A hearing aid hearing status detection method and system

Abstract

The present application relates to the technical field of hearing state determination, in particular to a hearing aid hearing state detection method and system, comprising the following steps: introducing a residual network to participate in difference structure representation, and filling in the missing paragraphs on the time axis, which can extract structure information directly related to gain offset from the original sound pressure level relationship, and complete node direction comparison and cross-band path reconstruction by combining dynamic time warping, the innovation lies in introducing two types of participants, offset direction marker and cross-band continuous path, which makes the offset change form a coherent trajectory in the time and frequency dimensions, and through path node deviation sorting, section division and numbered connection to establish a frequency band deviation level sequence, and removing continuous repeated numbers to form a hearing state number chain, which realizes that the hearing state determination result is driven by a continuous trajectory, and takes into account the time continuity, frequency band resolution and state label stability in the same detection process.

Description

Technical Field

[0001] This invention relates to the field of hearing status assessment technology, and in particular to a method and system for detecting the hearing status of a hearing aid. Background Technology

[0002] The field of hearing status assessment technology includes hearing aid signal processing and control, auditory perception modeling, gain and compression parameter setting, speech intelligibility assessment, and usage effect monitoring. The research object of this field is to construct a hearing function status model based on the electroacoustic signals in the hearing aid and usage parameters. Specifically, it involves digitally processing the input sound signal within the hearing aid and combining it with the user's wearing behavior parameters to determine the user's current hearing status category. The hearing status categories include compensation matching state, undercompensation state, overcompensation state, and abnormal working state. Multiple status categories are mapped to a set of hearing aid control parameters to guide the hearing aid's gain adjustment, compression strategy adjustment, and frequency band equalization adjustment in different states.

[0003] A hearing aid hearing status detection method is proposed. By setting up a hearing status detection process inside the hearing aid, the method analyzes the data during use and outputs a hearing status judgment result. The judgment result includes the hearing status category and the corresponding quantitative parameters. The method aims to reduce the number of adjustments, shorten the fitting and adjustment cycle, reduce the probability of compensation offset during long-term wear, and improve speech intelligibility and wearing comfort. It forms a hearing status trajectory for each user within the hearing aid's usage cycle. Through the status classification results and quantitative parameter output, it provides a direct control basis for automatic and semi-automatic parameter adjustment, thereby achieving stable maintenance of hearing aid compensation performance and improved long-term use effect.

[0004] Existing technologies for hearing status assessment focus on constructing a hearing function status model within the hearing aid based on electroacoustic signals and usage parameters. The assessment results are provided through a mapping between preset control parameter sets and status categories. The signal processing flow leans towards overall statistics and parameter level adjustment, paying insufficient attention to the difference structure between input and output sound pressure levels across frequency bands. It struggles to characterize the fluctuations and direction of compensation offset from a continuous time series. While status categories include compensation-matched, undercompensated, overcompensated, and abnormal operation states, the quantitative parameters for each state are mostly global attributes. Monitoring of usage effectiveness relies heavily on periodic follow-ups or external hearing assessments. The hearing status trajectory formed during the hearing aid's usage period is recorded in a coarse manner, lacking a structured description based on continuous time markers and frequency band numbers. This easily leads to situations where the hearing status remains stable on the label while the internal compensation relationship slowly deviates, resulting in decreased speech intelligibility and altered wearing comfort over a long period. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for detecting hearing status in hearing aids.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for detecting hearing status in a hearing aid, comprising the following steps:

[0007] S1: The sensor acquires the input sound pressure level, amplified output sound pressure level, and frequency band division information of the hearing aid microphone. The residual network subtracts the two types of sound pressure levels point by point according to the frequency band, rearranges the difference sequence values, and fills in the missing segment markers to generate a structured offset group.

[0008] S2: Based on the structured offset group, the offset sequence is extracted in time order in each frequency band and the rising segment and falling segment are recorded. After comparing the dynamic time warping direction in the same frequency band, they are connected into a continuous node sequence. The cross-frequency band path is reconstructed in each frequency band to obtain the offset trajectory mapping set.

[0009] S3: Based on the offset trajectory mapping set, extract the path node deviation in each frequency band and sort it by value. Divide the sorted sequence into a fixed number and give it a segment number. Then connect the segment numbers in the number order to form a unified sequence and establish a frequency band deviation level sequence.

[0010] S4: Based on the offset trajectory mapping set and the frequency band deviation level sequence, match the path node deviation value with the level segment number item by item and filter out nodes that do not match the direction. After rearranging the matching results according to the number order, remove consecutive duplicate numbers, establish a unified number sequence, and form a hearing status number chain.

[0011] S5: Based on the hearing status number chain, count the number of occurrences in each number and determine the peak number of occurrences. Map the peak number to the position of the corresponding name in the status name table, organize the text of the mapped name and maintain a single label output, and output the hearing status judgment result.

[0012] As a further aspect of the present invention, the structured offset group includes a difference sequence arranged by frequency band, a padded segment marker, and a rearranged sequence order; the offset trajectory mapping set includes cross-frequency band path nodes, node direction identifiers, and path connection order; the frequency band deviation level sequence specifically includes sorted node deviation segment numbers, segment segmentation order, and a unified connection number sequence; the hearing status number chain includes filtered number entries, a rearranged number order, and a deduplicated continuous number chain; and the hearing status determination result specifically refers to the peak number, the corresponding status name text, and the single label output content.

[0013] As a further aspect of the present invention, the specific steps for generating the structured offset group are as follows:

[0014] The sensor acquires the input sound pressure level, amplified output sound pressure level, and frequency band division information of the hearing aid microphone. At the same time position, the input value and output value are matched point by point and the sound pressure difference is recorded. All difference entries are sorted according to the division order, and a residual network is introduced during the sorting process to extract the difference structure, thus obtaining the original set of frequency band difference.

[0015] Based on the original set of frequency band differences, adjacent entries are checked in chronological order and the breakpoints are marked. The missing positions are filled with the mean of the preceding and following entries, and the original sorting order is maintained to make the difference sequence continuous and complete, thus obtaining the frequency band sequence correction set.

[0016] Based on the frequency band sequence correction set, the correction entries are rearranged in the time dimension, and the sorted entries are merged into a single set according to the division order. The set content is then output, and the whole is kept in a continuous structure to generate a structured offset group.

[0017] As a further aspect of the present invention, the residual network first receives a difference sequence arranged in frequency band order when organizing sound pressure level difference entries. The sequence is then fed into the layer-by-layer calculation structure of the network according to the input order. In each layer, the original input difference is maintained and the difference offset calculated by the current layer is superimposed. The superimposed result is then passed to the next layer. At the end structure, difference structure information that corresponds to the original difference sequence is output. The difference structure information is then added to the difference entry organization process to form the original set of frequency band differences.

[0018] As a further aspect of the present invention, the specific steps for obtaining the offset trajectory mapping set are as follows:

[0019] Based on the structured offset group, adjacent points are read sequentially from the offset record in chronological order. Values ​​greater than zero are marked as ascending segments and values ​​less than zero are marked as descending segments. Then, the segments are arranged in the original order to obtain the segment annotation sequence group.

[0020] Based on the paragraph annotation sequence group, the amplitude of adjacent paragraphs is extracted sequentially within the same sequence, and paragraphs with similar amplitude differences are connected by their start and end positions. Paragraphs with the same direction are continuously merged into a node chain to obtain a continuous node chain group.

[0021] Based on the continuous node chain group, each chain is arranged in sequence according to its starting position and the connection point difference is compared one by one. The connection relationship between the chains is calibrated by dynamic time warping. The valley difference chain is used as the connection target to form a cross-sequence continuous path structure and generate an offset trajectory mapping set.

[0022] As a further aspect of the present invention, the dynamic time warping first calibrates the connection relationship of continuous node chains. It receives multiple chain sequences arranged in order of their starting positions, inputs the node difference sequences of each chain into the warping process in sequence according to their current positions, calculates the distance between nodes during the sequence comparison process, records the movement direction of each step during the distance accumulation process, uses the recorded path trajectory to locate the alignment position of the difference between chains, and re-aligns the chain node order according to the path trajectory so that the connection relationship between chains is calibrated under the same warping path.

[0023] As a further aspect of the present invention, the specific steps for establishing the frequency band deviation level sequence are as follows:

[0024] Based on the offset trajectory mapping set, the node deviations are extracted sequentially according to the path order and the nodes are arranged one by one according to their numerical values. The arranged sequence is divided into several segments by equal division according to a fixed number. Each segment is assigned a number according to its order of appearance to obtain a segment number set.

[0025] Based on the set of segment numbers, the numbers are connected sequentially according to their size. During the connection process, the continuity of adjacent segment numbers is checked and redundant numbers are deleted where they are repeated, so that a single number sequence is formed after sorting, and a frequency band deviation level sequence is established.

[0026] As a further aspect of the present invention, the specific steps for forming the hearing status number chain are as follows:

[0027] Based on the offset trajectory mapping set and the frequency band deviation level sequence, the node deviation values ​​are read sequentially in the path record and each node is compared with the corresponding segment number according to its appearance order. The deviation direction is consistent with the segment direction as the screening basis, and the filtered numbers are rearranged according to their original order to obtain the direction screening number group.

[0028] Based on the direction of the number group, the number entries are arranged in ascending order of number value. During the arrangement process, duplicate numbers in adjacent positions are removed one by one. The remaining numbers are combined into a single number chain in a continuous connection manner, and the overall order of the chain is kept consistent to generate a hearing status number chain.

[0029] As a further aspect of the present invention, the specific steps for outputting the hearing status determination result are as follows:

[0030] Based on the hearing status number chain, the numbers are read sequentially and the same numbers are counted into the counting unit during reading. After traversing each position, the number of each counting unit is compared one by one, and the peak number is selected from the comparison results to obtain the peak number information set.

[0031] Based on the peak number information set, the name is located in the status name table by number and the corresponding name text is extracted from the table. The text spaces and separators are deleted and the characters are arranged by label. After the arrangement is completed, the label is output to obtain the hearing status judgment result.

[0032] A hearing aid hearing status detection system, the hearing aid hearing status detection system being used to perform the above-described hearing aid hearing status detection method, the system comprising:

[0033] Offset construction module: It acquires sound pressure level and frequency band division through sensors, derives the difference between input and output sound pressure levels and organizes the frequency band order, introduces a residual network to extract the difference structure, fills in the broken segments and rearranges the sequence to generate a structured offset group.

[0034] Trajectory generation module: Based on the structured offset group, it extracts the offset segments in time sequence and marks the direction. After completing the direction comparison by combining dynamic time warping, it connects the node segments and performs path combination and order sorting on the cross-frequency band segment chain to obtain the offset trajectory mapping set.

[0035] The grade composition module: Based on the offset trajectory mapping set, capture the node deviation and arrange it according to the value, divide the arrangement sequence into segments in an equal manner, label the segment contents with numbers and arrange the number sequence in order to establish a frequency band deviation grade sequence;

[0036] Numbering module: Based on the offset trajectory mapping set and the frequency band deviation level sequence, read the node deviation and perform directional filtering, rearrange the filtered numbers in order and delete consecutive duplicate numbers, and connect the sorted numbers to form a hearing status numbering chain;

[0037] Status output module: Based on the hearing status number chain, it counts the number and records the number of occurrences, extracts the peak number of the statistical results and compares it with the position in the name table, performs formatting and character arrangement on the name text, outputs the arrangement result as a label, and obtains the hearing status judgment result.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] 1. In this invention, by introducing a residual network to participate in the differential structure representation and filling in the missing segments on the time axis, compared with establishing a hearing function state model based solely on electroacoustic signals and usage parameters, it is possible to extract structural information directly related to gain shift from the original sound pressure level relationship.

[0040] 2. In this invention, node direction comparison and cross-band path reconstruction are completed by combining dynamic time warping. The innovation lies in the introduction of two types of participants: offset direction marker and cross-band continuous path, so that the offset change forms a coherent trajectory in both time and frequency dimensions.

[0041] 3. In this invention, a frequency band deviation level sequence is established by sorting path node deviations, segmenting and connecting them with numbers, removing consecutive duplicate numbers to form a hearing status number chain, and using the peak occurrence frequency to correspond to the status name during the numbering statistics process. This achieves the hearing status determination result being driven by a continuous trajectory, taking into account time continuity, frequency band resolution and status label stability in the same detection process. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0043] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0045] Example 1

[0046] Please see Figure 1 This invention provides a technical solution: a method for detecting hearing status in a hearing aid, comprising the following steps:

[0047] S1: The sensor acquires the input sound pressure level, amplified output sound pressure level, and frequency band division information of the hearing aid microphone. The residual network subtracts the two types of sound pressure levels point by point according to the frequency band, rearranges the difference sequence values, and fills in the missing segment markers to generate a structured offset group.

[0048] S2: Based on the structured offset group, the offset sequence is extracted in time order in each frequency band and the rising and falling segments are recorded. After comparing the dynamic time warping direction in the same frequency band, they are connected into a continuous node sequence. The cross-frequency band path is reconstructed in each frequency band to obtain the offset trajectory mapping set.

[0049] S3: Based on the offset trajectory mapping set, the path node deviation is extracted in each frequency band and sorted by value. The sorted sequence is divided into segments with a fixed number and given segment numbers. The segment numbers are then connected in order to form a unified sequence, thus establishing a frequency band deviation level sequence.

[0050] S4: Based on the offset trajectory mapping set and frequency band deviation level sequence, match the path node deviation value with the level segment number item by item and filter out nodes that do not match the direction. After rearranging the matching results according to the number order, remove consecutive duplicate numbers, establish a unified numbering sequence, and form a hearing status numbering chain.

[0051] S5: Based on the hearing status number chain, count the number of occurrences of each number and determine the peak number of occurrences. Map the peak number to the position of the corresponding name in the status name table, organize the text of the mapped name and keep a single label output, and output the hearing status judgment result.

[0052] The structured offset group includes a difference sequence arranged by frequency band, a padded segment marker, and a rearranged sequence order. The offset trajectory mapping set includes cross-frequency band path nodes, node direction identifiers, and path connection order. The frequency band deviation level sequence is specifically the sorted node deviation segment number, segment segmentation order, and unified connection number sequence. The hearing status number chain includes the filtered numbered entries, the rearranged number order, and the deduplicated continuous number chain. The hearing status judgment result specifically refers to the peak number, the corresponding status name text, and the single label output content.

[0053] The specific steps for generating structured offset groups are as follows:

[0054] The sensor acquires the input sound pressure level, amplified output sound pressure level, and frequency band division information of the hearing aid microphone. At the same time position, the input value and output value are matched point by point and the sound pressure difference is recorded. All difference entries are sorted according to the division order, and a residual network is introduced during the sorting process to extract the difference structure, thus obtaining the original set of frequency band difference.

[0055] Based on the original set of frequency band differences, adjacent entries are checked in chronological order and the breakpoints are marked. The missing positions are filled with the mean of the preceding and following entries, and the original sorting order is maintained to make the difference sequence continuous and complete, thus obtaining the frequency band sequence correction set.

[0056] Based on the frequency band sequence correction set, the correction entries are rearranged in the time dimension, and the sorted entries are merged into a single set according to the division order. The set content is then output, and the whole structure is maintained to generate a structured offset group.

[0057] The microphone input sound pressure level, amplified output sound pressure level, and frequency band division information are obtained by sensors. A residual network is used to derive the difference between the input and output values ​​at the same time position. During the difference derivation, the sound pressure value at each time moment is set to a single point value range of 0 to 120 dB. The length of each frequency band is set to 64 time points and input into the network in sequence. In the residual network, the input difference is calculated in a three-layer structure in the form of a sequence. Each layer consists of two one-dimensional sequence superposition operations. Each superposition operation is performed with a fixed step size of 1 and a fixed window of 3 time points. At the end of each layer, the difference is summed with a fixed weight of 1 to form the residual superposition content. The superposition content is then recorded as difference entries according to the original frequency band order and arranged item by item to generate the original set of frequency band difference values.

[0058] Based on the original set of frequency band differences, each difference entry is scanned continuously in chronological order. The scanning time step is set to once every 5 milliseconds. During continuous scanning, the change in the difference between two adjacent difference entries is used as the basis for determining the breakpoint. Entries with a change exceeding 25 dB are marked as breakpoints, and the time index and frequency band index when the breakpoint occurs are recorded. Each breakpoint is stored in the breakpoint record sequence, and a missing completion operation is performed at the breakpoint position. The completion step involves adding the values ​​of the entry before and after the breakpoint and multiplying by 0.5 to obtain the single-point filling value, which is in the range of 30 to 70 dB. During the filling process, the filling entries are written back to the original sequence position in a fixed order within each frequency band. After the breakpoint filling is completed, the entire sequence is rearranged again in chronological order, and a unified numbering mark is established for the rearrangement result. The numbering range is from 1 to 64, corresponding to 64 records in each frequency band. After the numbering is aligned, the corrected data is combined item by item to form a continuous difference sequence, generating a frequency band sequence correction set.

[0059] Based on the frequency band sequence correction set, the frequency band records are rearranged from 0 milliseconds to 320 milliseconds according to the time index. During the rearrangement, each frequency band has 64 records as a fixed group and they are sorted by time point. The sorted entries are grouped by frequency band number from 1 to 8 and each group of entries is stored as an independent set. When combining these sets, the splicing operation is performed in the order of frequency band number. The 64 records of frequency band 1 are placed at the beginning of the set, and the records of frequency band 2 to frequency band 8 are arranged sequentially. The original difference value range is retained in each group of records and written into a unified array structure according to the entry number. A complete output set contains 512 difference records. By storing the 512 records in 8 groups and reconstructing all entries according to a single time axis in the final output stage, all records are kept in continuous order from the 1st to the 512th record. At the same time, the time interval is set to increase every 5 milliseconds to form a complete sequence coordinate. The sorted set content is written into a structured data container in a fixed output format and the continuous structure is maintained to generate a structured offset group.

[0060] The residual network first receives the difference sequence arranged in frequency band order when organizing sound pressure level difference entries. The sequence is then fed into the layer-by-layer calculation structure of the network according to the input order. In each layer, the original input difference is maintained and the difference offset calculated by the current layer is superimposed. The superimposed result is then passed to the next layer. At the end of the network, difference structure information that corresponds to the original difference sequence is output. The difference structure information is then added to the difference entry organization process to form the original set of frequency band differences.

[0061] Residual networks, according to the formula:

[0062]

[0063] in: For the frequency band index Time index is Improved residual output value at the location, For the frequency band index Time index is The input sound pressure difference at that location, For the frequency band index Time index is The input sound pressure difference at that location, For the frequency band index Time index is The input sound pressure difference at that location, For the frequency band index Time index is The input sound pressure difference at that location, for The weighting coefficients, for The weighting coefficients, for The weighting coefficients, The bias constant is This is the frequency band adjustment coefficient. Add a coefficient to the time and location. For time index normalized values, Frequency band index The sum of the input differences at all 64 time points. Frequency band index The average difference For time-normalized components and their fixed weights, For index number, For time indexing, For frequency band indexing;

[0064] Execution process: Read the frequency band index point by point. With time index Input sound pressure difference , , The data is written to the calculation unit in a fixed order. When processing boundary points, missing positions are directly replaced with the current sound pressure difference value, and then weighted by a coefficient. , , After being applied to the three sets of differences, and then combined with the bias constant Perform an accumulation once to obtain the basic residual term, and simultaneously read all time points of this frequency band. It then performs summation and division to generate the average difference, which is used for the average term in the formula, and then indexes it by time. Calculate the normalized value The time addition term and time return term are then written in, and the above basic residual term is multiplied by the frequency band adjustment coefficient. Then, the time-dependent part, the frequency band averaging part, and the normalized time part are all added together in the same addition unit to obtain the final result. Index by frequency band With time index Write the output sequence sequentially to complete the improved residual derivation of the residual network at the time point.

[0065] The specific steps to obtain the offset trajectory mapping set are as follows:

[0066] Based on the structured offset group, adjacent points are read sequentially from the offset record in chronological order. Values ​​greater than zero are marked as rising segments and values ​​less than zero are marked as falling segments. Then, the segments are arranged in the original order to obtain the segment annotation sequence group.

[0067] Based on the paragraph annotation sequence group, the amplitude of adjacent paragraphs is extracted sequentially within the same sequence, and paragraphs with similar amplitude differences are connected by their start and end positions. Paragraphs with the same direction are continuously merged into a node chain to obtain a continuous node chain group.

[0068] Based on continuous node chain groups, each chain is arranged in sequence according to its starting position and the difference between the connection points is compared one by one. The connection relationship between the chains is calibrated by dynamic time warping. The valley difference chain is used as the connection target to form a cross-sequence continuous path structure and generate an offset trajectory mapping set.

[0069] Based on the structured offset group, adjacent points are read sequentially from the offset record in chronological order. When judging the difference between adjacent points, each time point is recorded as a single-point array format. Each array contains three items: time index (in milliseconds, ranging from 0 to 320), difference value (ranging from -30 dB to +30 dB), and frequency band number (from 1 to 8). When the difference is greater than zero, it is marked as an ascending segment and marked with mark code 1. When the difference is less than zero, it is marked as a descending segment and marked with mark code 2. The records are reordered according to the time index. During the sorting process, a segment number is generated for each record at a fixed interval of 20 milliseconds and written into the segment field. Each segment mark is written into the segment set and arranged according to the frequency band number and time order to form a segment list, generating a segment label sequence group.

[0070] Based on the paragraph label sequence group, the amplitude of adjacent paragraphs is extracted sequentially within the same sequence. During amplitude measurement, the maximum value of the absolute value of all differences recorded in two adjacent paragraphs is calculated and recorded as the amplitude value, which is recorded in integer form, generally ranging from 5 dB to 35 dB. When judging the amplitude difference, if the difference between two amplitude values ​​is less than 3 dB, the two paragraphs are marked as a connectable pair, and the connection pointer is recorded. The connectable pair is written into the connection table in the order of start and end, and sorted sequentially according to the starting paragraph number in the connection table. When merging paragraphs with the same direction, the direction code 1 in the paragraph label is used as the upward direction, and the code 2 is used as the downward direction. If the direction codes of two adjacent paragraphs are the same, the two paragraphs are merged, and the new paragraph number is used to overwrite the original number. The merged paragraph is written into the node table, and a node sequence is generated in the node table according to the paragraph order. The node sequence is stored as a node set according to the frequency band number, generating a continuous node chain group.

[0071] Based on a continuous node chain group, the chains are arranged sequentially according to their starting positions. During chain arrangement, the starting time index of each chain is used as the sorting key, and the chains are sorted in ascending order. When comparing the connection point differences, the difference at the end of each chain is paired with the starting difference of the next chain, and the absolute difference is recorded as the connection difference. The range of the connection difference is generally between 1 dB and 20 dB. A dynamic time warping algorithm is used to calibrate the connection relationships between the chains. During dynamic time warping, the difference sequences of two chains are written into a two-dimensional integer table. The rows of the two-dimensional integer table represent the length of the first chain, and the column index represents the length of the second chain. The length of the first chain is 15 nodes, and the length of the second chain is 1 node. With 8 nodes, when constructing the cost table, each cell in the table is written with the absolute value of the difference between the corresponding nodes of the two chains. Three movement methods are limited during movement: moving to the right means incrementing the column index by 1, moving down means incrementing the row index by 1, and moving diagonally means incrementing both the row index and the column index by 1. In the cost accumulation record, a fixed accumulation of 5 is added for each movement. After all movement records are completed, the minimum cost movement path is backtracked from the end of the cost table. The node index corresponding to the path is written into the alignment list. The aligned node matching relationship is written into the connection structure, and the valley difference chain is selected as the final connection target. The processed chains are combined in order to form a cross-sequence continuous path structure, generating an offset trajectory mapping set.

[0072] Dynamic time warping first calibrates the connection relationship of continuous node chains. It receives multiple chain sequences arranged in the initial order, inputs the node difference sequence of each chain into the warping process in sequence according to the current position, calculates the distance between nodes during the sequence comparison process, and records the movement direction of each step during the distance accumulation process. The recorded path trajectory is used to locate the alignment position of the difference between chains, and the chain node order is re-corresponded according to the path trajectory so that the connection relationship between chains is calibrated under the same warping path.

[0073] Dynamic time warping, according to the formula:

[0074]

[0075] in: For the cumulative alignment cost matrix at the index The cumulative alignment value at the location, For the second chain and The corresponding node index, For the first chain in the index The difference between nodes at each point, in decibels. For the first chain in the index The node difference at that point, For the second chain in the index The node difference at that point, For the second chain in the index The node difference at that point, For frequency band numbering, For frequency band The average of the differences between all nodes in the first inner chain. For frequency band The average of the differences between all nodes in the second inner chain. index for the first chain The corresponding time point, Second chain index The corresponding time point, The weight of the principal difference term. The forward interpolation compensation weight for the first chain. For the forward interpolation compensation weight of the second chain, The weight of the average difference term in the frequency band. For the weight of the time difference term, This represents the cumulative cost value at the corresponding position in the previous row of the cumulative matrix. This represents the cumulative cost value at the corresponding position on the left side of the cumulative matrix. This represents the cumulative cost of the top-left position in the cumulative matrix;

[0076] Execution process: First, when aligning consecutive node chains, the difference sequence between the two chains is indexed. and Write the regular structure and read it sequentially. and Simultaneously, read the forward difference from the linked list. and Frequency band numbering Sum the differences between all nodes and divide by the number of nodes to get and Then obtain from the time series and Weight , , , , After calculating the node differences using an offline record set, the local cost term is determined by multiplying the five differences by their corresponding weights and summing them during the runtime phase. Then, the three neighborhood terms are read from the cumulative matrix. , , Select the valley value and add it to the local cost term, then write it. Fill all index positions in row and column order, and backtrack from the end of the matrix to form the node alignment path for the offset trajectory mapping set.

[0077] The specific steps for establishing the frequency band deviation level sequence are as follows:

[0078] Based on the offset trajectory mapping set, the node deviations are extracted sequentially according to the path order and the nodes are arranged one by one according to their numerical values. The arranged sequence is divided into several segments in a fixed number of equal parts. Each segment is assigned a number according to its order of appearance, thus obtaining a set of segment numbers.

[0079] Based on the set of segment numbers, the numbers are connected sequentially according to their size. During the connection process, the continuity of adjacent segment numbers is checked and redundant numbers are deleted where numbers are repeated. After sorting, a single number sequence is formed, and a frequency band deviation level sequence is established.

[0080] Based on the offset trajectory mapping set, node deviations are extracted sequentially according to the path order and arranged one by one by the nodes according to their numerical values. During the node arrangement, each node deviation is written into a single-column array, which is typically 512 records long. Each record contains a node deviation value ranging from 1 dB to 40 dB, a time index ranging from 0 ms to 320 ms, and a frequency band number ranging from 1 to 8. The arrangement operation uses the node deviation value as the sorting key and performs the arrangement in ascending order. After the arrangement is completed, the arranged sequence is divided into a fixed number of equal segments, with a preset number of segments of 8 segments and a set of 64 records per segment. Specifically, the 512 records are assigned to segment 1 from number 1 to 64, segment 2 from number 65 to 128, and so on up to segment 8. When assigning segment numbers, the order in which the segment appears is used as the numbering basis, and each segment is assigned a number in the segment arrangement order, with a numbering range from 1 to 8. All segment numbers and corresponding node deviation records are written into a number set and sorted according to the segment order to generate a segment number set.

[0081] Based on the segment number set, the numbers are connected sequentially according to their size, using the number field in the number set as the connection key. The number range is from 1 to 8, and the numbers are rearranged in ascending order. During the rearrangement, the continuity of adjacent segment numbers is checked. Adjacent numbers with a difference of 0 are considered duplicate numbers and removed from the connection sequence by deletion. During deletion, the number position is checked sequentially using an array index, ranging from 1 to 8. If a number is the same as the previous number, the number record is directly cleared to maintain a non-repeating structure in the sequence. When connecting numbers, the remaining numbers are continuously spliced ​​in ascending order according to the array index. During the process, the actual number of numbers in the sequence is written into the output sequence, and the sequence structure is marked as a single number sequence. After the number splicing is completed, the number sequence is written into the level structure according to the original path order. The fields include number value, number order, and time index, and are arranged in a uniform manner. The number sequence is written into the deviation level structure table to establish the frequency band deviation level sequence.

[0082] The specific steps for forming a hearing status number chain are as follows:

[0083] Based on the offset trajectory mapping set and frequency band deviation level sequence, the node deviation values ​​are read sequentially in the path record and each node is compared with the corresponding segment number according to its appearance order. The deviation direction is consistent with the segment direction as the screening basis. The filtered numbers are then sorted according to their original order to obtain the direction screening number group.

[0084] Based on the direction-based numbering group, the number entries are arranged in ascending order of number value. During the arrangement process, duplicate numbers in adjacent positions are removed one by one. The remaining numbers are combined into a single number chain in a continuous connection manner while maintaining the overall order of the chain, thus generating a hearing status number chain.

[0085] Based on the offset trajectory mapping set and frequency band deviation level sequence, the node deviation values ​​are read sequentially from the path record, and each node is compared with the corresponding segment number according to its appearance order. When reading the node deviation, each node is read in the form of three fields: the deviation value is in decibels, ranging from 1 to 40; the deviation direction is represented by the values ​​1 and 2 to indicate rising and falling; and the path order is recorded in the form of milliseconds, ranging from 0 to 320. When comparing the segment numbers, the deviation value of each record is matched with the number field in the segment number set, ranging from 1 to 8, and the matching operation is performed according to the appearance order of the nodes. When filtering the deviation direction, the node direction code 1 is considered to be consistent with the segment direction code 1, and the direction code 2 is considered to be consistent with the segment direction code 2. When the directions are consistent, the node number is written into the filtering list. When sorting the number order, the filtering list is arranged from beginning to end according to the appearance order of the nodes and sorted by the node path record position. After sorting, each number record is written into the number group list to generate the direction filtering number group.

[0086] Based on direction-based numbering, the numbered entries are arranged in ascending order of their numerical values. During the numbering process, the number records are loaded as a single-column array, typically between 64 and 256 elements. The numerical values ​​range from 1 to 8 and are written as integers. During the arrangement, duplicate numbers are removed one by one. Specifically, the number array is scanned continuously from the first position to the last. If a number is identical to the previous one, it is deleted from the array structure, and subsequent numbers are shifted forward one position. After the duplicate number removal step, the remaining numbers are combined into a single number chain. The combination is performed sequentially from left to right, and each number position in the chain is marked with a serial number ranging from 1 to the total number of remaining entries. The overall order of the chain is maintained during chain construction. After the numbering is completed, the number chain is written into the final output structure, generating the hearing status number chain.

[0087] The specific steps for outputting the hearing status assessment results are as follows:

[0088] Based on the hearing status number chain, the numbers are read sequentially and the same numbers are counted into the counting unit during reading. After traversing each position, the number of each counting unit is compared one by one, and the peak number is selected from the comparison results to obtain the peak number information set.

[0089] Based on the peak number information set, the name is located in the status name table by number and the corresponding name text is extracted from the table. Spaces and separators in the text are deleted and the characters are arranged by label. After the arrangement is completed, the label is output to obtain the hearing status judgment result.

[0090] Based on the hearing status number chain, the numbers are read sequentially, and identical numbers are counted in the counting unit during reading. The number chain is loaded as an array during the number reading operation, with an array length typically between 64 and 256. The number values ​​range from 1 to 8 and are stored as integers. During counting unit initialization, an independent counting cell is created for each number, indexed by the number value. Numbers 1 to 8 correspond to counting cell positions 1 to 8, respectively. All counting cells are initially set to 0. During counting, the numbers are read sequentially according to the number chain, and the corresponding counting cell value is incremented by one time during reading, with the increment fixed at one time. After processing one number, a counting cell increment is performed. After traversing the number chain entries, the values ​​in the counting cells are compared one by one, from number 1 to 8, using the integer value in the counting cell as the comparison item. When selecting the largest number, the largest value is used as the peak number, and the peak number is written to the peak list, along with the corresponding occurrence count. After completing the peak recording, the results are organized into a structured array containing a peak number field and an occurrence count field, generating a peak number information set.

[0091] Based on the peak number information set, the name is located in the status name table by number and the corresponding name text is extracted from the table. When loading the status name table, it is loaded in a double list structure. List 1 is the number list, containing numbers 1 to 8, and List 2 is the name list, containing the status text corresponding to the number. Each text consists of no more than 8 Chinese characters. When locating by number, the number field in the peak number information set is used as the locating key to perform item-by-item comparison in the number list. When the numbers match, the text record at the same position in the name list is read. When performing text processing, the space characters in the read text are removed with a deletion command, the separators are cleared with a replacement command, and the remaining characters are written into the character array in order. The character array is written with characters in order from left to right, and the index range is from 1 to the text length. When arranging the tags, the characters in the character array are combined into a single tag string without gaps, and the tag is written into the output buffer. After the character combination is completed, the tag string is written into the output structure in a fixed output format to generate the hearing status judgment result.

[0092] Please see Figure 2 A hearing aid hearing status detection system, the system comprising:

[0093] Offset construction module: It acquires sound pressure level and frequency band division through sensors, derives the difference between input and output sound pressure levels and organizes the frequency band order, introduces a residual network to extract the difference structure, fills in the broken segments and rearranges the sequence to generate a structured offset group.

[0094] The trajectory generation module extracts time-sequential offset segments based on structured offset groups and marks their directions. After completing the direction comparison by combining dynamic time warping, it connects the node segments and performs path combination and order sorting on the cross-frequency band segment chain to obtain the offset trajectory mapping set.

[0095] The grade composition module: Based on the offset trajectory mapping set, it captures the node deviation and arranges them by value. The arrangement sequence is divided into segments in an equal manner. The segment contents are labeled with numbers and the number sequence is arranged in order to establish the frequency band deviation grade sequence.

[0096] Numbering module: Based on the offset trajectory mapping set and frequency band deviation level sequence, read the node deviation and perform direction filtering, rearrange the filtered numbers in order and delete consecutive duplicate numbers, and connect the sorted numbers to form a hearing status numbering chain.

[0097] Status output module: Based on the hearing status number chain, it counts the number and records the number of occurrences, extracts the peak number of the statistical results and compares it with the position in the name table, performs formatting and character arrangement on the name text, outputs the arrangement result as a label, and obtains the hearing status judgment result.

[0098] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A hearing aid hearing status detection method, characterized in that, The method comprises the following steps: S1: obtaining the hearing aid microphone input sound pressure level, amplification output sound pressure level and frequency band division information item through the sensor, subtracting the two types of sound pressure levels point by point in each frequency band through the residual network, rearranging the difference value sequence and filling in the missing paragraph mark to generate a structured offset group; S2: based on the structured offset group, extracting the offset sequence in time sequence in each frequency band and recording the rising section and the falling section, comparing the dynamic time warping direction in the same frequency band, connecting the continuous node sequence after the comparison, reconstructing the cross-frequency band path in each frequency band, and obtaining the offset trajectory mapping set; S3: based on the offset trajectory mapping set, extracting the path node deviation in each frequency band, sorting the values, cutting the sorted sequence by a fixed number and giving the section number, connecting the section numbers in order to form a unified sequence, and establishing a frequency band deviation level sequence; S4: based on the offset trajectory mapping set and the frequency band deviation level sequence, matching the path node deviation value with the level section number item by item and screening out the inconsistent direction nodes, rearranging the matching results in order by number and removing the continuous repeated numbers to establish a unified number sequence, forming a hearing status number chain; S5: based on the hearing status number chain, counting the number of occurrences in each number and determining the peak value number, mapping the peak value number with the corresponding name in the state name table, arranging the mapping name text and keeping a single label output, and outputting the hearing status determination result; The residual network first arranges the difference value sequence in the order of frequency band, inputs the sequence into the layer-by-layer calculation structure of the network in the input order, maintains the original amount of input difference value in each layer structure and superimposes the difference value offset calculated by the current layer, continues to pass the superimposed result to the next layer structure, outputs the difference value structure information corresponding to the original difference value sequence at the end of the structure, and adds the difference value structure information to the difference value item arrangement process to form the frequency band difference value original set; The dynamic time warping first calibrates the connection relationship of the continuous node chain group, receives a plurality of chain sequences arranged in the starting order, inputs the node difference value sequence of each chain in turn into the warping process according to the current position, calculates the distance between nodes in the sequence comparison process and records the moving direction of each step in the distance accumulation process, uses the recorded path trajectory to locate the alignment position of the difference value between the chains, and re-corresponds the chain node order according to the path trajectory, so that the connection relationship between the chains is calibrated under the same warping path.

2. The hearing aid hearing status detection method of claim 1, wherein, The structured offset group includes the difference value sequence arranged by frequency band, the filled paragraph mark and the rearranged sequence order, the offset trajectory mapping set includes the cross-frequency band path node, the node direction identifier and the path connection order, the frequency band deviation level sequence specifically is the sorted node deviation section number, the section division order and the unified connected number sequence, the hearing status number chain includes the screened number item, the rearranged number order and the de-duplicated continuous number chain, and the hearing status determination result specifically refers to the peak value number, the corresponding state name text and the single label output content.

3. The hearing aid hearing status detection method of claim 1, wherein, The specific steps for generating the structured offset group are: The input sound pressure level of a hearing aid microphone, the amplified output sound pressure level and the frequency band division information item are acquired by a sensor, the input value and the output value are corresponded and recorded at the same time position, the difference value is recorded, all the difference value items are arranged in order according to the division, the residual error network is introduced to extract the difference value structure in the arrangement process, and the original set of frequency band difference values is obtained; Based on the original set of frequency band difference values, adjacent items are checked in time sequence and the breakpoint position is marked, the missing position is filled with the average of the previous and next items, the original arrangement order is maintained, the difference value sequence is continuous and complete, and the sequence correction set of frequency bands is obtained; Based on the sequence correction set of frequency bands, the correction items are rearranged in the time dimension, the arranged items are combined into a single set according to the division order, the set content is output, the overall structure is maintained, and the structured offset group is generated.

4. The hearing aid hearing status detection method of claim 1, wherein, The specific steps for obtaining the offset trajectory mapping set are: Based on the structured offset group, adjacent two points are read in time sequence from the offset record, the value greater than zero is marked as an upward segment and the value less than zero is marked as a downward segment, each segment is arranged in the original order after the value greater than zero is marked as an upward segment and the value less than zero is marked as a downward segment, and the paragraph annotation sequence group is obtained; Based on the paragraph annotation sequence group, the amplitude of adjacent paragraphs is extracted in the same sequence, the amplitude difference is close to the paragraph, the start and end order is connected, the direction consistent paragraphs are continuously combined into a node chain, and the continuous node chain group is obtained; Based on the continuous node chain group, the chains are arranged in order according to the chain starting position, and the difference value of the connection point is compared piece by piece, the connection relationship between the chains is calibrated through dynamic time warping, the valley value difference chain is taken as the connection target to form a cross-sequence continuous path structure, and the offset trajectory mapping set is generated.

5. The hearing aid hearing status detection method of claim 1, wherein, The specific steps for establishing the frequency band deviation level sequence are: Based on the offset trajectory mapping set, the node deviation is extracted in path order, the nodes are arranged piece by piece according to the value, a fixed number of arranged sequences are divided into several sections in equal division, each section is assigned a number according to the appearance order, and the section number set is obtained; Based on the section number set, the numbers are connected in order according to the size, the continuity of adjacent section numbers is checked during the connection process, and the redundant numbers are deleted at the repeated numbers, a single number sequence is formed in the arranged order, and the frequency band deviation level sequence is established.

6. The hearing aid hearing status detection method of claim 1, wherein, The specific steps for forming the hearing state number chain are: Based on the offset trajectory mapping set and the frequency band deviation level sequence, the node deviation value is read in the path record, each node is compared with the corresponding section number item according to the appearance order, the deviation direction is consistent with the section direction as the screening basis, the screened numbers are arranged in the original arrangement order, and the direction screening number group is obtained; Based on the direction screening number group, the number items are arranged in order from small to large according to the number value, the adjacent position repeated numbers are removed in the arrangement process, the remaining numbers are combined into a single number chain in a continuous connection manner, and the overall order of the chain is kept consistent, and the hearing state number chain is generated.

7. The hearing aid hearing status detection method of claim 1, wherein, The specific steps for outputting the hearing state judgment result are: Based on the hearing state number chain, read the numbers in order and count the same number when reading, compare the number of each counting unit after traversing each position, and select the number of the peak value from the comparison result to obtain the peak number information set; Based on the peak number information set, locate the name in the state name table according to the number and extract the corresponding name text from the table, delete the text spaces and separators, arrange the characters according to the label, and output the label after the arrangement is completed to form the label, and obtain the hearing state determination result.

8. A hearing aid hearing status detection system, characterized by The hearing aid hearing state detection method according to any one of claims 1-7, the system comprises: Offset modeling module: obtain the collected sound pressure level and frequency band division content through the sensor, derive the input-output sound pressure level difference and arrange the frequency band order, introduce residual network to extract the difference structure, fill in the broken paragraphs and reorder the columns, and generate the structured offset group; Trajectory generation module: based on the structured offset group, extract the time-ordered offset paragraphs and label the direction, combine the dynamic time warping to complete the direction comparison and connect the node paragraphs, perform path combination and order arrangement on the cross-band paragraph chain, and obtain the offset trajectory mapping set; Grade composition module: based on the offset trajectory mapping set, grab the node deviation and arrange it by value, cut the arranged sequence into sections in equal amounts, label the section content with numbers and arrange the number sequence in order, and establish the frequency band deviation grade sequence; Number generation module: based on the offset trajectory mapping set and the frequency band deviation grade sequence, read the node deviation and perform direction filtering, rearrange the filtered numbers in order and delete the continuous repeated numbers, uniformly connect the arranged numbers, and form the hearing state number chain; State output module: based on the hearing state number chain, count the numbers and record the number of occurrences, extract the peak number of the statistical result and compare the name table position, perform format arrangement and character arrangement on the name text, output the arrangement result as a label, and obtain the hearing state determination result.

Images