Biological recognition elevator calling method based on wavelet tight frame

Abstract

The invention discloses a biological recognition elevator calling method based on a wavelet tight frame. Aiming at the problem of poor user experience caused by easy missing detection and false detection of wake-up words and floor instructions in the existing voice elevator calling, a wavelet tight frame (such as redundant wavelet transform) is firstly utilized to perform noise reduction and reconstruction on voice signals in an elevator car, then voiceprint features are extracted in combination with models such as X-Vectors, and an ID distinguishing method is constructed based on a large number of voice-ID samples. Voiceprints are monitored according to the single-floor operation interval of the elevator, the voiceprints and floor matching information are stored in a candidate list according to the inner calling condition, and after the number of times of occurrence of the voiceprints and a corresponding floor proportion threshold value are met, the candidate list is converted into a formal list. And when the voiceprint is detected again, floor calling is automatically matched, and the accuracy can be further improved by combining a content recognition model of the wake-up word and the floor instruction. The elevator calling convenience and accuracy are improved, the intelligent service level of the elevator is enhanced by fusing multiple technologies, and the elevator calling method has high application value in various elevator scenes.

Description

Technical Field

[0001] The invention relates to the technical field of elevator control, in particular to a biometric elevator calling method based on wavelet tight frame. Background Art

[0002] In the field of modern elevator technology, the voice call function has emerged as an important means to improve the intelligence level of elevators and user convenience. Its original intention is to allow passengers to quickly and accurately call the elevator to the designated floor through simple voice commands without manually operating the elevator control panel, so that passengers can easily use the elevator service even when their hands are busy (such as carrying items, pushing wheelchairs, etc.) or it is inconvenient to touch the elevator buttons.

[0003] However, the current voice call elevator system that relies on designated wake-up words and floor instructions has exposed many defects in practical applications. On the one hand, voice recognition technology is greatly affected by environmental factors. In the elevator car, a relatively closed space with high mobility of people, the sources of environmental noise are extensive and complex. For example, the mechanical noise during the operation of the elevator, the airflow sound generated by the car ventilation system, the noise when passengers enter and exit the car, and the mixed voice background formed by multiple people talking or talking at the same time, etc., will cause serious interference to the voice signal. These interferences make it difficult for the voice recognition system to accurately capture the wake-up word and floor instructions, which leads to frequent missed detections, that is, the effective instructions issued by the user are not recognized by the system, and the elevator cannot respond, which seriously affects the user's normal elevator needs.

[0004] On the other hand, even in a relatively quiet environment, the voice recognition system may make false detections due to the diversity and ambiguity of voice commands. Different users have different pronunciation habits, speaking speeds, intonations, and accents, which makes the same wake-up word or floor command show a large range of variation in voice features. For example, the dialect accents in certain regions may cause the pronunciation of the wake-up word to deviate significantly from the standard pronunciation preset by the system, and thus be misjudged as an invalid command by the system; or when the user speaks fast or pronounces unclearly, the floor command may be mistakenly identified as other floors or commands, causing an erroneous elevator call, which not only wastes the elevator's operating resources, but may also cause inconvenience to other passengers.

[0005] In summary, the existing voice elevator calling function cannot meet user needs stably and efficiently in actual use due to the problems of missed detection or false detection of wake-up words and floor instructions, which seriously restricts the improvement of the quality of intelligent elevator services. An innovative technical solution is urgently needed to overcome these drawbacks in order to achieve more accurate, intelligent and convenient elevator calling operations. Summary of the invention

[0006] The purpose of the present invention is to provide a biometric elevator calling method based on wavelet tight frame, which can realize automatic elevator calling based on voiceprint only when the user does not trigger the specified wake-up word or floor command. For example, when the user is talking or talking in the elevator and it is inconvenient to issue a specific voice command, the elevator calling operation can still be performed according to the voiceprint feature matching the corresponding floor, which greatly improves the convenience of users using the elevator and improves the overall user experience.

[0007] In order to achieve the above object, the present invention is implemented by the following technical solution: a biometric elevator calling method based on wavelet tight frame, comprising the following steps:

[0008] Speech Signal Processing

[0009] The present invention firstly uses a wavelet tight frame to decompose the speech signal. The wavelet tight frame has good time-frequency localization characteristics and can decompose the speech signal at different scales and frequencies, thereby effectively extracting the characteristic information of the signal. In the decomposition process, a basis function that meets compact support and redundancy (such as redundant wavelet transform) is selected, so that the adaptability to the signal can be improved while ensuring the accuracy of signal processing.

[0010] The decomposed wavelet coefficients are threshold processed. By setting the threshold reasonably, the wavelet coefficients corresponding to the noise are removed, the main characteristic coefficients of the speech signal are retained, and then the signal is reconstructed to achieve noise reduction and reconstruction of the speech signal. The speech signal after noise reduction processing can effectively reduce the influence of interference factors such as environmental noise on subsequent voiceprint extraction and recognition, and improve the accuracy and reliability of the entire system.

[0011] Voiceprint feature extraction and ID identification

[0012] The reconstructed speech signal is combined with advanced voiceprint extraction models (such as X-Vectors and SpeakerNet models) to extract voiceprint features. These models can deeply mine the unique features of voiceprints from speech signals and convert speech signals into representative voiceprint feature vectors.

[0013] Based on a large number of different voice-ID samples, the distribution differences of voiceprint features corresponding to different IDs in the feature space are statistically analyzed to construct an ID differentiation method based on voiceprint features. This method does not identify specific user IDs, but focuses on distinguishing whether different audios come from different identity individuals, thereby providing an effective basis for identity differentiation for subsequent voiceprint and floor matching without affecting user privacy.

[0014] Voiceprint and floor matching during elevator operation

[0015] The monitoring starts from the time when the elevator door opens on a certain floor and ends before the door opens on the next floor. During this period, when a new voiceprint is matched in the car, voiceprint matching is performed through a specific matching mechanism (such as distance calculation, including Euclidean distance or cosine similarity, etc., to calculate the matching degree between voiceprint features).

[0016] If there is a new internal call during this period, the new floor match and the voiceprint corresponding ID are added to the new entry of the candidate match list; if there is no new internal call and multiple internal calls have been registered, all registered internal calls and the voiceprint corresponding ID are added to the new entry of the candidate match list. In this way, the association between the voiceprint and the possible floor call demand is gradually established.

[0017] In the candidate matching list, certain matching rules are set, that is, when a voiceprint appears more than n times and corresponds to a certain floor m times, if m / n is greater than the preset threshold, it is considered that there is a strong correlation between the voiceprint and the floor, and the pairing combination of the voiceprint and the floor is put into the formal list. The pairing combination in the formal list will be used as the basis for subsequent automatic elevator calls.

[0018] Automatic elevator call

[0019] When a certain voiceprint appears again in the elevator, the system quickly matches the corresponding floor from the official list and automatically registers the internal call to realize the automatic elevator call operation. In addition, when the user triggers the specified wake-up word or floor command, the method of the present invention can be combined with the content recognition model to further improve the accuracy of the automatic elevator call. The content recognition model can perform more accurate analysis and judgment on the user's voice commands, complement the voiceprint recognition results, and further improve the performance and reliability of the entire elevator call system.

[0020] The present invention has the following beneficial effects:

[0021] 1. Improve the convenience of elevator calling: The method of the present invention can realize automatic elevator calling based on voiceprint only when the user does not trigger the specified wake-up word or floor command. For example, when the user is talking or talking in the elevator and it is inconvenient to issue a specific voice command, the elevator calling operation can still be performed according to the voiceprint feature matching the corresponding floor, which greatly improves the convenience of users using the elevator and improves the overall user experience.

[0022] 2. Optimize voiceprint processing accuracy: Decompose and reconstruct voice signals through wavelet tight framework, effectively reduce noise interference in voice signals, and improve the accuracy of subsequent voiceprint extraction. Advanced voiceprint extraction models such as X-Vectors and SpeakerNet further ensure the effectiveness and reliability of voiceprint feature extraction, thus laying a solid foundation for accurate voiceprint matching and floor calling.

[0023] 3. Enhance the reliability of voiceprint matching: The voiceprint feature-based ID differentiation method is constructed using a large number of voice-ID samples. It does not involve the storage of specific user ID information and is only used for matching. Without infringing on user privacy, it accurately distinguishes and statistically analyzes different voiceprint features, combines the multi-stage matching mechanism of the candidate matching list and the formal list, such as the ratio threshold judgment of the number of voiceprint appearances and the number of corresponding floor appearances, etc., which significantly improves the reliability and accuracy of voiceprint and floor matching, reduces the occurrence of mismatching, and makes elevator calling more intelligent and accurate.

[0024] 4. Integrate multiple technologies to improve performance: When the user triggers the specified wake-up word or floor command, the accuracy of automatic elevator calling can be further improved by combining the content recognition model. The voiceprint recognition technology and the traditional voice command recognition technology are organically integrated to give full play to their respective advantages, further improving the performance and adaptability of the entire elevator calling system. It can meet the needs of different user scenarios and usage habits, and effectively improve the intelligent service level of elevators. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0026] Figure 1 It is a flow chart of a biometric elevator calling method based on wavelet tight frame according to an embodiment of the present invention;

[0027] Figure 2 This is the framework intention of a biometric elevator calling method based on a wavelet tight frame in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention.

[0029] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, upper end, lower end, top, bottom...) are only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0030] In the present invention, unless otherwise clearly defined and limited, the term "connection" should be understood in a broad sense, for example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0031] In addition, in the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features; in addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0032] The following is combined with Figure 1 and attached Figure 2 The present invention is described in further detail.

[0033] A biometric elevator calling method based on wavelet tight frame includes the following steps:

[0034] 1. Voice signal acquisition and preprocessing:

[0035] A high-sensitivity microphone array is installed in the elevator car to collect voice signals in the car. The microphone array can collect sounds from different directions and enhance the strength and clarity of the voice signal through signal processing algorithms.

[0036] The collected voice signals are digitized and converted into a digital signal format that can be processed by a computer, and preliminary filtering is performed to remove some obvious high-frequency noise and low-frequency interference signals.

[0037] 2. Speech denoising based on wavelet tight frame:

[0038] The preprocessed speech signal is decomposed by redundant wavelet transform. The appropriate wavelet basis function (such as Daubechies wavelet basis) and the number of decomposition layers (such as 3-5 layers) are selected, and the redundancy (such as 2-3 times redundancy) is determined according to the characteristics of the speech signal.

[0039] Threshold processing is performed on the decomposed wavelet coefficients. The soft threshold or hard threshold method is used to determine the threshold value according to the noise level estimate. For wavelet coefficients whose absolute value is less than the threshold, the soft threshold method sets them to 0, while the hard threshold method directly discards them. Then the signal is reconstructed through inverse wavelet transform to obtain the denoised speech signal.

[0040] 3. Voiceprint feature extraction and ID distinction model training:

[0041] The denoised speech signal is input into the X-Vectors or SpeakerNet voiceprint extraction model. These models are built on deep learning frameworks (such as TensorFlow or PyTorch) and trained with a large amount of speech data.

[0042] Collect a large number of voice-ID sample data of different people, input the voice signal into the voiceprint extraction model to obtain the voiceprint feature vector, and then count the mean, variance and other statistics of the voiceprint feature vectors corresponding to different IDs in various dimensions to build an ID distinction model based on voiceprint features. For example, cluster analysis methods can be used to cluster the voiceprint feature vectors of different IDs, determine the boundary features between different clusters, and thus distinguish whether different audios belong to different IDs.

[0043] 4. Voiceprint and floor matching during elevator operation:

[0044] When the elevator opens its door at a certain floor, the voiceprint and floor matching monitoring program is started and ends before the door of the next floor opens. During this period, the voice signal in the car is continuously collected and processed.

[0045] For the newly collected voice signal, the voiceprint features are extracted and matched with the existing voiceprint template. The distance between the voiceprint features is calculated using Euclidean distance or cosine similarity. When the distance is less than the set matching threshold, it is considered that the new voiceprint is matched.

[0046] If there is a new internal call operation during this period, the newly added floor and the ID corresponding to the matched voiceprint will be added to the new entry in the candidate matching list; if there is no new internal call and multiple internal calls have been registered, all registered internal calls and the ID corresponding to the voiceprint will be added to the candidate matching list.

[0047] The candidate matching list is analyzed regularly, and the number of times each voiceprint appears n and the number of times each floor appears m are counted. When m / n is greater than a preset threshold (such as 0.6), the pairing combination of the voiceprint and the floor is put into the official list.

[0048] 5. Automatic elevator call execution:

[0049] When a certain voiceprint is detected in the elevator again, the system searches for the corresponding floor information in the official list. If a match is found, the internal call operation is automatically registered and the elevator is controlled to go to the corresponding floor.

[0050] When the user triggers the specified wake-up word or floor command, the content recognition model parses the voice command and extracts the wake-up word and floor information. At the same time, the voiceprint recognition system continues to work, matching the voiceprint corresponding to the command with the existing voiceprint template to confirm the user's identity. If both are matched successfully, the corresponding elevator call operation is performed, further improving the accuracy and reliability of the elevator call.

[0051] Example:

[0052] This wavelet tight frame-based biometric elevator calling method is applied in the elevator system of a high-rise office building.

[0053] First, four high-sensitivity microphones are evenly distributed on the top of each elevator car to form a microphone array. The sampling frequency of the microphone array is set to 16kHz, which can effectively collect all-round voice signals in the car.

[0054] After the voice signal is collected, it enters the preprocessing stage. For example, during an elevator operation, the collected original voice signal is mixed with obvious mechanical humming of the elevator and the slight background sound of passengers talking in the car. After digital conversion, a high-pass filter with a cutoff frequency of 50Hz is used to remove the low-frequency mechanical humming, and a low-pass filter with a cutoff frequency of 8kHz is used to filter out the high-frequency weak interference signal to obtain a preliminary purified voice signal.

[0055] Then, speech denoising based on wavelet tight frame was performed. Daubechies wavelet basis was selected, the number of decomposition layers was set to 4, and the redundancy was 2.5 times. Soft threshold processing was applied to the decomposed wavelet coefficients, and the threshold was determined to be 0.3 based on the statistical characteristics of the environmental noise of the office building elevator. After reconstruction, the noise in the speech signal was significantly suppressed, and the originally unclear speech details became clearly discernible.

[0056] Then, voiceprint feature extraction and ID differentiation model training were performed. Voice-ID sample data of 500 different employees in the office building were collected, and voiceprint feature extraction was performed using the X-Vectors model built based on the TensorFlow framework. After a large amount of data training, the constructed ID differentiation model can effectively distinguish the voiceprint features of different employees. For example, when two employees talk in the car, the model can accurately determine that their voiceprints come from different individuals, even though the specific identity information of the employees is unknown.

[0057] During the voiceprint and floor matching stage of the elevator operation, when the elevator opens the door from the 1st floor, three passengers enter the car, and the voice of one of the passengers during the conversation is collected by the system. After processing, a new voiceprint is matched. During the elevator ascent, when it reaches the 3rd floor, a passenger presses the intercom button for the 10th floor. At this time, the system adds the ID corresponding to the 10th floor and the new voiceprint to the candidate matching list. After that, the elevator continues to run, and two more passengers enter the car and talk. During this period, the voiceprint is detected again many times, and there is no new intercom operation, but the intercoms for the 15th and 20th floors have been registered before, so the IDs corresponding to these two floors and the voiceprint are also added to the candidate matching list. When the voiceprint appears 8 times in total, 3 times for the 10th floor, 2 times for the 15th floor, and 3 times for the 20th floor, since 3 / 8>0.6 (preset threshold), the pairing combination of the 10th floor, 15th floor, and 20th floor with the voiceprint is put into the official list.

[0058] Finally, in the automatic elevator call execution phase, when the passenger takes the elevator again and speaks in the car, the system quickly matches the corresponding floor in the official list. If the passenger also triggers the specified wake-up word and the floor command of the 25th floor, the content recognition model accurately interprets the command and combines voiceprint recognition to confirm the identity. The elevator will not only go to the previously matched floor, but also add the 25th floor to the elevator call task sequence, further improving the accuracy and flexibility of elevator calls, and providing efficient and convenient elevator service experience for employees in office buildings.

[0059] It can be seen from this implementation case that the elevator calling method of the present invention can operate effectively in actual complex elevator usage scenarios, overcomes many problems of traditional voice elevator calling, and significantly improves the intelligence level of elevator calling and user experience.

[0060] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. Improve the convenience of elevator calling: The method of the present invention can realize automatic elevator calling based on voiceprint only when the user does not trigger the specified wake-up word or floor command. For example, when the user is talking or talking in the elevator and it is inconvenient to issue a specific voice command, the elevator calling operation can still be performed according to the voiceprint feature matching the corresponding floor, which greatly improves the convenience of users using the elevator and improves the overall user experience.

[0062] 2. Optimize voiceprint processing accuracy: Decompose and reconstruct voice signals through wavelet tight framework, effectively reduce noise interference in voice signals, and improve the accuracy of subsequent voiceprint extraction. Advanced voiceprint extraction models such as X-Vectors and SpeakerNet further ensure the effectiveness and reliability of voiceprint feature extraction, thus laying a solid foundation for accurate voiceprint matching and floor calling.

[0063] 3. Enhance the reliability of voiceprint matching: The voiceprint feature-based ID differentiation method is constructed using a large number of voice-ID samples. It does not involve the storage of specific user ID information and is only used for matching. Without infringing on user privacy, it accurately distinguishes and statistically analyzes different voiceprint features, combines the multi-stage matching mechanism of the candidate matching list and the formal list, such as the ratio threshold judgment of the number of voiceprint appearances and the number of corresponding floor appearances, etc., which significantly improves the reliability and accuracy of voiceprint and floor matching, reduces the occurrence of mismatching, and makes elevator calling more intelligent and accurate.

[0064] 4. Integrate multiple technologies to improve performance: When the user triggers the specified wake-up word or floor command, the accuracy of automatic elevator calling can be further improved by combining the content recognition model. The voiceprint recognition technology and the traditional voice command recognition technology are organically integrated to give full play to their respective advantages, further improving the performance and adaptability of the entire elevator calling system. It can meet the needs of different user scenarios and usage habits, and effectively improve the intelligent service level of elevators.

[0065] The technical solutions provided by the embodiments of the present invention are introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the embodiments of the present invention. The description of the above embodiments is only applicable to help understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, according to the embodiments of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A biometric elevator calling method based on wavelet tight frame, characterized in that: The following steps are involved: Step S1, decomposing the speech signal based on the wavelet tight frame, transforming it using the basis function that meets the tight support and redundancy, reconstructing the signal after threshold processing the wavelet coefficients, and realizing the noise reduction and reconstruction of the speech signal; Step S2, extracting voiceprint features using the reconstructed speech signal combined with a voiceprint extraction model; Step S3: Based on a large number of different voice-ID samples, the voiceprint differences of different IDs are counted, and an ID differentiation method based on voiceprint features is constructed; Step S4, starting from the time when the elevator door opens on a certain floor and ending before the door opens on the next floor, if a new voiceprint is matched in the car during this period, matching is performed based on distance calculation or matching mechanism: if there is a new internal call during this period, the new floor match and the ID corresponding to the voiceprint are added to the new entry of the candidate matching list; if there is no new internal call and multiple internal calls have been registered, all registered internal calls and the ID corresponding to the voiceprint are added to the new entry of the candidate matching list; Step S5: In the candidate matching list, when a voiceprint appears more than n times and corresponds to a certain floor m times, if m / n is greater than a preset threshold, the pairing combination of the voiceprint and the floor is put into the official list; Step S6: When a certain voiceprint appears in the elevator, the corresponding floor is matched from the official list and the internal call is automatically registered.

2. The biometric elevator calling method based on wavelet tight frame according to claim 1 is characterized by: The voiceprint extraction model includes but is not limited to an X-Vectors model and a SpeakerNet model.

3. The biometric elevator calling method based on wavelet tight frame according to claim 1 is characterized by: The ID distinguishing method of the voiceprint feature only distinguishes whether the audio belongs to different IDs, and does not identify the specific ID.

4. The biometric elevator calling method based on wavelet tight frame according to claim 1 is characterized by: The wavelet tight framework uses redundant wavelet transform to decompose and reconstruct speech signals.

5. The biometric elevator calling method based on wavelet tight frame according to claim 1 is characterized by: The ID differentiation method based on voiceprint features constructs a differentiation model by statistically analyzing the distribution differences of voiceprint feature vectors in different dimensions.

6. The biometric elevator calling method based on wavelet tight frame according to claim 1 is characterized by: The distance calculation in the matching mechanism uses Euclidean distance or cosine similarity to calculate the matching degree between voiceprint features.

7. The biometric elevator calling method based on wavelet tight frame according to claim 1 is characterized by: When the user triggers a specified wake-up word or floor command, the content recognition model is used to improve the accuracy of automatic elevator calls.

Images