A speaker cloud service platform system

Abstract

This invention relates to a speaker cloud service platform system, addressing the technical problem of limited application and ineffective regulation of relationships between speakers. By employing a positioning detection unit at the speaker location, the cloud server executes the following steps: Step 1: The cloud server receives positioning data from the working speakers; Step 2: The cloud server determines the status of the operating speakers based on the positioning data. If the speaker's positioning distance is less than a predefined threshold, the corresponding speaker is marked as "suspected same group," and a "suspected same group confirmation message" is sent to the corresponding speaker; Step 3: The system receives the feedback result of the "suspected same group confirmation message." If the result is "yes," the speakers in the same group transmit data uniformly to any speaker in the same group, and any speaker is controlled to transmit data within the same group. If the result is "NO," playback data is transmitted sequentially according to weight priority, and the volume is controlled. This technical solution effectively solves the problem and can be used in speaker cloud services.

Description

Technical Field

[0001] This invention relates to the field of speaker cloud services, and more specifically to a speaker cloud service platform system. Background Technology

[0002] A smart speaker is an upgraded version of a traditional speaker, serving as a tool for home consumers to access the internet via voice commands, such as playing songs, shopping online, or checking the weather forecast. It can also control smart home devices, such as opening curtains, setting refrigerator temperatures, and preheating water heaters. On June 11, 2018, Baidu launched its first self-branded smart speaker, the "Xiaodu Smart Speaker," in Beijing. On June 1, 2019, Baidu's AI assistant, the "Xiaodu Smart Speaker Da Jin Gang," was launched on the Xiaodu Mall. On November 25, the Huawei SoundX smart speaker, jointly developed by Huawei and Devialet, was officially released.

[0003] Existing cloud service platform systems suffer from limitations such as single-application functionality and inability to effectively manage the relationships between speakers. This invention addresses these issues by providing a speaker cloud service platform system. Summary of the Invention

[0004] The technical problem this invention aims to solve is the limitation of existing technologies in their reliance on single applications and their inability to effectively regulate the relationships between speakers. This invention provides a novel speaker cloud service platform system that features diverse applications and effectively handles conflicts and relationships between speakers.

[0005] To solve the above technical problems, the following technical solution is adopted:

[0006] A speaker cloud service platform system includes a speaker comprising a voice input module, a network connection unit, and a playback device. The system includes a cloud server and a network connection unit for connecting the cloud server and the speaker. A positioning detection unit is installed at the speaker, and the cloud server receives data from the positioning detection unit in real time. The cloud server performs a speaker conflict detection step.

[0007] Step 1: The cloud server receives the location data from the speaker that is in operation;

[0008] Step 2: The cloud server determines the status of the speaker based on the location data. If the speaker's location distance is less than a predefined threshold, the corresponding speaker is marked as "suspected to be in the same group" and "suspected to be in the same group confirmation information" is sent to the corresponding speaker.

[0009] Step 3: Receive feedback results of "suspected same group confirmation information". If the result is "yes", transmit unified data of the same group speakers to any speaker in the same group and control any speaker to transmit data within the same group. If the result is "NO", transmit playback data in order of weight priority and control the volume.

[0010] The working principle of this invention: This invention uses speaker location information as the basis for determining whether they are "suspected to be in the same group," and effectively adjusts the relationship between potentially related or conflicting speakers based on feedback results. Simultaneously, the cloud server can control various internet applications.

[0011] In the above scheme, for optimization, the weight priority is further determined by the cloud server in the following way:

[0012] Step 1.1: Determine the speaker's network startup time; the earlier the time, the higher the priority.

[0013] Step 1.2: Determine the speaker's self-test status. A speaker with a good self-test status has higher priority.

[0014] Furthermore, the cloud server can also invoke other applications on the Internet according to the speaker's instructions.

[0015] Furthermore, the cloud server also receives voice control signals from the speaker for voice recognition, including:

[0016] Step 1: Establish a historical speech feature map library. The historical speech feature map is created by extracting features from pre-input or historical speech sentences and drawing speech speech feature maps. The speech speech feature map includes character, word, and sentence feature maps.

[0017] Step two: Extract features from the real-time voice recordings captured by the speaker and draw a target voice feature map; arbitrarily select a voice feature map from the historical voice feature map library as the reference image, and use the target voice feature map as the target image.

[0018] Step 3, transfer the target image I C Binarization is performed, with a value of 1 indicating the presence of speech features and 0 indicating the absence of speech features. The binarized feature map is then divided into a grid using a unit grid. The origin is defined as the first point (x1, y1) of the grid map, and the retrieval matching step size is defined as L. Starting from the origin, the retrieval proceeds along the x-direction. If a value of 1 is found, the position and value of that point are recorded and sequentially labeled; otherwise, the retrieval matching continues.

[0019] Step 4: Update point (x1, y1+N*L) to the origin, return to step 3, and continue until the x and y directions have been searched and matched to complete the preliminary positioning and retrieval matching, where N is an integer and L is a constant;

[0020] Step 5: Take out the points with a value of 1 in sequence, update the origin with the point with the value of 1 taken out in the current time, update the search matching step size to L / 2, and perform search matching in sequence along the x direction. Points that have been searched and matched before will not be searched and matched again. If the search matching is out of range, the search matching step size will be halved automatically and the search matching will continue until the step size is reduced to the minimum. If a new point with a value of 1 appears during the search matching process, it is defined as a new point that needs to be searched and matched in the y direction. Execute Step 6; otherwise, execute Step 7.

[0021] Step 6: The search and matching step size remains unchanged at L / 2. Search and match sequentially along the y-direction. Points that have already been searched and matched will not be searched and matched again. If the search and matching exceeds the range, the search and matching step size will be automatically halved and the search and matching will continue until the step size is reduced to the minimum. If a new point with a value of 1 appears during the search and matching process, it is defined as a new point that needs to be searched and matched in the x-direction. Step 5 is executed. Otherwise, Step 7 is executed.

[0022] Step 7: Continue searching and matching until no new points need to be searched and matched. Then, end the search and match process and collect the regions of the searched and matched 1-value points as the valid target image.

[0023] Step 8: Search and match the valid target images in the historical speech feature library;

[0024] Step 9: Invoke the corresponding strategy based on the recognition results.

[0025] Furthermore, step eight also includes image correction processing, including:

[0026] Step a, define the effective target image as I is defined as any reference image selected from the historical speech feature library. C ;

[0027] Step b: Define the reference image IC and the target image after polar coordinate transformation. The relationships are as follows:

[0028] Where, α z This is the scale offset parameter. For rotation offset parameters;

[0029] Step c, calculate the reference image I C Radial projection in polar coordinate system target image Radial projection K C (i) and Taking the logarithm gives LK C (i) and LK C (i) and The translation difference is used as the scale offset parameter α. z ;

[0030]

[0031] i = 1, 2, ..., n r ,

[0032]

[0033] For K i =K max The number of samples in the angular direction, ce() represents the smallest integer greater than or equal to the value in parentheses, fl() represents the largest integer less than or greater than the value in parentheses; the size of the target image is 2K. max ×2K max n r =K max n is the number of samples in the radial direction. φ =8K i This represents the number of samples taken in the angular direction.

[0034] Step d: Calculate the reference image I based on the scale offset parameters from step c. C and target image Projection in radial and angular directions:

[0035]

[0036]

[0037] right and Perform normalization calculations to determine the translation of the highest point. according to Calculate the rotation offset parameters

[0038] Step e, adjust the rotation offset parameter φ z and scale offset parameter α z The target image is corrected by substituting it into step A, and at the same time, according to... Calculate ∈ z The location point corresponding to the minimum value Use the center point of the target image to complete the image correction process.

[0039] Furthermore, the search matching analysis in step eight also includes:

[0040] Step A, the target image Concentric circles are drawn with the center point as the center to segment the speech feature image into B annular regions. Finally, each annular region is divided into K sector regions, where K and B are predefined constants.

[0041] Step B, calculate S for each sector. sq Sector speech feature value V sqθ As Code1;

[0042]

[0043] Among them, F sqθ (x, y) represents the sector region S sq The grayscale values ​​of each pixel, P sqθ S represents the sector region sq The average value of the inner pixel grayscale, n sq For the annular region S sq The number within, 0 < sq ≤ B × K - 1, θ = {0°, (360° / K), 2*(360° / K), 3*(360° / K), ... ≤ 180°};

[0044] Step C: After rotating the speech feature image (180° / K), repeat step B to extract each sector S. sq Sector speech feature value V sqθ As Code2;

[0045] Step E: Rotate Code1 and Code2 by R×(360° / K) (R=0,1,2...K-1) to obtain Code1' and Code2' respectively;

[0046] Step F: Input Code1 and Code2, Code1' and Code2' from step E into the historical speech feature library for matching.

[0047] The beneficial effects of this invention are as follows: This invention uses speaker positioning information as the basis for determining whether speakers are "suspected to be in the same group," and effectively adjusts the relationship between potentially related or conflicting speakers based on feedback results. Simultaneously, the cloud server can control various internet applications. Converting speech feature recognition into overall feature map recognition results in higher recognition efficiency. Through pre-cross-checking and positioning processing of feature images, the accuracy and efficiency of control are improved. Attached Figure Description

[0048] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0049] Figure 1 A schematic diagram of the speaker conflict detection steps. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0051] Example 1

[0052] This embodiment provides a speaker cloud service platform system. The speaker includes a voice input module, a network connection unit, and a playback device. The speaker cloud service platform system includes a cloud server and a network connection unit for connecting the cloud server and the speaker. A positioning detection unit is installed at the speaker, and the cloud server receives data from the positioning detection unit in real time. The cloud server performs a speaker conflict detection step.

[0053] Step 1: The cloud server receives the location data from the speaker that is in operation;

[0054] Step 2: The cloud server determines the status of the speaker based on the location data. If the speaker's location distance is less than a predefined threshold, the corresponding speaker is marked as "suspected to be in the same group" and "suspected to be in the same group confirmation information" is sent to the corresponding speaker.

[0055] Step 3: Receive feedback results of "suspected same group confirmation information". If the result is "yes", transmit unified data of the same group speakers to any speaker in the same group and control any speaker to transmit data within the same group. If the result is "NO", transmit playback data in order of weight priority and control the volume.

[0056] This embodiment uses speaker location information as the basis for determining whether they are "suspected to be in the same group," and effectively adjusts the relationship between potentially related or conflicting speakers based on feedback results. Simultaneously, the cloud server can control various internet applications.

[0057] Specifically, the weight priority is determined by the cloud server in the following way:

[0058] Step 1.1: Determine the speaker's network startup time; the earlier the time, the higher the priority.

[0059] Step 1.2: Determine the speaker's self-test status. A speaker with a good self-test status has higher priority.

[0060] Specifically, the cloud server can also invoke other applications on the Internet according to the speaker's instructions.

[0061] Preferably, the cloud server also receives voice control signals from the speaker for voice recognition, including:

[0062] Step 1: Establish a historical speech feature map library. The historical speech feature map is created by extracting features from pre-input or historical speech sentences and drawing speech speech feature maps. The speech speech feature map includes character, word, and sentence feature maps.

[0063] Step two: Extract features from the real-time voice recordings captured by the speaker and draw a target voice feature map; arbitrarily select a voice feature map from the historical voice feature map library as the reference image, and use the target voice feature map as the target image.

[0064] Step 3, transfer the target image I C Binarization is performed, with a value of 1 indicating the presence of speech features and 0 indicating the absence of speech features. The binarized feature map is then divided into a grid using a unit grid. The origin is defined as the first point (x1, y1) of the grid map, and the retrieval matching step size is defined as L. Starting from the origin, the retrieval proceeds along the x-direction. If a value of 1 is found, the position and value of that point are recorded and sequentially labeled; otherwise, the retrieval matching continues.

[0065] Step 4: Update point (x1, y1+N*L) to the origin, return to step 3, and continue until the x and y directions have been searched and matched to complete the preliminary positioning and retrieval matching, where N is an integer and L is a constant;

[0066] Step 5: Take out the points with a value of 1 in sequence, update the origin with the point with the value of 1 taken out in the current time, update the search matching step size to L / 2, and perform search matching in sequence along the x direction. Points that have been searched and matched before will not be searched and matched again. If the search matching is out of range, the search matching step size will be halved automatically and the search matching will continue until the step size is reduced to the minimum. If a new point with a value of 1 appears during the search matching process, it is defined as a new point that needs to be searched and matched in the y direction. Execute Step 6; otherwise, execute Step 7.

[0067] Step 6: The search and matching step size remains unchanged at L / 2. Search and match sequentially along the y-direction. Points that have already been searched and matched will not be searched and matched again. If the search and matching exceeds the range, the search and matching step size will be automatically halved and the search and matching will continue until the step size is reduced to the minimum. If a new point with a value of 1 appears during the search and matching process, it is defined as a new point that needs to be searched and matched in the x-direction. Step 5 is executed. Otherwise, Step 7 is executed.

[0068] Step 7: Continue searching and matching until no new points need to be searched and matched. Then, end the search and match process and collect the regions of the searched and matched 1-value points as the valid target image.

[0069] Step 8: Search and match the valid target images in the historical speech feature library;

[0070] Step 9: Invoke the corresponding strategy based on the recognition results.

[0071] Preferably, step eight further includes image correction processing, including:

[0072] Step a, define the effective target image as I is defined as any reference image selected from the historical speech feature library. C ;

[0073] Step b, define reference image I C and the target image after polar coordinate transformation The relationships are as follows:

[0074] Where, α z This is the scale offset parameter. For rotation offset parameters;

[0075] Step c, calculate the reference image I C Radial projection in polar coordinate system target image Radial projection K C (i) and Taking the logarithm gives LK C (i) and LK C (i) and The translation difference is used as the scale offset parameter α. z ;

[0076]

[0077] i = 1, 2, ..., n r ,

[0078]

[0079] For K i =K max The number of samples in the angular direction, ce() represents the smallest integer greater than or equal to the value in parentheses, fl() represents the largest integer less than or greater than the value in parentheses; the size of the target image is 2K. max ×2K max n r =K max n is the number of samples in the radial direction. φ =8K i This represents the number of samples taken in the angular direction.

[0080] Step d: Calculate the reference image I based on the scale offset parameters from step c. C and target image Projection in radial and angular directions:

[0081]

[0082]

[0083] right and Perform normalization calculations to determine the translation of the highest point. according to Calculate the rotation offset parameters

[0084] Step e, adjust the rotation offset parameter φ z and scale offset parameter α z The target image is corrected by substituting it into step A, and at the same time, according to... Calculate ∈ z The location point corresponding to the minimum value Use the center point of the target image to complete the image correction process.

[0085] Preferably, the search matching analysis in step eight further includes:

[0086] Step A, the target image Concentric circles are drawn with the center point as the center to segment the speech feature image into B annular regions. Finally, each annular region is divided into K sector regions, where K and B are predefined constants.

[0087] Step B, calculate S for each sector. sq Sector speech image feature value V sqθ As Code1;

[0088]

[0089] Among them, F sqθ (x, y) represents the sector region S sq The grayscale values ​​of each pixel, P sqθ S represents the sector region sq The average value of the inner pixel grayscale, n sq For the annular region S sq The number within, 0 < sq ≤ B × K - 1, θ = {0°, (360° / K), 2*(360° / K), 3*(360° / K), ... ≤ 180°};

[0090] Step C: After rotating the speech feature image (180° / K), repeat step B to extract each sector S. sq Sector speech feature value V sqθ As Code2;

[0091] Step E: Rotate Code1 and Code2 by R×(360° / K) (R=0,1,2...K-1) to obtain Code1' and Code2' respectively;

[0092] Step F: Input Code1 and Code2, Code1' and Code2' from step E into the historical speech feature library for matching.

[0093] This embodiment uses speaker location information as the basis for determining whether speakers are "suspected to be in the same group," and effectively adjusts the relationship between potentially related or conflicting speakers based on feedback results. Simultaneously, the cloud server can control various internet applications. Converting speech feature recognition into overall feature map recognition achieves higher recognition efficiency. Pre-crossover verification and location processing of feature images improve control accuracy and efficiency.

[0094] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, the invention is not limited to the scope of the specific embodiments. For those skilled in the art, all inventions utilizing the concept of the present invention are protected as long as various variations are within the spirit and scope of the invention as defined and determined by the appended claims.

Claims

1. A speaker cloud service platform system, wherein the speaker includes a voice input module, a network connection unit, and a playback device, characterized in that: The speaker cloud service platform system includes a cloud server and a network connection unit for connecting the cloud server and the speaker; a positioning detection unit is installed at the speaker, and the cloud server receives data from the positioning detection unit in real time. The cloud server performs the following speaker conflict detection steps: Step 1: The cloud server receives the location data from the speaker that is in operation; Step 2: The cloud server determines the status of the speaker based on the location data. If the speaker's location distance is less than a predefined threshold, the corresponding speaker is marked as "suspected to be in the same group" and a "suspected to be in the same group confirmation message" is sent to the corresponding speaker. Step 3: Receive feedback results of "suspected same group confirmation information". If the result is "yes", the same data of the same group speakers will be transmitted to any speaker in the same group, and any speaker will be controlled to transmit data within the same group. If the result is "NO", the playback data will be transmitted in order of weight priority and the volume will be controlled.

2. The speaker cloud service platform system according to claim 1, characterized in that: The weight priority is determined by the cloud server in the following way: Step 1.1: Determine the speaker's network startup time; the earlier the time, the higher the priority. Step 1.2: Determine the speaker's self-test status. A speaker with a good self-test status has higher priority.

3. The speaker cloud service platform system according to claim 1, characterized in that: The cloud server can also invoke other applications on the Internet according to the speaker's instructions.

4. The speaker cloud service platform system according to any one of claims 1-3, characterized in that: The cloud server also receives voice control signals from the speaker for voice recognition, including: Step 1: Establish a historical speech feature map library. The historical speech feature map is created by extracting features from pre-input or historical speech sentences and drawing speech speech feature maps. The speech speech feature map includes character, word, and sentence feature maps. Step two: Extract features from the real-time voice recordings captured by the speaker and draw a target voice feature map; arbitrarily select a voice feature map from the historical voice feature map library as the reference image, and use the target voice feature map as the target image. Step 3: Binarize the target image. A value of 1 indicates the presence of speech features, while 0 indicates the absence of speech features. Divide the binarized feature map into a grid using a unit grid. Define the first point (x1, y1) of the grid as the origin and define the retrieval matching step size as L. Starting from the origin, search along the x-direction. If a point with a value of 1 is found, record its position and value and label it sequentially. Otherwise, continue the retrieval matching. Step 4: Update point (x1, y1+N*L) to the origin, return to step 3, and continue until the x and y directions have been searched and matched to complete the preliminary positioning and retrieval matching, where N is an integer and L is a constant; Step 5: Take out the points with a value of 1 in sequence, update the origin with the point with the value of 1 taken out in the current time, update the search matching step size to L / 2, and perform search matching in sequence along the x direction. Points that have been searched and matched before will not be searched and matched again. If the search matching is out of range, the search matching step size will be halved automatically and the search matching will continue until the step size is reduced to the minimum. If a new point with a value of 1 appears during the search matching process, it is defined as a new point that needs to be searched and matched in the y direction. Execute Step 6; otherwise, execute Step 7. Step 6: The search and matching step size remains unchanged at L / 2. Search and match sequentially along the y-direction. Points that have already been searched and matched will not be searched and matched again. If the search and matching exceeds the range, the search and matching step size will be automatically halved and the search and matching will continue until the step size is reduced to the minimum. If a new point with a value of 1 appears during the search and matching process, it is defined as a new point that needs to be searched and matched in the x-direction. Step 5 is executed. Otherwise, Step 7 is executed. Step 7: Continue searching and matching until no new points need to be searched and matched. Then, end the search and match process and collect the regions of the searched and matched 1-value points as the valid target image. Step 8: Search and match the valid target images in the historical speech feature library; Step 9: Invoke the corresponding strategy based on the recognition results.

5. The speaker cloud service platform system according to claim 4, characterized in that: Step eight also includes image correction processing, including: Step a, define the effective target image as Select any reference image from the historical speech feature library and define it as... ; Step b, define the reference image and the target image after polar coordinate transformation The relationships are as follows: ;in, For scale offset parameters, For rotation offset parameters; Step c, calculate the reference image Radial projection in polar coordinate system Target image Radial projection ,Will and Taking the logarithm yields and ,Will and The translation difference is used as the scale offset parameter. ; ; ， ； ; , ; for Number of samples at the angular direction, Represents the smallest integer greater than or equal to the value within the parentheses. This represents the largest integer less than or equal to the value within the parentheses; the size of the target image is... , This represents the number of samples in the radial direction. This represents the number of samples taken in the angular direction. Step d: Calculate the reference image based on the scale offset parameters from step c. and target image Projection in terms of angle: ; ; right and Perform normalization calculations to determine the translation of the highest point. ,according to Calculate the rotation offset parameters ; Step e, set the rotation offset parameter and scale offset parameters Substitute the values ​​from step a into the target image for correction, and simultaneously, based on... Calculate The location point corresponding to the minimum value The center point of the target image is used to complete the image correction process.

6. The speaker cloud service platform system according to claim 4, characterized in that: The eighth step, search and matching analysis, also includes: Step A, the target image Concentric circles are drawn with the center point as the center to segment the speech feature image into B annular regions. Finally, each annular region is divided into K sector regions, where K and B are predefined constants. Step B: Calculate each sector sector speech feature values As Code1; ； in, For a fan-shaped area The grayscale values ​​of each pixel, Represents a sector area The average grayscale value of the inner pixel. A ring-shaped region The number within, , ; Step C, rotate the speech feature image Then, repeat step B to extract each sector. sector speech feature values As Code2; Step E: Rotate Code1 and Code2 respectively. Get Code1' and Code2'; Step F: Input Code1 and Code2, Code1' and Code2' from step E into the historical speech feature library for matching.

Images