AI voice rate adjustment method and system for AI platform
By analyzing the broadband and narrowband spectrograms of voice data, core speech segments were identified, and the voice rate of the AI platform was adjusted to match user preferences. This solved the problem of inaccurate voice rate adjustment on the AI platform and improved the user experience.
Patent Information
- Application Number
- CN202411139857.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-20
AI Technical Summary
Existing automotive AI platforms struggle to automatically adjust to different users' voice rate preferences, leading to a decline in user experience, especially with severe interference in multi-person communication environments.
By analyzing broadband and narrowband spectrograms, the core speech segments of the target user are identified, and the speech rate of the AI platform is adjusted according to the user's speech rate to match the user's speech rate.
It enables automated adaptive adjustment of the AI platform to the user's voice rate, improving the user experience and avoiding inaccurate speech rate adjustment caused by interference from others.
Smart Images

Figure CN118658450B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of natural language processing, and in particular to an AI speech rate adjustment method and system for an AI platform. Background Art
[0002] With the development of science and technology, the popularization of intelligent voice in cars currently allows users to perform basic operations and basic communication through communication with artificial intelligence (AI). In the current automotive AI platform, the AI voice rate adjustment method and system are important components of speech synthesis technology. With the rapid development of artificial intelligence and natural language processing technology, the speech synthesis system can not only generate high-quality speech, but also adjust the speech rate according to user needs to provide a more personalized and optimized user experience.
[0003] Currently, cars are equipped with intelligent AI voice platforms so that users can better perform car-related operations and basic communication. Different users may have different preferences. For example, some people may prefer a slightly faster voice rate, while others may prefer a slower rate. As for the current AI voice rate adjustment problem, the AI platform generally fixes the default AI voice rate or manually adjusts the speech rate parameters to determine the AI voice rate. This makes it difficult for operators to determine the appropriate AI voice rate for a while, and it is impossible to achieve automatic speech rate matching for different users. Especially when communicating with AI, since there may be multiple people speaking at the same time in the communication environment, this will also interfere with the communication between the user and AI, and thus affect the determination of the AI voice rate that matches the user. Summary of the invention
[0004] In order to solve the technical problem that when a user in a car communicates with AI, the improper AI speech speed will affect the user experience, the purpose of the present invention is to provide an AI voice rate adjustment method and system for an AI platform, and the technical solution adopted is as follows:
[0005] In a first aspect, the present invention provides an AI voice rate adjustment method for an AI platform, comprising:
[0006] Obtain the voice data collected by the AI platform in the current AI communication environment;
[0007] Generating a broadband spectrogram and a narrowband spectrogram corresponding to the speech data;
[0008] Determining a core speech segment of a target user in the speech data by analyzing the broadband spectrogram and the narrowband spectrogram;
[0009] According to the voice rate corresponding to the core voice segment, the voice rate of the AI platform when communicating with the target user through AI is adjusted.
[0010] Optionally, determining a core speech segment of a target user in the speech data by analyzing the wideband spectrogram and the narrowband spectrogram includes:
[0011] Determine the main frequency performance of the target user at each speech moment covered by the speech data by analyzing the frequency changes in the narrowband spectrogram;
[0012] Dividing the speech data into a plurality of speech segments by analyzing frequency changes in the broadband spectrogram;
[0013] Calculating the confidence of each speech segment in the plurality of speech segments as a core speech segment according to the main frequency performance;
[0014] Based on the confidence level, core speech segments of the target user are screened from the multiple speech segments.
[0015] Optionally, determining the main frequency performance of the target user at each speech moment covered by the speech data by analyzing the frequency change in the narrowband spectrogram includes:
[0016] Determine an expandable time window corresponding to each voice moment covered by the voice data, and window information corresponding to the expandable time window, wherein the window information at least includes a window size of the expandable time window;
[0017] Based on the narrowband spectrogram, extracting first frequency information in each of the extended time windows, the first frequency information at least including a first frequency value in the narrowband spectrogram at each moment in the extended time window, and a first frequency quantity of all the first frequency values contained in the extended time window;
[0018] The main frequency performance at each speech moment is calculated according to the window information and the first frequency information corresponding to each speech moment.
[0019] Optionally, dividing the speech data into a plurality of speech segments by analyzing the frequency changes in the broadband spectrogram comprises:
[0020] Determine second frequency information at each of the speech moments based on the broadband spectrogram, the second frequency information at least including a second frequency value of each of the speech moments in the broadband spectrogram;
[0021] Calculating, according to the second frequency information, each of the voice moments as a judgment value of a speaking rhythm pause moment;
[0022] Based on the judgment value, selecting at least one speaking rhythm pause point moment from a plurality of speech moments;
[0023] The voice data is segmented at each pause point of the speaking rhythm to obtain a plurality of voice segments corresponding to the voice data.
[0024] Optionally, calculating the confidence of each speech segment in the multiple speech segments as a core speech segment according to the main frequency performance includes:
[0025] Marking the first frequency value corresponding to each of the main frequency expressions in the narrowband spectrogram as a target frequency value, and calculating the user main frequency basis degree corresponding to each of the target frequency values;
[0026] For any one of the multiple speech segments, the confidence of the current speech segment as a core speech segment is calculated based on the target frequency value and the user main frequency basis.
[0027] Optionally, the calculating the user main frequency basis corresponding to each target frequency value includes:
[0028] In the narrowband spectrogram, counting the overall frequency quantity of the plurality of target frequency values corresponding to all the speech moments, and the second frequency quantity corresponding to each target frequency value;
[0029] A user main frequency basis degree corresponding to each of the target frequency values is calculated according to the second frequency quantity and the overall frequency quantity.
[0030] Optionally, for any one of the multiple speech segments, respectively calculating the confidence of the current speech segment as a core speech segment based on the target frequency value and the user main frequency basis, includes:
[0031] Determine any one of the multiple speech segments as the current speech segment;
[0032] Determine the speech paragraph information of the current speech paragraph, the speech paragraph information at least including the number of moments existing in the current speech paragraph, the frequency mean value corresponding to the second frequency value at each speech moment in the current speech paragraph, and the target frequency value and the user main frequency basis degree at each speech moment in the current speech paragraph;
[0033] Substitute the speech segment information into a preset calculation formula to calculate the confidence that the current speech segment is a core speech segment.
[0034] Optionally, screening the core voice segments of the target user from the multiple voice segments based on the confidence level includes:
[0035] The speech segment corresponding to the maximum confidence among the multiple speech segments is determined as the core speech segment of the target user.
[0036] Optionally, adjusting the speech rate of the AI platform and the target user during AI communication according to the speech rate corresponding to the core speech segment includes:
[0037] Determining an initial speech rate when the AI platform communicates with the target user through AI;
[0038] The initial speech rate is adjusted to the speech rate corresponding to the core speech segment.
[0039] In a second aspect, an embodiment of the present invention further provides an AI speech rate adjustment system for an AI platform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above-mentioned methods when executing the computer program.
[0040] The present invention has the following beneficial effects: through the technical solution provided by the present invention, the core voice segment of the target user is first determined in the voice data by converting the voice data into the form of a spectrogram, and based on the broadband spectrogram and narrowband spectrogram obtained by the conversion; then, the voice rate of the AI platform and the target user for AI communication is adjusted according to the voice rate corresponding to the core voice segment, so that the user and AI maintain the same speaking speed. The present invention can realize automatic adaptive adjustment of the AI voice speed according to the actual speaking speed of the target user communicating with the AI, so that the final adjusted AI voice rate fully matches the speech rate of the target user, which can improve the experience of AI communication. And by screening the core voice segments and determining the AI voice speed to be adjusted based on the core voice segments, it is possible to avoid the interference of other people making phone calls when the target user communicates with the AI, which may cause the final AI speed adjustment to be inaccurate.
[0041] It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present invention. Other features and advantages of the present invention will be described in detail in the following specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are 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.
[0043] Figure 1 A flowchart of an AI speech rate adjustment method of an AI platform provided by an embodiment of the present invention;
[0044] Figure 2 A flowchart of an AI speech rate adjustment method for an AI platform provided in accordance with another embodiment of the present invention. DETAILED DESCRIPTION
[0045] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the AI speech rate adjustment method and system of an AI platform proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0046] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0047] The following is a detailed description of a method and system for adjusting the AI speech rate of an AI platform provided by the present invention in conjunction with the accompanying drawings.
[0048] See also Figure 1 , which shows a flowchart of a method for adjusting the AI speech rate of an AI platform provided by an embodiment of the present invention, the method comprising the following steps:
[0049] Step 110: Obtain voice data collected by the AI platform in the current AI communication environment.
[0050] In specific application scenarios, current cars are often equipped with intelligent AI voice platforms (i.e., AI platforms), and target users can use the AI platform to perform relevant operational controls and basic communication with the car. In the current AI communication environment, the AI platform can first communicate with the target user at the default AI voice communication rate in the platform to collect voice data of the target user communicating with the AI platform, and use this as the basis for the AI platform to automatically adjust the AI voice rate when the target user communicates with the AI platform in the future. Among them, the target user is the user who communicates with the AI platform, who can be any user riding in the car, or a designated user riding in the car (such as the main driver or the co-driver, etc.), which can be personalized according to the actual application scenario, and is not specifically limited here.
[0051] It should be noted that in order to save the AI platform's storage resources for voice data, when the target user is a designated user riding in a car, in the current AI communication environment, the AI platform can selectively collect voice data based on the target user's voice characteristics (such as at least one of timbre, sound quality, pitch and sound intensity) when collecting voice data, such as only collecting voice data that contains the voice characteristics.
[0052] Step 120: Generate a wideband spectrogram and a narrowband spectrogram corresponding to the speech data.
[0053] For the embodiments of the present disclosure, the currently acquired speech data can be converted into corresponding spectrograms using tools such as MATALB, so as to facilitate subsequent specific analysis. For example, the speech data can be simultaneously converted into a broadband spectrogram with a short window length of 20ms and a narrowband spectrogram with a short window length of 2ms.
[0054] Step 130: Determine the core speech segment of the target user in the speech data by analyzing the broadband spectrogram and the narrowband spectrogram.
[0055] Among them, the core voice segment is the voice segment that is least interfered with in the voice data and best reflects the information characteristics of the communication between the target user and AI.
[0056] In the current AI communication environment, when the target user communicates with the AI platform, the communication between the target user and the AI platform may be interfered by the communication between other people in the car or external noise, so the collected voice data has interference factors, which is not conducive to the relevant operations of AI voice processing. Given that the frequency of human speech is generally in the low-frequency range (generally between 100-300hz), the target user's speaking frequency in the collected voice data is in the low-frequency range, and since the communication between the target user and AI is generally relatively continuous, the frequency characteristics corresponding to the target user are low-frequency and relatively continuous in time. Even when the target user communicates with AI, there is interference from other people speaking, and the frequency of the target user and other people is different in the low-frequency range, and there is also discontinuity when other people speak. Therefore, for such a situation, the main frequency range of the target user can be analyzed and determined. At the same time, silence detection is performed on the audio data to segment the audio data, and each segment of data is analyzed and compared with the main frequency range of the target user to determine the voice segment with the least interference as the core voice segment of the target user.
[0057] Step 140: According to the voice rate corresponding to the core voice segment, adjust the voice rate of the AI platform when communicating with the target user.
[0058] For the embodiments of the present disclosure, the speech rate of the core voice segment can be obtained and used as the basis for adjusting the AI speech rate to match the speech rate of the target user, so that the target user and the AI maintain the same speech rate, thereby improving the user experience during communication.
[0059] In summary, according to the AI voice rate adjustment method of an AI platform provided by the present invention, the core voice segment of the target user is first determined in the voice data by converting the voice data into the form of a spectrogram, and based on the broadband spectrogram and narrowband spectrogram obtained by the conversion; then, the voice rate of the AI platform and the target user for AI communication is adjusted according to the voice rate corresponding to the core voice segment, so that the user and AI maintain the same speaking speed. The present invention can realize automatic adaptive adjustment of the AI voice speed according to the actual speaking speed of the target user communicating with the AI, so that the final adjusted AI voice rate fully matches the speech rate of the target user, which can improve the experience of AI communication. And by screening the core voice segments and determining the AI voice speed to be adjusted based on the core voice segments, it is possible to avoid the interference of other people making phone calls when the target user communicates with the AI, which leads to inaccurate final AI speed adjustment.
[0060] based on Figure 1 The embodiment shown is a refinement and extension of the above embodiment. In order to fully illustrate the specific implementation process of the method of this embodiment, this embodiment provides the following Figure 2 The specific method shown. Figure 2 based on Figure 1 As shown in 2, the method comprises the following steps:
[0061] Step 210: Obtain voice data collected by the AI platform in the current AI communication environment.
[0062] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 110 of the embodiment, which will not be repeated here.
[0063] Step 220: Generate a wideband spectrogram and a narrowband spectrogram corresponding to the speech data.
[0064] For the embodiment of the present disclosure, the specific implementation process can refer to the relevant description in step 120 of the embodiment, which will not be repeated here.
[0065] Step 230: Determine the main frequency performance of the target user at each speech moment covered by the speech data by analyzing the frequency changes in the narrowband spectrogram.
[0066] Among them, the horizontal and vertical coordinates of the narrowband spectrogram are time and the first frequency value, respectively. For the disclosed embodiment, the narrowband spectrogram converted from the voice data can be used to perform a specific analysis and comparison of the main frequency performance. This is because the narrowband spectrogram focuses more on frequency resolution than the broadband spectrogram, and can provide more detailed and accurate frequency information. Taking into account the relative continuity of the communication between the target user and the AI, as well as the low frequency of the speaking frequency, an expansive time window analysis can be performed based on each moment here. This means that based on each moment as the intermediate time point, a fixed time range is determined before and after it, so as to determine the expansive time window and analyze the user's main frequency range based on this. The fixed time range is preset, and the same fixed time range size U / 2 is selected before and after each moment, which directly determines the expansive time window size as U, and the size of U can be 20ms.
[0067] Here, after the size of the expandable time window is determined as U at each moment, the data performance of each expandable time window in the narrowband spectrogram can be analyzed, and the adjacent frequency differences of each first frequency value in the row corresponding to each expandable time window of the narrowband spectrogram can be traversed (that is, the frequency data changes based on the time series), and the user's main frequency performance at each speech moment can be determined based on the adjacent frequency differences.
[0068] Correspondingly, for the embodiments of the present disclosure, the steps of the embodiments may include: determining the extended time window corresponding to each speech moment covered by the speech data, and the window information corresponding to the extended time window, the window information at least including the window size of the extended time window; based on the narrowband spectrogram, extracting the first frequency information under each extended time window, the first frequency information at least including the first frequency value in the narrowband spectrogram at each moment in the extended time window, and the first frequency quantity of all first frequency values contained in the extended time window; according to the window information and the first frequency information corresponding to each speech moment, calculating the main frequency performance at each speech moment.
[0069] Among them, when calculating the main frequency performance at each voice moment according to the window information and the first frequency information corresponding to each voice moment, the window information and the first frequency information corresponding to each voice moment can be substituted into the main frequency performance calculation formula to calculate the main frequency performance at each voice moment.
[0070] The formula characteristic description corresponding to the main frequency performance calculation formula is: ;
[0071] In the formula, Indicates the main frequency performance at speech time t; represents the first frequency value at any time in the extendable time window corresponding to the t voice time, F represents the number of frequencies existing in the extendable time window corresponding to the t voice time, that is, the number of first frequencies of all first frequency values contained in the extendable time window; U represents the window size of the extendable time window corresponding to the t voice time; represents the frequency difference between the u-th moment and the adjacent moment about the first frequency value in the expandable time window; is the middle value of the low-frequency range of human speech, which is a known value, such as 200 Hz; Indicates that in this extended time window, the first frequency value at the uth moment is frequency difference.
[0072] The main frequency performance calculation formula uses max() to filter out the maximum user frequency performance at the current voice moment t, which is used as the main frequency performance of the target user.
[0073] Step 240: Divide the speech data into a plurality of speech segments by analyzing the frequency changes in the broadband spectrogram.
[0074] Here, we can consider dividing the voice data into paragraphs. This is because when the target user communicates with the AI, there may be interference such as other people in the car making phone calls. Therefore, we consider dividing the voice data into paragraphs based on the discontinuity of the speaking frequency, and compare and analyze the divided voice paragraphs with the main frequency of the user at the corresponding moment. When dividing the voice paragraphs, you can use the broadband spectrogram of the voice data. This is because the broadband spectrogram usually has a higher time resolution and can more accurately capture the rapid changes and instantaneous characteristics of the voice signal. Therefore, compared with the narrowband spectrogram, the broadband spectrogram performs better in analyzing rapid voice feature changes and instantaneous events in the voice. Among them, the horizontal and vertical coordinates of the broadband spectrogram are time and the second frequency value, respectively.
[0075] Because there are certain pauses in the rhythm of human speech, and generally the pauses correspond to different parts of the speech sentence, when there is a pause, the voice frequency will be interrupted (it will be temporarily sharply reduced or approach 0), so based on this, the broadband spectrogram of the speech data can be divided. Here, the frequency data below 300hz in the broadband spectrogram can be analyzed, which is determined based on the low-frequency range of human speech, and the frequencies above 300hz may be affected by the interference of the external environment (even if the car is relatively closed, it is impossible to completely block the external sound). Specifically, the second frequency value at each speech moment in the broadband spectrogram can be traversed, and it can be compared with the second frequency values at subsequent multiple speech moments. When the difference between the second frequency value at each speech moment and the second frequency values at subsequent multiple speech moments is large, it means that the probability of corresponding to the speech pause here is greater.
[0076] Correspondingly, for the embodiments of the present disclosure, the embodiment steps may include: determining the second frequency information at each speech moment based on the broadband spectrogram, the second frequency information at least including the second frequency value of each speech moment in the broadband spectrogram; calculating each speech moment as a judgment value of a speaking rhythm pause moment based on the second frequency information; based on the judgment value, screening at least one speaking rhythm pause point moment from multiple speech moments; performing paragraph segmentation on the speech data at each speaking rhythm pause point moment to obtain multiple speech paragraphs corresponding to the speech data.
[0077] Among them, when calculating each voice moment as the judgment value of the speaking rhythm pause moment based on the second frequency information, the second frequency information at each voice moment can be substituted into the judgment value calculation formula of the speaking rhythm pause point to calculate the judgment value of each voice moment as the speaking rhythm pause moment.
[0078] The formula characteristic description corresponding to the judgment value calculation formula is: ;
[0079] In the formula, Indicates the judgment value of t speech moment as a pause point in the speaking rhythm; Y indicates the number of moments selected for comparison based on t speech moment; The second frequency value representing the speech time t (there may be multiple frequency values at the speech time t, and the largest frequency value can be used as the second frequency value); It represents the second frequency value of the yth speech moment after the t speech moment (similarly, based on its maximum frequency value); t represents any speech moment covered in the speech data; y represents the speech moment after the t speech moment for comparison with the t speech moment for the second frequency value.
[0080] Through the above judgment value calculation formula, the judgment value of the speaking rhythm pause point of each voice moment can be calculated (this value includes positive and negative, and the larger the value, the more likely it is to correspond to the speaking rhythm pause point). Here, among multiple voice moments, voice moments with relatively large corresponding judgment values can be selected as speaking rhythm pause point moments, and the voice data can be segmented at each speaking rhythm pause point moment, so that multiple voice paragraphs can be segmented. For example, S-1 speaking rhythm pause point moments can be selected from multiple voice moments, and the voice data can be segmented at each speaking rhythm pause point moment, so that S voice paragraphs can be segmented.
[0081] Step 250: Calculate the confidence of each speech segment in the multiple speech segments as a core speech segment based on the main frequency performance.
[0082] Accordingly, for the embodiment of the present disclosure, the step 250 of calculating the confidence of each speech segment in the plurality of speech segments as a core speech segment according to the main frequency performance may include the following steps:
[0083] Step 250-1: Mark the first frequency value corresponding to each main frequency expression in the narrowband spectrogram as a target frequency value, and calculate the user main frequency basis corresponding to each target frequency value.
[0084] For the embodiments of the present disclosure, the first frequency value corresponding to each main frequency expression in the narrowband spectrogram can be marked as the target frequency value to facilitate subsequent operations. For the target user, there are usually not too many emotions and rhythm changes when communicating with AI, so the corresponding frequency range is relatively constant, and considering that the communication between the target user and AI is relatively continuous, even if there may be interference such as someone making a phone call during the period, the target user still occupies a dominant position in the overall voice data performance. Therefore, the target frequency value marked at each moment is counted here, and its frequency in the narrowband spectrogram is counted based on each target frequency value. The frequency corresponding to each target frequency value is compared with the frequency of the overall frequency value on the narrowband spectrogram, so as to determine the user main frequency basis corresponding to each target frequency value. , that is, the user main frequency basis degree corresponding to each target frequency value = the frequency corresponding to each target frequency value / the frequency of the overall frequency value.
[0085] Accordingly, for the embodiments of the present disclosure, when calculating the user main frequency basis degree corresponding to each target frequency value, the embodiment steps may include: in the narrow-band spectrogram, counting the overall frequency quantity of multiple target frequency values corresponding to all speech moments, and the second frequency quantity corresponding to each target frequency value; based on the second frequency quantity and the overall frequency quantity, calculating the user main frequency basis degree corresponding to each target frequency value.
[0086] Step 250 - 2 : For any one of the multiple speech segments, the confidence of the current speech segment as a core speech segment is calculated based on the target frequency value and the user main frequency basis.
[0087] The above process determines the division of speech segments, where the core speech segments of the target user can be determined by comparing each speech segment with the user's main frequency performance (a narrowband spectrogram is also used for comparative analysis). Accordingly, for the disclosed embodiment, the embodiment steps may include: determining any one of the multiple speech segments as the current speech segment; determining the speech segment information of the current speech segment, the speech segment information at least including the number of moments existing in the current speech segment, the frequency mean of the first frequency value corresponding to each speech moment in the current speech segment, and the target frequency value and the user's main frequency basis at each speech moment in the current speech segment; substituting the speech segment information into the preset confidence calculation formula to calculate the confidence of the current speech segment as the core speech segment.
[0088] The formula characteristic description of the preset reliability calculation formula is: ;
[0089] In the formula, Indicates the confidence that the sth speech segment is the core speech segment; K indicates the number of moments in the current speech segment; Indicates the user main frequency basis degree at the kth speech moment of the sth speech segment; Indicates the frequency mean of multiple first frequency values in the narrowband spectrogram corresponding to the kth speech moment of the sth speech paragraph (there may be multiple first frequency values at the current moment, and the mean value is used as the basis, but the overall range is limited to less than 300 Hz); It represents the target frequency value corresponding to the user's main frequency performance at the kth speech moment of the sth speech segment; s is any speech segment among multiple speech segments; k is any speech moment in the speech segment.
[0090] The above-mentioned preset confidence calculation formula as a whole compares the first frequency value of each speech moment in the current speech segment with the target frequency value of the corresponding speech moment, while considering the user's main frequency basis of the corresponding speech moment, so as to determine the confidence of each speech segment as the user's main speech segment (i.e., the core speech segment); thereby, the confidence of each speech segment as a core speech segment can be determined.
[0091] Step 260: Based on the confidence level, core speech segments of the target user are screened from multiple speech segments.
[0092] Among them, the core voice segment refers to the voice segment among the multiple voice segments after division, which can better express the information characteristics of the target user’s communication with AI, that is, the voice segment that is least affected by factors such as people around making phone calls or external interference.
[0093] For the embodiment of the present disclosure, the embodiment steps may include: determining a speech segment with the largest corresponding confidence among multiple speech segments as a core speech segment of the target user.
[0094] Step 270: According to the voice rate corresponding to the core voice segment, adjust the voice rate of the AI platform when communicating with the target user.
[0095] For the embodiments of the present disclosure, the embodiment steps may include: determining the initial voice rate when the AI platform communicates with the target user through AI; adjusting the initial voice rate to the voice rate corresponding to the core voice segment. The initial voice rate may be the default AI voice communication rate in the AI platform, or may be the AI voice speed adjusted by the target user and the AI platform in the previous round of AI communication, etc., which is not specifically limited here.
[0096] In summary, the technical solution in this application first converts the voice data into a spectrogram form, and based on the broadband spectrogram and narrowband spectrogram obtained by the conversion, determines the core voice segment of the target user in the voice data; then adjusts the voice rate of the AI platform and the target user for AI communication according to the voice rate corresponding to the core voice segment, so that the user and AI maintain the same speaking speed. The present invention can realize automatic adaptive adjustment of the AI voice speed according to the actual speaking speed of the target user communicating with the AI, so that the final adjusted AI voice rate fully matches the speech rate of the target user, which can improve the experience of AI communication. And by screening the core voice segments and determining the AI voice speed to be adjusted based on the core voice segments, it is possible to avoid the interference of other people making phone calls when the target user communicates with the AI, which may cause inaccurate adjustment of the final AI speed.
[0097] Based on the same inventive concept as the above method, an embodiment of the present invention also provides an AI voice rate adjustment system for an AI platform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned methods for adjusting the AI voice rate of an AI platform are implemented.
[0098] In summary, the embodiment of the present invention provides an AI voice rate adjustment method for an AI platform, which first converts voice data into a spectrogram form, and determines the core voice segment of the target user in the voice data based on the spectrogram information; then adjusts the voice rate of the AI platform and the target user for AI communication according to the voice rate corresponding to the core voice segment, so that the user and the AI maintain the same speech speed. The present invention can realize the automation of AI voice speed adjustment, and can avoid the situation where the final AI speed adjustment is inaccurate due to interference from factors such as other people making phone calls when the user is communicating with the AI, fully matches the user's voice rate, and can improve the experience of AI communication.
[0099] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0100] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0101] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for adjusting the AI speech rate of an AI platform, characterized in that: The method comprises: Obtain the voice data collected by the AI platform in the current AI communication environment; Generating a broadband spectrogram and a narrowband spectrogram corresponding to the speech data; Determining a core speech segment of a target user in the speech data by analyzing the broadband spectrogram and the narrowband spectrogram; According to the voice rate corresponding to the core voice segment, adjusting the voice rate when the AI platform communicates with the target user through AI; The step of determining a core speech segment of a target user in the speech data by analyzing the broadband spectrogram and the narrowband spectrogram comprises: Determine the main frequency performance of the target user at each speech moment covered by the speech data by analyzing the frequency changes in the narrowband spectrogram; Dividing the speech data into a plurality of speech segments by analyzing frequency changes in the broadband spectrogram; Calculating the confidence of each speech segment in the plurality of speech segments as a core speech segment according to the main frequency performance; Based on the confidence level, core speech segments of the target user are screened from the multiple speech segments.
2. The AI speech rate adjustment method of an AI platform according to claim 1, characterized in that: Determining the main frequency performance of the target user at each speech moment covered by the speech data by analyzing the frequency changes in the narrowband spectrogram includes: Determine an expandable time window corresponding to each voice moment covered by the voice data, and window information corresponding to the expandable time window, wherein the window information at least includes a window size of the expandable time window; Based on the narrowband spectrogram, extracting first frequency information in each of the extended time windows, the first frequency information at least including a first frequency value in the narrowband spectrogram at each moment in the extended time window, and a first frequency quantity of all the first frequency values contained in the extended time window; The main frequency performance at each speech moment is calculated according to the window information and the first frequency information corresponding to each speech moment.
3. The AI speech rate adjustment method of an AI platform according to claim 1, characterized in that: The method of dividing the speech data into a plurality of speech segments by analyzing the frequency changes in the broadband spectrogram comprises: Determine second frequency information at each of the speech moments based on the broadband spectrogram, the second frequency information at least including a second frequency value of each of the speech moments in the broadband spectrogram; Calculate each of the voice moments as a judgment value of a speaking rhythm pause moment according to the second frequency information; Based on the judgment value, selecting at least one speaking rhythm pause point moment from a plurality of speech moments; The voice data is segmented at each pause point of the speaking rhythm to obtain a plurality of voice segments corresponding to the voice data.
4. The AI speech rate adjustment method of an AI platform according to claim 2, characterized in that: The step of calculating the confidence of each speech segment in the plurality of speech segments as a core speech segment according to the main frequency performance includes: Marking the first frequency value corresponding to each of the main frequency expressions in the narrowband spectrogram as a target frequency value, and calculating the user main frequency basis degree corresponding to each of the target frequency values; For any one of the multiple speech segments, the confidence of the current speech segment as a core speech segment is calculated based on the target frequency value and the user main frequency basis.
5. The AI speech rate adjustment method of an AI platform according to claim 4, characterized in that: The calculating the user main frequency basis corresponding to each target frequency value includes: In the narrowband spectrogram, counting the overall frequency quantity of the plurality of target frequency values corresponding to all the speech moments, and the second frequency quantity corresponding to each target frequency value; A user main frequency basis degree corresponding to each of the target frequency values is calculated according to the second frequency quantity and the overall frequency quantity.
6. The AI speech rate adjustment method of an AI platform according to claim 4, characterized in that: The step of calculating the confidence of the current speech segment as a core speech segment for any one of the multiple speech segments based on the target frequency value and the user main frequency basis respectively includes: Determine any one of the multiple speech segments as the current speech segment; Determine the speech paragraph information of the current speech paragraph, the speech paragraph information at least including the number of moments existing in the current speech paragraph, the frequency mean value corresponding to the first frequency value at each speech moment in the current speech paragraph, and the target frequency value and the user main frequency basis degree at each speech moment in the current speech paragraph; Substitute the speech segment information into a preset confidence calculation formula to calculate the confidence of the current speech segment as a core speech segment.
7. The AI speech rate adjustment method of an AI platform according to claim 1, characterized in that: The step of screening the core speech segments of the target user from the multiple speech segments based on the confidence level includes: The speech segment corresponding to the maximum confidence among the multiple speech segments is determined as the core speech segment of the target user.
8. The AI speech rate adjustment method of an AI platform according to claim 1, characterized in that: According to the voice rate corresponding to the core voice segment, adjusting the voice rate when the AI platform communicates with the target user through AI, including: Determining an initial speech rate when the AI platform communicates with the target user through AI; The initial speech rate is adjusted to the speech rate corresponding to the core speech segment.
9. An AI speech rate adjustment system of an AI platform, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of the AI speech rate adjustment method of an AI platform as described in any one of claims 1 to 8 are implemented.
