A neonatal intelligent identity recognition system based on speech recognition

Abstract

The present invention relates to the technical field of intelligent newborn identification technology. The present invention relates to a newborn intelligent identification system based on voice recognition. The system comprises an audio database establishment unit, an audio age calculation unit, a newborn audio prediction unit, a prediction error analysis unit, and an audio identity identification unit. The audio database establishment unit is used to collect historical audio data of newborns, assign a unique identity tag to each newborn, and establish an audio database. After audio data to be identified enters the system, a status determination module identifies the status, and an identity determination module compares the similarity between the predicted audio data and the audio data to be identified, and determines the identity based on the error range. This system can quickly and accurately identify newborns in environments with complex personnel flow, such as hospitals, to prevent safety incidents such as wrongly taken babies.

Description

Technical Field

[0001] The present invention relates to the technical field of neonatal intelligent identity recognition, and more specifically, to a neonatal intelligent identity recognition system based on voice recognition. Background Art

[0002] In the medical field, neonatal identity recognition is of great importance. Currently, the methods adopted include traditional bracelet identification, ankle bracelet identification, etc., which can, to a certain extent, initially distinguish the identities of neonates and help medical staff identify neonatal identities in daily care, examinations, treatments, etc.

[0003] Currently, bracelet and ankle bracelet identifications are prone to being damaged, falling off, or becoming unclear due to neonatal activities, sweating, bathing, etc., resulting in difficulties in identity recognition. Moreover, these identifications mainly rely on manual visual recognition. When medical staff are busy and stressed, misreading and misjudgment are extremely likely to occur, thereby increasing the risk of safety accidents such as taking the wrong neonate, seriously threatening the safety of neonates, and running counter to the purpose of accurately identifying neonatal identities. In order to reduce this situation, a neonatal intelligent identity recognition system based on voice recognition is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a neonatal intelligent identity recognition system based on voice recognition to solve the problems raised in the above background art.

[0005] To achieve the above purpose, a neonatal intelligent identity recognition system based on voice recognition is provided, including an audio database establishment unit, an audio age calculation unit, a neonatal audio prediction unit, a prediction error analysis unit, and an audio identity recognition unit;

[0006] The audio database establishment unit is used to collect historical audio data of neonates, set a unique identity tag for each neonate, and establish an audio database;

[0007] The audio age calculation unit is used to obtain the birth date of each neonate, and at the same time obtain the acquisition date of each piece of historical audio data, and calculate the neonatal age for the historical audio data by combining the birth date and the acquisition date;

[0008] The neonatal audio prediction unit is used to perform status recognition on the historical audio data of each neonate, obtain the status data corresponding to each piece of historical audio data, then extract the historical audio data with the same status data for audio change rate analysis, and predict the audio data of the neonate according to the audio change rate in combination with the historical audio data and the real-time date;

[0009] The prediction error analysis unit is used to calculate the unupdated time for the latest historical audio data and real-time date of each newborn, and then analyze the error range of different states by combining the unupdated time with the average audio change rate of the newborn at different ages;

[0010] The audio identity recognition unit is used to obtain the audio data to be recognized, perform status recognition on the audio data to be recognized, extract the predicted audio data of the same status of the stored newborn according to the recognition result, combine the error range for comparison, and perform newborn identity recognition on the audio data to be recognized according to the comparison result.

[0011] As a further improvement of this technical solution, the audio database establishment unit includes an audio collection module and an audio distribution module;

[0012] The audio collection module is used to collect the historical audio data of newborns, and at the same time obtain the identity information of newborns, and set a unique identity label for each newborn according to the identity information;

[0013] The audio distribution module is used to match the collected historical audio data according to the data source and the newborns, match each segment of historical audio data to the corresponding newborns, and then establish an audio database according to the historical audio data corresponding to each newborn.

[0014] As a further improvement of this technical solution, during the process of collecting historical audio data by the audio collection module, the historical audio data is first preprocessed to remove background noise, and then the continuous audio signal in the audio data is segmented into small audio segments.

[0015] As a further improvement of this technical solution, the audio age calculation unit includes a date acquisition module and an age calculation module;

[0016] The date acquisition module is used to obtain the birth date of each newborn in the audio database according to the identity information of the newborn, and at the same time obtain the acquisition date of each segment of historical audio data;

[0017] The age calculation module is used to calculate the difference between the birth date of the newborn and the acquisition date of the historical audio data, so as to obtain the difference time between the date and the birth date, and then use the difference time as the age of the newborn corresponding to the historical audio data.

[0018] As a further improvement of this technical solution, the newborn audio prediction unit includes a status recognition module, an audio change module, and an audio prediction module;

[0019] The status recognition module is used to perform status recognition on the historical audio data corresponding to each newborn, obtain the status data corresponding to each segment of historical audio data, and classify the historical audio data of each newborn according to the status data;

[0020] The audio change module is used to extract the historical audio data of the same state data of the newborn in the audio database, then sort the historical audio data according to the age of the corresponding newborn, and then analyze the audio change rate of the sorted historical audio data of each state data to obtain the audio change rate of different state data in the different age differences of each newborn;

[0021] The audio prediction module is used to combine the audio change rate of different state data corresponding to each newborn and the historical audio data with the real-time date to predict the audio data, so as to obtain the predicted audio data corresponding to each newborn.

[0022] As a further improvement of the technical solution, during the prediction process, the audio prediction module performs predictions on different states, so that the predicted audio data includes the predicted audio corresponding to different state data of the newborn.

[0023] As a further improvement of the technical solution, the steps for the audio prediction module to obtain the predicted audio data corresponding to each newborn are as follows:

[0024] ;

[0025] where A st，j is the predicted audio of the j-th state data at the real-time time, Δz is the audio change rate, LSTM is the time series prediction model, z t ,z t-1 ,z t-2 ,...,z t-n are the historical audio data corresponding to the newborn, n ∈ [0, +∞) and is an integer, D st is the real-time time.

[0026] As a further improvement of the technical solution, the prediction error analysis unit includes a time difference module and a range analysis module;

[0027] The time difference module is used to extract the latest historical audio data of each newborn closest to the real-time date, and at the same time calculate the difference time by combining the acquisition time of the latest historical audio data with the real-time date to obtain the unupdated time corresponding to each newborn;

[0028] The range analysis module is used to analyze the error range of different states by combining the unupdated time of each newborn with the average audio change rate of different ages, so as to obtain the error range corresponding to the predicted audio data of each newborn in different states;

[0029] The longer the unupdated time, the larger the error range;

[0030] The shorter the time without update, the smaller the error range.

[0031] As a further improvement of the present technical solution, the audio identity recognition unit includes a state determination module and an identity determination module;

[0032] The state determination module is used to obtain audio data to be identified, and then perform state identification on the audio data to be identified, thereby obtaining state data corresponding to the newborn in the audio data to be identified;

[0033] The identity determination module is used to extract the predicted audio data of the same state data according to the state data obtained by the state determination module and perform similarity comparison with the audio data to be identified to obtain the similarity between each segment of the predicted audio data and the audio data to be identified;

[0034] Obtain the audio difference between the predicted audio data with the highest similarity value and the audio data to be identified, and compare the audio difference with the corresponding error range. When the audio difference exceeds the error range, the audio data to be identified is sent to the audio database as a new newborn for identity setting. Conversely, when the audio difference does not exceed the error range, it is determined that the audio data to be identified and the predicted audio data are the same newborn.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] In this newborn intelligent identity recognition system based on voice recognition, after the audio data to be recognized enters the system, the state determination module recognizes the state, and the identity determination module compares the similarity between the predicted audio data and the audio data to be recognized, and judges the identity in combination with the error range. In an environment with complex personnel flow such as a hospital, the identity of the newborn can be quickly and accurately identified to prevent safety accidents such as wrong holding. At the same time, the error range is calculated based on the non-updated time and the average audio change rate. The longer the non-updated time is, the larger the error range is, and vice versa. This dynamic adjustment mechanism makes identity recognition more flexible and accurate, and adapts to the changes in the audio characteristics of newborns over time. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is the overall structural principle diagram of the present invention.

[0038] The meaning of each number in the figure is:

[0039] 10. Audio database establishment unit; 20. Audio age calculation unit; 30. Newborn audio prediction unit; 40. Prediction error analysis unit; 50. Audio identity recognition unit. DETAILED DESCRIPTION

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0041] Please refer to Figure 1 As shown, the purpose of this embodiment is to provide a neonatal intelligent identity recognition system based on speech recognition, including an audio database establishment unit 10, an audio age calculation unit 20, a neonatal audio prediction unit 30, a prediction error analysis unit 40, and an audio identity recognition unit 50;

[0042] The audio database establishment unit 10 is used to collect the historical audio data of newborns, set a unique identity tag for each newborn, and establish an audio database;

[0043] The audio database establishment unit 10 includes an audio collection module and an audio distribution module;

[0044] The audio collection module is used to collect the historical audio data of newborns, obtain the identity information of newborns at the same time, and set a unique identity tag for each newborn according to the identity information;

[0045] Deploy 4 - 8 microphone arrays in the neonatal care area to form a spatial coverage network, and synchronously record audio at a sampling rate of 44.1 kHz and a bit depth of 16 bits to obtain the historical audio data of each newborn;

[0046] Obtain the basic information of newborns (name and date of birth) from the hospital HIS system, and generate a unique identifier through the hash algorithm.

[0047] During the process of collecting historical audio data, the audio collection module first pre - processes the historical audio data to remove background noise, and then divides the continuous audio signal in the audio data into small audio segments.

[0048] The audio distribution module is used to match the collected historical audio data with the newborns according to the data source (newborn source), match each segment of historical audio data with the corresponding newborn, and then establish an audio database according to the historical audio data corresponding to each newborn.

[0049] The audio age calculation unit 20 is used to obtain the date of birth of each newborn, obtain the acquisition date of each segment of historical audio data at the same time, and calculate the age of the newborn for the historical audio data by combining the date of birth with the acquisition date;

[0050] The audio age calculation unit 20 includes a date acquisition module and an age calculation module;

[0051] The date acquisition module is used to obtain the birth date of each newborn in the audio database according to the identity information of the newborn, and at the same time obtain the acquisition date of each piece of historical audio data;

[0052] Extract the birth date from the identity information of the newborn, which can retrieve the birth date of each newborn from the database;

[0053] Similarly, extract the acquisition date of each piece of historical audio data from the audio database. Each piece of audio has a different timestamp, indicating the date when they were collected or recorded;

[0054] The age calculation module is used to calculate the difference between the birth date of the newborn and the acquisition date of the historical audio data, so as to obtain the difference time between the date and the birth date, and then use the difference time as the age of the newborn corresponding to the historical audio data.

[0055] The age of the newborn = the acquisition date of the audio data - the birth date of the newborn;

[0056] Store the calculated difference time as the age annotation of the historical audio data, indicating the age of the newborn corresponding to the audio data at the time of recording.

[0057] The neonatal audio prediction unit 30 is used to perform status recognition on the historical audio data of each newborn, obtain the status data corresponding to each piece of historical audio data, then extract the historical audio data with the same status data for audio change rate analysis, and predict the audio data of the newborn according to the audio change rate in combination with the historical audio data and the real-time date;

[0058] The neonatal audio prediction unit 30 includes a status recognition module, an audio change module, and an audio prediction module;

[0059] The status recognition module is used to perform status recognition on the historical audio data corresponding to each newborn, obtain the status data corresponding to each piece of historical audio data, and classify the historical audio data of each newborn according to the status data. The specific steps are as follows:

[0060] Status recognition: The status in the audio data usually refers to the specific "status" represented by the characteristics of the audio, including emotional status (pain, hunger, sleepiness, etc.), and physiological status (whether the newborn is in a waking, sleeping or crying state);

[0061] Feature extraction: To perform state recognition, features need to be extracted from the audio data. These features can help identify the state of the audio, including Mel Frequency Cepstral Coefficients (used to describe the frequency characteristics of audio and widely used in speech recognition and sentiment analysis), zero-crossing rate (describing the number of zero points in the signal waveform, which can reflect the "roughness" of the sound), spectral centroid (describing the brightness perception of the audio signal), pitch, tone, and volume changes;

[0062] State classification: Based on the above features, a machine learning model can be used for state recognition. The process is as follows:

[0063] Step 1, label each newborn's audio data together with the corresponding state labels to form a training dataset. For example, label whether the audio represents crying, being awake, or other physiological states;

[0064] Step 2, extract features from the audio (such as MFCC, zero-crossing rate, etc.). If the feature dimension is too high, dimensionality reduction methods (such as PCA, LDA, etc.) can be used to reduce the complexity of the feature space;

[0065] Step 3, use a machine learning algorithm (neural network) to classify the states. The input of the model is the features extracted from the audio data, and the output is the corresponding state labels;

[0066] Step 4, use the cross-validation method to evaluate the performance of the model to ensure classification accuracy;

[0067] State classification for each newborn: Process the audio data of each newborn, extract features, and use the trained model to classify the state of the audio data. Each audio segment will be assigned a state label, such as: crying, quiet, hungry, awake, etc. Then, store the audio data and corresponding state labels of each newborn in the database for subsequent query and analysis.

[0068] The audio change module is used to extract historical audio data of the same state data for newborns in the audio database, then sort the historical audio data according to the corresponding newborn age, and then perform audio change rate analysis on the sorted historical audio data of each state data to obtain the audio change rate of different state data in different age differences of each newborn. The specific steps are as follows:

[0069] Extract historical audio data of the same state data: Extract all audio data from the database according to the identity information of the newborns, screen the data according to the state labels of each audio (such as "crying", "quiet", "hungry", etc.), and then store the audio data grouped by their states (such as "crying");

[0070] Sort audio data according to the age of the newborn: For each piece of audio data, calculate the age of the newborn, and sort the historical audio data according to the age of the newborn, arranging them in ascending order of age;

[0071] Audio change rate analysis: After sorting the data for each state, next analyze the audio change rate of the state data for each newborn at different age stages. The formula is as follows:

[0072] ;

[0073] where Δz is the audio change rate, t is the timestamp, z t is the historical audio data at time t, and z t-1 is the historical audio data at time t - 1;

[0074] For each newborn, perform audio change rate analysis on all state data (such as emotions, language, etc.).

[0075] The audio prediction module is used to combine the audio change rates of different state data corresponding to each newborn and the historical audio data with the real-time date for audio data prediction, so as to obtain the predicted audio data corresponding to each newborn.

[0076] During the prediction process, the audio prediction module performs predictions on different states, so that the predicted audio data includes the predicted audio corresponding to different state data of the newborn. The specific formula is as follows:

[0077] ;

[0078] where A st，j is the predicted audio of the jth state data at the real-time time, LSTM is a time series prediction model (long short-term memory network), z t , z t-1 , z t-2 ,..., z t-n are the historical audio data corresponding to the newborn, n ∈ [0, +∞) and is an integer, and D st is the real-time time.

[0079] The prediction error analysis unit 40 is used to calculate the non-updated time by combining the latest historical audio data of each newborn and the real-time date, and then perform error range analysis for different states by combining the non-updated time with the average audio change rate of the newborn at different ages;

[0080] The prediction error analysis unit 40 includes a time difference module and a range analysis module;

[0081] The time difference module is used to extract the latest historical audio data of each newborn closest to the real-time date, and at the same time calculate the difference time by combining the acquisition time of the latest historical audio data with the real-time date to obtain the unupdated time corresponding to each newborn;

[0082] The unupdated time refers to the time difference from the last audio data acquisition time of the newborn to the current date. The larger this time difference is, the older the update time of the audio data is.

[0083] The range analysis module is used to analyze the error range of different states by combining the unupdated time of each newborn with the average audio change rate of different ages, so as to obtain the error range corresponding to the predicted audio data of each newborn in different states;

[0084] The longer the unupdated time is, the larger the error range is;

[0085] The shorter the unupdated time is, the smaller the error range is. The specific formula is as follows:

[0086] ;

[0087] where E is the error range, representing the prediction error of the audio data, and R avg (Δz) is the average audio change rate of different ages, and T wg is the unupdated time;

[0088] The unupdated time T wg The longer it is, the larger the error range E is, because the audio change rate may change over time.

[0089] The audio identity recognition unit 50 is used to obtain the audio data to be recognized, perform state recognition on the audio data to be recognized, extract the predicted audio data of the same state of the stored newborn according to the recognition result and combine the error range for comparison, and perform newborn identity recognition on the audio data to be recognized according to the comparison result.

[0090] The audio identity recognition unit 50 includes a state determination module and an identity determination module;

[0091] The state determination module is used to obtain the audio data to be recognized, and then perform state recognition on the audio data to be recognized (using the same steps as above for state recognition), so as to obtain the state data corresponding to the newborn in the audio data to be recognized.

[0092] The identity determination module is used to extract the predicted audio data of the same status data according to the status data obtained by the status determination module, combine it with the audio data to be recognized, and perform a similarity comparison to obtain the similarity between each piece of predicted audio data and the audio data to be recognized (using similarity metrics such as cosine similarity, Euclidean distance, etc.). The specific steps for comparing the similarity between the audio data to be recognized and the predicted audio data are as follows:

[0093] ;

[0094] where C(Z st , A st，j ) is the similarity between the predicted audio data and the audio data to be recognized, Z st is the audio data to be recognized, 〈Z st , A st，j 〉 is the dot product of the vectors Z st and A st，j , ||Z st || and ||A st，j || are the L2 norms representing the vectors Z st and A st，j .

[0095] By calculating the similarity, the matching degree between the audio data to be recognized and the predicted audio is obtained, and then it is judged whether they are similar.

[0096] Obtain the audio difference between the predicted audio data with the highest similarity value and the audio data to be recognized, and compare the audio difference with the corresponding error range. When the audio difference exceeds the error range, the audio data to be recognized is sent to the audio database as a new neonate for identity setting. Conversely, when the audio difference does not exceed the error range, it is judged that the audio data to be recognized and the predicted audio data are the same neonate.

[0097] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A neonatal intelligent identity recognition system based on speech recognition, characterized in that: It includes an audio database establishment unit (10), an audio age calculation unit (20), a neonatal audio prediction unit (30), a prediction error analysis unit (40), and an audio identity recognition unit (50); The audio database establishment unit (10) is used to collect the historical audio data of newborns, set a unique identity tag for each newborn, and establish an audio database; The audio age calculation unit (20) is used to obtain the birth date of each newborn, and at the same time obtain the acquisition date of each piece of historical audio data, and calculate the age of the newborn for the historical audio data by combining the birth date with the acquisition date; The neonatal audio prediction unit (30) is used to perform status recognition on the historical audio data of each newborn, obtain the status data corresponding to each piece of historical audio data, then extract the historical audio data with the same status data for audio change rate analysis, and predict the audio data of the newborn according to the audio change rate combined with the historical audio data and the real-time date; The prediction error analysis unit (40) is used to calculate the non-update time for the latest historical audio data and the real-time date of each newborn, and then analyze the error range of different states by combining the non-update time with the average audio change rate of the newborn at different ages; The audio identity recognition unit (50) is used to obtain the audio data to be recognized, perform status recognition on the audio data to be recognized, extract the predicted audio data of the existing newborns with the same status according to the recognition result for comparison in combination with the error range, and perform neonatal identity recognition on the audio data to be recognized according to the comparison result.

2. The neonatal intelligent identity recognition system based on speech recognition according to claim 1, characterized in that: The audio database establishment unit (10) includes an audio collection module and an audio allocation module; The audio collection module is used to collect the historical audio data of newborns, and at the same time obtain the identity information of the newborns, and set a unique identity tag for each newborn according to the identity information; The audio allocation module is used to match the collected historical audio data according to the data source and the newborns, match each piece of historical audio data to the corresponding newborn, and then establish an audio database according to the historical audio data corresponding to each newborn.

3. The neonatal intelligent identity recognition system based on speech recognition according to claim 2, characterized in that: During the process of collecting historical audio data, the audio collection module first preprocesses the historical audio data to remove background noise, and then divides the continuous audio signal in the audio data into small audio segments.

4. The neonatal intelligent identity recognition system based on speech recognition according to claim 1, characterized in that: The audio age calculation unit (20) includes a date acquisition module and an age calculation module; The date acquisition module is used to obtain the birth date of each newborn in the audio database according to the identity information of the newborn, and at the same time obtain the acquisition date of each piece of historical audio data; The age calculation module is used to calculate the difference between the birth date of the newborn and the acquisition date of the historical audio data, so as to obtain the difference time between the date and the birth date, and then use the difference time as the age of the newborn corresponding to the historical audio data.

5. The neonatal intelligent identity recognition system based on voice recognition according to claim 1, characterized in that: The neonatal audio prediction unit (30) includes a status recognition module, an audio change module, and an audio prediction module; The state recognition module is used to perform state recognition on the historical audio data corresponding to each newborn, obtain the state data corresponding to each segment of historical audio data, and classify the historical audio data of each newborn according to the state data; The audio change module is used to extract the historical audio data with the same state data for the newborns in the audio database, then sort the historical audio data according to the age of the corresponding newborns, and then perform audio change rate analysis on the sorted historical audio data of each state data to obtain the audio change rate of different state data in the different age differences of each newborn; The audio prediction module is used to combine the audio change rates of different state data corresponding to each newborn and the historical audio data with the real-time date to perform audio data prediction, so as to obtain the predicted audio data corresponding to each newborn.

6. The neonatal intelligent identity recognition system based on voice recognition according to claim 5, characterized in that: During the prediction process, the audio prediction module performs predictions on different states, so that the predicted audio data includes the predicted audio corresponding to different state data of the newborn.

7. An intelligent neonatal identity recognition system based on speech recognition according to claim 5, characterized in that: The steps for the audio prediction module to obtain the predicted audio data corresponding to each newborn are as follows: ； Among them, A st，j is the predicted audio of the j-th state data at the real-time, Δz is the audio change rate, LSTM is the time series prediction model, z t , z t-1 , z t-2 ,..., z t-n are the historical audio data corresponding to the newborn, n ∈ [0, +∞) and is an integer, D st is the real-time.

8. The neonatal intelligent identity recognition system based on voice recognition according to claim 1, characterized in that: The prediction error analysis unit (40) includes a time difference module and a range analysis module; The time difference module is used to extract the latest historical audio data of each newborn closest to the real-time date, and at the same time calculate the difference time by combining the acquisition time of the latest historical audio data with the real-time date, so as to obtain the unupdated time corresponding to each newborn; The range analysis module is used to perform different state error range analysis by combining the unupdated time of each newborn with the average audio change rate of different ages, so as to obtain the error range corresponding to the predicted audio data in different states of each newborn; The longer the unupdated time, the larger the error range; The shorter the unupdated time, the smaller the error range.

9. The neonatal intelligent identity recognition system based on speech recognition according to claim 1, characterized in that: The audio identity recognition unit (50) includes a state determination module and an identity determination module; The state determination module is used to obtain the audio data to be recognized, and then perform state recognition on the audio data to be recognized, so as to obtain the state data corresponding to the newborn in the audio data to be recognized; The identity determination module is used to extract the predicted audio data with the same state data according to the state data obtained by the state determination module, and perform similarity comparison by combining the audio data to be recognized, so as to obtain the similarity between each segment of predicted audio data and the audio data to be recognized; Obtain the audio difference between the predicted audio data with the highest similarity value and the audio data to be recognized, and compare the audio difference with the corresponding error range. When the audio difference exceeds the error range, send the audio data to be recognized as a new newborn to the audio database for identity setting. On the contrary, when the audio difference does not exceed the error range, it is determined that the audio data to be recognized and the predicted audio data are the same newborn.

Images