[0044] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the audio processing methods described above.

[0045] In this application, after determining the target audio type for the audio to be processed, the phonemes to be confused are selected in the audio. The target confusion data corresponding to the phonemes to be confused is used to process the audio within the time interval corresponding to the phonemes to be confused. Since only some phonemes are processed, it can be ensured that the audio sounds similar to the original audio. However, since the voiceprint features contained in the audio have changed, the confusion of the user's voiceprint features is achieved, thus protecting the user's voiceprint. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of an audio processing method provided in an embodiment of this application;

[0048] Figure 2 This application provides an example of an audio diagram after phoneme recognition.

[0049] Figure 3 A phoneme spectrum feature map of a user's original audio provided in an embodiment of this application;

[0050] Figure 4 A schematic diagram of the frequency response of an FIR bandpass filter provided in an embodiment of this application;

[0051] Figure 5a A filtered phoneme spectrum feature map provided in an embodiment of this application;

[0052] Figure 5b This is one of the schematic diagrams of an audio processing procedure provided in an embodiment of this application;

[0053] Figure 6 A frequency and energy spectrum diagram of the target fundamental frequency and target overtone corresponding to the phoneme to be confused, provided in an embodiment of this application;

[0054] Figure 7a A spectrum diagram of synthesized phonemes corresponding to a time interval is provided in an embodiment of this application;

[0055] Figure 7b This is a second schematic diagram of an audio processing procedure provided in an embodiment of this application;

[0056] Figure 8 This is a schematic diagram of the structure of an audio processing device provided in an embodiment of this application;

[0057] Figure 9 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0060] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0061] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0062] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0063] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0064] This application provides an audio processing method, apparatus, device, and medium. The method includes: for an audio to be processed, using a pre-trained recognition model to determine the target audio category of the audio; identifying the starting position of each phoneme in the audio and selecting a phoneme to be confused; determining the target confusion data corresponding to the phoneme to be confused in the target audio category from pre-saved confusion data corresponding to each phoneme of each audio category; determining the time interval in the audio where the phoneme to be confused is located, and processing the audio within the time interval according to the target confusion data.

[0065] In order to protect the user's voiceprint and make the altered audio sound similar to the original audio, this application provides an audio processing method, apparatus, device and medium.

[0066] Figure 1 This is a schematic diagram illustrating an audio processing method provided in an embodiment of this application. Figure 1 As shown, the process includes the following steps:

[0067] S101: For the audio to be processed, a pre-trained recognition model is used to determine the target audio type of the audio.

[0068] The audio processing method provided in this embodiment is applied to electronic devices, such as mobile phones, PCs, tablets, speakers, smart TVs, and other devices with audio processing capabilities.

[0069] The key aspects of voiceprint recognition include feature extraction and pattern recognition. Feature extraction involves processing user audio files to obtain the voiceprint features hidden in the sound; pattern recognition involves matching the voiceprint features with a pre-learned feature library, identifying a matching feature based on similarity or clustering algorithms, and thus identifying the user's identity.

[0070] Voiceprint features are primarily the acoustic characteristics of the human vocalization mechanism, such as the frequency spectrum, cepstral spectrum, formants, fundamental tone, and reflection coefficient. Changing one or two of these five acoustic features can alter the voiceprint characteristics. However, directly changing these features may result in the altered user audio sounding unlike the original user audio, or even disrupt the original user audio's voice interaction content, making accurate voice interaction recognition impossible.

[0071] To make the modified audio sound more like the original audio, the target audio category needs to be determined first. One audio category identifies a type of user. The audio category can identify the user from one dimension, such as the user's age or gender. However, to identify the user more clearly and accurately, the audio category can be identified from the perspective of the user's gender, age group, and voice style.

[0072] Once the target audio category is determined, the audio is then processed accordingly to avoid mixing male and female voices, or children's and elderly people's voices, which would result in a poor user experience.

[0073] S102: Identify the starting position of each phoneme in the audio and select the phonemes to be confused.

[0074] A phoneme is the smallest unit of speech defined based on the natural attributes of speech. It is analyzed based on the articulation actions within a syllable; one action constitutes one phoneme. To make altered audio sound similar to the original while simultaneously altering voiceprint features, the audio content is broken down into phonemes. Several phonemes or a set number of phonemes can be randomly selected as the phonemes to be confused. Processing a feature of the selected phonemes can preserve most of the user's voice interaction features, such as timbre and background noise, making the processed audio sound more like the original. However, because some phoneme features are transformed, when a voiceprint recognition algorithm processes the audio, it cannot extract features exactly the same as the original audio, thus failing to accurately locate the user and protect their voiceprint.

[0075] Specifically, identifying the starting position of each phoneme in the audio can be achieved using methods such as learnable segment feature phoneme boundary detection, syllable / vowel / phoneme boundary detection based on auditory attention cues, phoneme boundary detection based on phoneme posterior probability and hierarchical agglomerative clustering algorithms, or phoneme boundary detection based on spectral energy. This process is existing technology and is not limited in this application.

[0076] Figure 2This is a schematic diagram of an audio signal after phoneme recognition, provided in an embodiment of this application. The horizontal axis represents time, and the dashed lines represent the starting position of each phoneme. Figure 2 As shown, after processing, the starting positions of different phonemes in the entire audio file can be obtained, and the phonemes to be confused can be selected from them. Specifically, the learnable segment feature phoneme boundary detection method in the above method can be used to realize phoneme boundary detection, identify the starting position of each phoneme in the audio, and mark a timestamp at the starting position of each phoneme, that is, the time point at which the phoneme appears in the audio.

[0077] S103: Determine the target obfuscation data corresponding to the phoneme to be obfuscated for the target audio type from the pre-saved obfuscation data corresponding to each phoneme of each audio type.

[0078] To achieve voiceprint protection, in this embodiment, obfuscation data corresponding to each phoneme of each audio type is pre-stored, including but not limited to: obfuscated audio corresponding to each phoneme of each audio type, and obfuscated audio of different lengths can be stored for each phoneme. Based on the target audio type and the selected phoneme to be obfuscated, the target obfuscation data corresponding to the phoneme to be obfuscated in that target audio type can be determined.

[0079] S104: Determine the time interval where the phoneme to be confused is located in the audio, and process the audio in the time interval according to the target confusion data.

[0080] Once the target obfuscation data corresponding to the phonemes to be obfuscated for the target audio type is determined, this target obfuscation data can be used to replace the feature information corresponding to that phoneme in the audio. First, the phonemes to be obfuscated can be identified in the audio.

[0081] Because the above detection method can detect the starting position of each phoneme in the audio and mark the starting position of each phoneme with a timestamp, the time interval of the phoneme to be confused in the audio can be determined. This time interval is from the starting position of the phoneme to be confused to the starting position of the next phoneme in the audio.

[0082] The audio within the time interval is processed based on the target obfuscation data, that is, the phonemes to be obfuscated within the time interval are adjusted to achieve obfuscation of the user's voiceprint features.

[0083] In this embodiment of the application, after determining the target audio type for the audio to be processed, the phonemes to be confused are selected in the audio, and the target confusion data corresponding to the phonemes to be confused are used to process the audio in the time interval corresponding to the phonemes to be confused. Since only some phonemes are processed, it is possible to ensure that the audio sounds similar to the original audio, but because the voiceprint features contained in the audio have changed, the confusion of the user's voiceprint features is achieved, thus protecting the user's voiceprint.

[0084] To preserve as much ambient noise and user vocal characteristics as possible during user interaction, based on the above embodiments, in this embodiment, the obfuscated data consists of the frequency and energy values ​​of the fundamental frequency and overtones. The processing of the audio within the time interval based on the target obfuscated data includes:

[0085] Filter the original fundamental frequency and overtones of the phonemes to be confused within the time interval of the audio; add the frequency and energy values ​​of the target fundamental frequency and target overtones to the time interval of the audio.

[0086] Because users have different acoustic structures, such as the thickness and shape of their vocal cords, the frequency of their vocalizations will vary. Figure 3 A phoneme spectrum feature map of a user's original audio provided in this application embodiment, such as Figure 3 As shown, the horizontal axis represents frequency in Hz, and the vertical axis represents energy value in dB. The spectral characteristics of a phoneme include the fundamental frequency and overtones, as well as their energy values. Overtones encompass multiple frequency bands, including the first to sixth overtones, where the frequencies of the fundamental frequency to the sixth overtone increase exponentially.

[0087] Specifically, such as Figure 3 As shown, the fundamental frequency is 632Hz with an energy value of 27dB; the first overtone has a frequency of 1264Hz with an energy value of 34dB; the second overtone has a frequency of 1896Hz with an energy value of 30dB; the third overtone has a frequency of 2528Hz with an energy value of 31dB; the fourth overtone has a frequency of 3160Hz with an energy value of 26dB; the fifth overtone has a frequency of 3792Hz with an energy value of 33dB; and the sixth overtone has a frequency of 4424Hz with an energy value of 29dB.

[0088] This application embodiment processes the frequencies of the first to sixth overtones in the key fundamental frequency and overtones, i.e., it processes the fundamental frequency and overtones. It filters the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval of the audio, that is, it filters the original fundamental frequency and the frequencies of the first to sixth overtones corresponding to the phoneme to be confused within the time interval of the audio that needs to be processed. The spectrum diagram corresponding to the audio can be obtained using an oscilloscope or spectrum analyzer, and then the original fundamental frequency and the frequencies of the first to sixth overtones of the phoneme to be confused can be determined based on the time interval of the phoneme in the audio. Specifically, a finite impulse response (FIR) bandpass filter from digital signal processing can be used to filter out the original fundamental frequency and the frequencies of the first to sixth overtones.

[0089] Figure 4 A schematic diagram of the frequency response of an FIR bandpass filter provided in an embodiment of this application is shown below. Figure 4 As shown in the figure, the horizontal axis represents frequency and the vertical axis represents energy value. After passing through this FIR bandpass filter, the frequencies in the following bands are filtered out: 568Hz-695 Hz, 1137Hz-1390 Hz, 1706Hz-2085 Hz, 2275Hz-2780 Hz, 2844Hz-3476 Hz, 3412Hz-4171 Hz, and 3981Hz-4866 Hz.

[0090] Combination Figure 3 As shown, it has been filtered out Figure 3 The spectral characteristics of the fundamental frequency and the frequencies of the first to sixth overtones of the phonemes are shown in the figure below. Figure 5a As shown, Figure 5a This is a filtered phoneme spectrum feature map provided in an embodiment of this application. Figure 5a The horizontal axis represents frequency, and the vertical axis represents energy value.

[0091] The frequency and energy values ​​of the target fundamental frequency and target overtone corresponding to the target audio type to be confused are added to the time interval of the target audio to be confused.

[0092] Figure 5b This is one of the schematic diagrams of an audio processing procedure provided in an embodiment of this application. For example... Figure 5b As shown, the process includes the following steps:

[0093] S501: For the audio to be processed, a pre-trained recognition model is used to determine the target audio type of the audio.

[0094] S502: Identify the starting position of each phoneme in the audio and select the phonemes to be confused.

[0095] S503: Determine the target fundamental frequency and target overtone frequency and energy value of the target audio type's phoneme to be confused from the pre-saved fundamental frequency and overtone frequency and energy value of each phoneme corresponding to each audio type.

[0096] S504: Filter the original fundamental frequency and overtones corresponding to the phonemes to be confused within the time interval of the audio; add the frequency and energy values ​​of the target fundamental frequency and target overtones to the time interval of the audio.

[0097] In this embodiment of the application, the obfuscated data is the frequency and energy value of the fundamental frequency and overtones. By filtering the original fundamental frequency and overtones of the phoneme to be obfuscated within the time interval of the audio, and then adding the frequency and energy value of the target fundamental frequency and target overtones to the time interval, the frequency and energy value of the phoneme to be obfuscated in the audio are changed, thereby changing the voiceprint characteristics of the audio. While protecting the user's voiceprint, the ambient sound and the user's distinctive voice characteristics that may exist during the user's communication are preserved to the greatest extent.

[0098] To ensure that the energy value of the processed audio frequency does not change significantly, based on the above embodiments, in this embodiment, after determining the time interval where the phoneme to be confused is located in the audio, and before filtering the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval of the audio, the method further includes:

[0099] Determine the original energy sum of the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval; determine the target energy sum of the target fundamental frequency and target overtones; determine the gain attenuation coefficient based on the ratio of the original energy sum to the target energy sum;

[0100] Adding the frequency and energy values ​​of the target fundamental frequency and target overtones to the time interval of the audio includes: processing the energy values ​​of the target fundamental frequency and target overtones using the gain attenuation coefficient, and adding the frequency and processed energy values ​​of the target fundamental frequency and target overtones to the time interval of the audio.

[0101] If the frequency and energy values ​​of the target fundamental frequency and target overtones are directly added to the time interval of the audio to be confused, the energy values ​​of the frequencies in the processed time interval may be too large, making the audio unnatural and not conducive to the user's voice interaction.

[0102] To ensure that the energy values ​​of frequencies within the processed time interval do not change significantly, the original energy sum of the fundamental frequency and overtones of the phonemes to be confused within the time interval to be processed is first determined, and the target energy sum of the target fundamental frequency and target overtones of the phonemes to be confused is determined; the gain attenuation coefficient is determined based on the ratio of the original energy sum to the target energy sum.

[0103] The energy values ​​of the target fundamental frequency and target overtones are then processed using the gain attenuation coefficient, which means attenuating the energy values ​​of the target fundamental frequency and target overtones by the gain attenuation coefficient multiple, and adding the frequency of the target fundamental frequency and target overtones and the processed energy values ​​to the time interval of the audio that needs to be processed.

[0104] Specifically, as mentioned above Figure 3 Taking the original fundamental frequency and the frequencies of the first to sixth overtones corresponding to the phonemes in the text as an example, such as Figure 3 As shown, the energy values ​​corresponding to the original fundamental frequency and the first to sixth overtones are 27dB, 34dB, 30dB, 31dB, 26dB, 33dB, and 29dB, respectively. Summing these energy values ​​yields the original energy sum of the original fundamental frequency and overtones, which is 210dB.

[0105] Figure 6 This application provides a frequency and energy spectrum diagram of the target fundamental frequency and target overtone corresponding to a phoneme to be confused, as shown in the embodiments of this application. Figure 6 As shown, the energy values ​​corresponding to the fundamental frequency and the first to sixth overtones of the target fundamental frequency and target overtones are 57.4dB, 67.1dB, 65.9dB, 61.4dB, 59.1dB, 61.9dB, and 51.2dB, respectively. Summing these energy values ​​yields a target energy sum of 424dB for the target fundamental frequency and target overtones. The gain attenuation coefficient is the ratio of the original energy sum to the target energy sum: γ = 210 / 424 = 0.495. After applying this gain attenuation coefficient to the energy values ​​of the target fundamental frequency and target overtones by a gain attenuation of 0.495, the frequencies of the target fundamental frequency and target overtones, along with the attenuated energy values, are then synthesized with the filtered audio within the same time interval. Figure 7a This application provides a spectrum diagram of synthesized phonemes corresponding to a time interval.

[0106] Assuming the original fundamental frequency and overtones of the phoneme to be confused are represented by f(t), w is the frequency of the target fundamental frequency and the target overtones, and P0, P1, P2, P3, P4, P5, and P6 represent the energy values ​​corresponding to the target fundamental frequency and the target overtones, respectively, the audio synthesized from the frequencies of the phoneme to be confused and the target fundamental frequency and the target overtones can be represented by the following function:

[0107] f(t)+γ*P0*sin(2*pi*w*t)+γ*P1*sin(2*pi*w*t*2)+γ*P2*sin(2*pi*w*t*3)+γ*P3*sin (2*pi*w*t*4)+γ*P4*sin(2*pi*w*t*5)+γ*P5*sin(2*pi*w*t*6)+γ*P6*sin(2*pi*w*t*7)

[0108] Where f(t) is the original fundamental frequency and the audio after overtone filtering of the phoneme to be confused, γ is the gain attenuation coefficient, P0, P1, P2, P3, P4, P5, and P6 represent the energy values ​​corresponding to the target fundamental frequency and the target overtone, respectively, w is the frequency of the target fundamental frequency and the target overtone, t represents the length of the time interval of the target fundamental frequency and the target overtone, and pi is 3.14159.

[0109] In this embodiment of the application, a gain attenuation coefficient is determined based on the sum of the original energy of the original fundamental frequency and overtones of the phonemes to be confused within the time interval, and the sum of the target energy of the target fundamental frequency and the target overtones. Then, the gain attenuation coefficient is used to process the energy values ​​of the target fundamental frequency and the target overtones, so as to ensure that the energy values ​​of the processed audio frequencies do not change significantly, the processed audio is similar to the original audio, and does not sound too harsh.

[0110] In order to make the processed audio sound similar to the original audio, based on the above embodiments, in this embodiment of the application, the step of using a pre-trained recognition model to determine the target audio type of the audio includes:

[0111] Using a pre-trained first model, the age group and gender of the target user to which the audio belongs are determined; using a pre-trained second model, the similarity between the audio and audios of various preset voice styles is determined; based on the similarity with audios of various preset voice styles, the target voice style of the audio is determined; and the age group, gender, and target voice style are determined as the target audio category of the audio.

[0112] To accurately identify the audio category of a target user, this embodiment identifies the user's audio category across multiple dimensions, including age group, gender, and voice style. To accurately determine the target audio category of a user's audio, it is necessary to identify age group, gender, and voice style separately.

[0113] In this embodiment, a pre-trained first model is used to determine the age group and gender of the target user to whom the audio belongs. Specifically, different labels can be assigned based on the differences in voice change between men and women at different ages to train the first model, and then the age group and gender of the target user to whom the audio belongs can be determined through the trained first model.

[0114] The following is a specific example to illustrate this:

[0115] Age group and gender Label 0-6 years old Infants and young children 7-12 years old Children Ages 13-17 (for boys) juvenile Ages 13-22 (for females) girl 18-35 years old (applicable to men) young men Ages 23-35 (for women) young woman 36-69 middle aged >69 elderly

[0116] Table 1

[0117] As shown in Table 1 above, the labels corresponding to age groups and genders include infants, teenagers, young men, young women, middle-aged, and elderly. The labels are divided according to the differences in voice changes between men and women in different age groups. For example, the difference between male and female voices is not significant for infants aged 0-6 years. Therefore, the age group and gender label of the target user to which the audio belongs in this age group can be directly determined as infants. Similarly, audio clips for ages 7-12 are tagged as children, audio clips for ages 13-17 and males are tagged as teenagers, audio clips for ages 13-22 and females are tagged as teenage girls, audio clips for ages 18-35 and males are tagged as young men, audio clips for ages 23-35 and females are tagged as young women, audio clips for ages 36-69 are tagged as middle-aged, and audio clips for ages over 69 are tagged as elderly.

[0118] The first model can employ existing classification models, such as those using a Platform of Artificial Intelligence (PAI). Alternatively, it can determine age group and gender by comparing the audio's voiceprint features with pre-saved audio voiceprint feature templates for different age groups and genders, calculating feature similarity.

[0119] Using a pre-trained second model, the similarity between the audio and audio in various preset voice styles can be determined. Based on the similarity with these preset voice styles, the target voice style of the audio can be determined. Specifically, the voice style corresponding to the highest similarity value can be used as the user's voice style, or the voice style corresponding to the lowest similarity value can be used as the user's voice style.

[0120] In this embodiment, the second model can be trained by classifying different sound styles according to the composition and intensity characteristics of each frequency component of the audio. Specifically, the classified sound styles include: booming, crisp, golden, silvery, lion-like, mixed, hoarse, plain, and throaty sounds.

[0121] The second model, after training, determines the similarity between the audio and various preset voice styles, such as {loud bell: 80%, crisp thunder: 50%, golden jade: 35%, silver bell: 42%, lion and tiger: 60%, mixed: 70%, hoarse voice: 20%, white voice: 33%, throat: 59%}. Based on the similarity with these preset voice styles, the target voice style is determined. For example, silver bell or lion and tiger can be used as the target style. The combination of age group, gender, and target voice style determines the target audio category. For example, if the determined age group, gender, and target style are 23-35 years old, female, and silver bell, then the target category is "young woman silver bell".

[0122] To determine a suitable target sound style, based on the above embodiments, in this embodiment, determining the target sound style of the audio based on the similarity includes:

[0123] If there is a similarity between the audio and the preset audio of each sound style that falls within the range of a first similarity threshold and a second similarity threshold, then the sound style of the audio with the highest similarity within the range is determined as the target sound style; wherein the first similarity threshold is greater than the second similarity threshold;

[0124] If there is no similarity within the range, and all similarities are greater than the first similarity threshold, then the audio style corresponding to the smallest similarity is determined as the target audio style.

[0125] If there is no similarity within the specified range, and all similarities are less than the second similarity threshold, then the audio style corresponding to the highest similarity is determined as the target audio style.

[0126] Because the target audio style can be determined based on the target sound style, and the corresponding obfuscated data containing that style can be used to process the audio, if the similarity between the target sound style and the audio of the preset sound styles is too high, the voiceprint change may be minimal. Conversely, if the similarity between the target sound style and the audio of the preset sound styles is too low, the processed audio may sound harsh. To determine a suitable target sound style, a first similarity threshold and a second similarity threshold are preset, where the first similarity threshold is greater than the second similarity threshold, and the difference between the first and second similarity thresholds is not significant. Specifically, the first similarity threshold can be set to 60%, and the second similarity threshold to 50%, or the first similarity threshold to 65%, and the second similarity threshold to 45%, thus ensuring that the selected target sound style is appropriate.

[0127] If the similarity with audio samples of various preset sound styles falls within the range of 50%-60%, the sound style with the highest similarity within this range is determined as the target sound style. If no similarity falls within this range, and all similarities are greater than 60%, the sound style with the lowest similarity is determined as the target sound style. If no similarity falls within this range, and all similarities are less than 50%, the sound style with the highest similarity is determined as the target sound style. If no similarity falls within this range, and both similarities less than 50% and greater than 60% exist, the sound style of the audio sample with the closest similarity of 50% or 60% is determined as the target sound style.

[0128] For example, if the similarity calculation results with the preset voice styles are: {Bell: 80%, Crisp Thunder: 50%, Golden Jade: 35%, Silver Bell: 60%, Lion and Tiger: 46%, Combined: 70%, Humorous: 20%, Plain Voice: 33%, Throat: 59%}, and the similarity in the range of 50%-60% are {Crisp Thunder: 50%, Silver Bell: 60%, Throat: 59%}, then the voice style corresponding to the highest similarity of 60% - Silver Bell - is determined as the target voice style.

[0129] The method described in the above embodiments for determining the target voice style can avoid causing too much change in the voiceprint and avoid making the processed audio sound harsh, thus achieving a natural adjustment of the audio.

[0130] In order to select a suitable phoneme to be confused, based on the above embodiments, in this embodiment of the application, the selection of the phoneme to be confused includes:

[0131] The number of phonemes to be confused is determined based on the preset confusion ratio and the total number of phonemes appearing in the audio.

[0132] Based on the stated quantity and the total quantity, determine the number of phonemes contained in each group;

[0133] Based on the order in which the phonemes appear in the audio, the phonemes are grouped according to their number, and one phoneme from each group is randomly selected as the phoneme to be confused.

[0134] Since the audio needs to be processed based on the selected phonemes to be confused, if too many phonemes are selected and their time intervals are too concentrated, the processed audio will sound unlike the original audio, resulting in a poor user voice interaction experience. If too few phonemes are selected and their time intervals are too scattered, the processed audio voiceprint may not change much, thus failing to protect the user's voiceprint.

[0135] Therefore, in this embodiment, a confusion ratio is preset. Based on the preset confusion ratio and the total number of phonemes appearing in the audio, the number of phonemes to be confused is determined. The total number is calculated based on the starting position of each marked phoneme. Repeated phonemes are counted as different phonemes because of their different positions.

[0136] Based on this quantity and the total quantity, determine the number of phonemes in each group; group the phonemes according to their order of appearance in the audio, and arbitrarily select one phoneme from each group as the phoneme to be confused. This ensures that the number of selected phonemes to be confused and their time intervals are appropriate. For example, the phoneme at a fixed position in each group can also be used as the phoneme to be confused, such as the first phoneme in each group or the last phoneme in each group.

[0137] Based on the above Figure 2 For example, Figure 2 As shown, the total number of phonemes appearing in the audio is 20. Assuming the preset confusion ratio is 20%, the number of phonemes to be confused is 4. Each group contains 5 phonemes. The phonemes are grouped according to their order of appearance in the audio. One phoneme is randomly selected from each group as the phoneme to be confused. That is, based on the order of appearance of the phonemes in the audio, one phoneme is randomly selected from every 5 phonemes as the phoneme to be confused. For example, the selected phonemes to be confused are as follows:

[0138] Select phoneme number Phoneme time interval phoneme 3 247-360ms / i / 6 488-614ms / a: / 12 1364-1464ms / I / 18 1804-1912ms / e /

[0139] Table 2

[0140] The column for selected phonemes in the table represents the order in which the selected phonemes to be confused appear in the audio. The column for phoneme time intervals represents the time intervals in which the selected phonemes to be confused appear in the audio. Specifically, the second row indicates that the selected phoneme to be confused is the 3rd phoneme in the audio, with a time interval of 247-360ms, and the phoneme is / i / ; the third row indicates that the selected phoneme to be confused is the 6th phoneme in the audio, with a time interval of 488-614ms, and the phoneme is / a: / ; the fourth row indicates that the selected phoneme to be confused is the 12th phoneme in the audio, with a time interval of 1364-1464ms, and the phoneme is / I / ; the fifth row indicates that the selected phoneme to be confused is the 18th phoneme in the audio, with a time interval of 1804-1912ms, and the phoneme is / e / .

[0141] By using the scheme described in the above embodiments to determine the phonemes to be confused, it can be ensured that the number of phonemes to be confused and the time interval in which they appear in the audio are appropriate, thereby ensuring that the processed audio is more natural.

[0142] To make the processed audio sound similar to the original audio, but with altered voiceprints, based on the above embodiments, in this application embodiment, the obfuscated data is obfuscated audio. After determining the time interval containing the phoneme to be obfuscated in the audio, the processing of the audio within the time interval according to the target obfuscated data includes:

[0143] The target obfuscated audio is sampled and synthesized with the audio within the time interval.

[0144] The obfuscated data can also be obfuscated audio. In the pre-saved obfuscated audio corresponding to each phoneme of each audio type, the target obfuscated audio corresponding to the phoneme to be obfuscated of the target audio type is determined.

[0145] After determining the time interval of the phonemes to be confused in the audio, the target confused audio can be sampled and synthesized with audio within that time interval. Since the target confused audio corresponds to the phonemes to be confused in the target audio category of the original audio, the processed audio sounds similar to the original audio, but the voiceprint is altered.

[0146] Figure 7b This is a second schematic diagram illustrating an audio processing procedure provided in an embodiment of this application. For example... Figure 7b As shown, the process includes the following steps:

[0147] S701: For the audio to be processed, a pre-trained recognition model is used to determine the target audio type of the audio.

[0148] S702: Identify the starting position of each phoneme in the audio and select the phoneme to be confused.

[0149] S703: Among the pre-saved confused audios corresponding to each phoneme of each audio type, determine the target confused audio corresponding to the phoneme to be confused of the target audio type.

[0150] S704: Sample and synthesize the target obfuscated audio with the audio within the time interval.

[0151] In this embodiment of the application, the obfuscated data is obfuscated audio. After determining the time interval where the phonemes to be obfuscated are located in the audio, the target obfuscated audio is sampled and synthesized with the audio in the time interval, so that the processed audio sounds similar to the original audio, but the voiceprint is changed to protect the user's voiceprint.

[0152] Based on the above embodiments, this application also provides an audio processing apparatus. Figure 8 This is a schematic diagram of the structure of an audio processing device provided in an embodiment of this application, as shown below. Figure 8 As shown, the device includes:

[0153] The determination module 801 is used to determine the target audio category of the audio to be processed using a pre-trained recognition model; identify the starting position of each phoneme in the audio and select the phoneme to be confused; and determine the target confusion data corresponding to the phoneme to be confused in the target audio category from the pre-saved confusion data corresponding to each phoneme of each audio category.

[0154] The processing module 802 is used to determine the time interval in the audio where the phoneme to be confused is located, and to process the audio in the time interval according to the target confusion data.

[0155] In one possible implementation, the processing module 802 is specifically used to filter the original fundamental frequency and overtones corresponding to the phonemes to be confused within the time interval of the audio; and to add the frequency and energy values ​​of the target fundamental frequency and target overtones to the time interval of the audio.

[0156] In one possible implementation, the determining module 801 is specifically used to determine the original energy sum of the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval; determine the target energy sum of the target fundamental frequency and target overtones; determine a gain attenuation coefficient based on the ratio of the original energy sum to the target energy sum; and add the frequency and energy value of the target fundamental frequency and target overtones to the time interval of the audio, which includes: processing the energy value of the target fundamental frequency and target overtones using the gain attenuation coefficient, and adding the frequency and processed energy value of the target fundamental frequency and target overtones to the time interval of the audio.

[0157] In one possible implementation, the processing module 802 is specifically used to sample and synthesize the target obfuscated audio with the audio within the time interval.

[0158] In one possible implementation, the determining module 801 is specifically used to: use a pre-trained first model to determine the age group and gender of the target user to which the audio belongs; use a pre-trained second model to determine the similarity between the audio and audios of various preset voice styles; determine the target voice style of the audio based on the similarity with audios of various preset voice styles; and determine the age group, gender, and target voice style as the target audio category of the audio.

[0159] In one possible implementation, the determining module 801 is specifically configured to: if there is a similarity between the audio samples with preset similarity to each audio sample with a first similarity threshold and a second similarity threshold, then determine the audio sample with the largest similarity within the range as the target audio sample; wherein the first similarity threshold is greater than the second similarity threshold; if there is no similarity within the range and all similarities are greater than the first similarity threshold, then determine the audio sample with the smallest similarity as the target audio sample; if there is no similarity within the range and all similarities are less than the second similarity threshold, then determine the audio sample with the largest similarity as the target audio sample.

[0160] In one possible implementation, the determining module 801 is specifically used to determine the number of selected phonemes to be confused based on a preset confusion ratio and the total number of phonemes appearing in the audio; determine the number of phonemes included in each group based on the number and the total number; group the phonemes according to the number of phonemes based on the order of appearance of the phonemes in the audio, and randomly select one phoneme from each group as the phoneme to be confused.

[0161] The device can be specifically deployed in a terminal, and other functions of the terminal are described in the other embodiments above.

[0162] Based on the above embodiments, this application also provides an electronic device. Figure 9 This is a schematic diagram of an electronic device structure provided in this application. Figure 9 As shown, it includes: processor 901, communication interface 902, memory 903 and communication bus 904, wherein processor 901, communication interface 902 and memory 903 communicate with each other through communication bus 904.

[0163] The memory 903 stores a computer program, which, when executed by the processor 901, causes the processor 901 to complete the steps of any of the above-described audio processing methods.

[0164] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0165] The communication interface 902 is used for communication between the above-mentioned electronic device and other devices.

[0166] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0167] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0168] Based on the above embodiments, the present invention provides a computer-readable storage medium storing a computer program executable by an electronic device, wherein computer-executable instructions are used to cause a computer to execute the process performed by any of the aforementioned audio processing methods.

[0169] The aforementioned computer-readable storage medium can be any available medium or data storage device that can be accessed by the processor in an electronic device, including but not limited to magnetic storage such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), optical storage such as CDs, DVDs, BDs, HVDs, etc., and semiconductor storage such as ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs), etc.

[0170] In this application, after determining the target audio type for the audio to be processed, the phonemes to be confused are selected in the audio. The target confusion data corresponding to the phonemes to be confused is used to process the audio within the time interval corresponding to the phonemes to be confused. Since only some phonemes are processed, it can be ensured that the audio sounds similar to the original audio. However, since the voiceprint features contained in the audio have changed, the confusion of the user's voiceprint features is achieved, thus protecting the user's voiceprint.

[0171] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0173] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0175] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An audio processing method, characterized in that, The method includes: For the audio to be processed, a pre-trained recognition model is used to determine the target audio category of the audio. Identify the starting position of each phoneme in the audio and select the phonemes to be confused; In the pre-saved obfuscation data corresponding to each phoneme of each audio type, determine the target obfuscation data corresponding to the phoneme to be obfuscated of the target audio type; The time interval of the phoneme to be confused is determined in the audio, and the audio in the time interval is processed according to the target confusion data; The obfuscated data consists of the fundamental frequency and overtone frequencies and energy values. The processing of the audio within the time interval based on the target obfuscated data includes: Filter the original fundamental frequency and overtones of the phonemes to be confused within the time interval of the audio; The frequency and energy values ​​of the target fundamental frequency and target overtone are added to the time interval of the audio.

2. The method according to claim 1, characterized in that, After determining the time interval where the phoneme to be confused is located in the audio, and before filtering the original fundamental frequency and overtones corresponding to the phoneme to be confused within the time interval of the audio, the method further includes: Determine the original energy of the fundamental frequency and overtones of the phoneme to be confused within the time interval; Determine the target energy of the target fundamental frequency and the target overtone; The gain attenuation coefficient is determined based on the ratio of the original energy to the target energy. Adding the frequency and energy values ​​of the target fundamental frequency and target overtone to the time interval of the audio includes: The energy values ​​of the target fundamental frequency and the target overtone are processed using the gain attenuation coefficient, and the frequencies of the target fundamental frequency and the target overtone, along with the processed energy values, are added to the time interval of the audio.

3. The method according to claim 1, characterized in that, The obfuscated data is obfuscated audio. After determining the time interval where the phoneme to be obfuscated is located in the audio, the processing of the audio within the time interval based on the target obfuscated data includes: The target obfuscated audio is sampled and synthesized with the audio within the time interval.

4. The method according to claim 1, characterized in that, The method of using a pre-trained recognition model to determine the target audio type of the audio includes: Using a pre-trained first model, the age group and gender of the target user to whom the audio belongs are determined; Using a pre-trained second model, the similarity between the audio and audios of various preset sound styles is determined; based on the similarity between the audio and audios of various preset sound styles, the target sound style of the audio is determined. The age group, gender, and target voice style are determined as the target audio category of the audio.

5. The method according to claim 1, characterized in that, Determining the target sound style of the audio based on the similarity includes: If there is a similarity between the audio and the preset audio of each sound style that falls within the range of a first similarity threshold and a second similarity threshold, then the sound style of the audio with the highest similarity within the range is determined as the target sound style; wherein the first similarity threshold is greater than the second similarity threshold; If there is no similarity within the range, and all similarities are greater than the first similarity threshold, then the audio style corresponding to the smallest similarity is determined as the target audio style. If there is no similarity within the specified range, and all similarities are less than the second similarity threshold, then the audio style corresponding to the highest similarity is determined as the target audio style.

6. The method according to claim 1, characterized in that, The selected phonemes to be confused include: The number of phonemes to be confused is determined based on the preset confusion ratio and the total number of phonemes appearing in the audio. Based on the stated quantity and the total quantity, determine the number of phonemes contained in each group; Based on the order in which the phonemes appear in the audio, the phonemes are grouped according to their number, and one phoneme from each group is randomly selected as the phoneme to be confused.

7. An audio processing device, characterized in that, The device includes: The determination module is used to determine the target audio category of the audio to be processed using a pre-trained recognition model; identify the starting position of each phoneme in the audio and select the phoneme to be confused; and determine the target confusion data corresponding to the phoneme to be confused in the target audio category from the pre-saved confusion data corresponding to each phoneme of each audio category. The processing module is used to determine the time interval in the audio where the phoneme to be confused is located, and to process the audio in the time interval according to the target confusion data; The obfuscated data consists of the fundamental frequency and overtone frequencies and energy values. The processing of the audio within the time interval based on the target obfuscated data includes: Filter the original fundamental frequency and overtones of the phonemes to be confused within the time interval of the audio; The frequency and energy values ​​of the target fundamental frequency and target overtone are added to the time interval of the audio.

8. An electronic device, characterized in that, The electronic device includes at least a processor and a memory, the processor being configured to implement the steps of the audio processing method as described in any one of claims 1-6 when executing a computer program stored in the memory.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the audio processing method as described in any one of claims 1-6.

Images