Passenger gender identification method for online car-hailing, server and medium

Abstract

The invention provides an online car-hailing passenger gender identification method, a server and a medium, and relates to the technical field of data processing, and the method comprises the steps: obtaining a target audio in an online car-hailing after the online car-hailing arrives at a boarding point of an order; detecting whether the target audio contains a preset number of continuous numbers or not; if yes, acoustic features are obtained from the target audio, context information influencing gender judgment is obtained, and whether the target audio contains female sound or not is judged according to the context information and the acoustic features; and judging whether to trigger a protection strategy for the passenger or not according to the judgment result and the sex of the online car-hailing driver. According to the embodiment of the invention, the sex of the passenger can be indirectly deduced by judging whether the detected audio contains the female voice and combining the known sex information of the driver to construct a reverse deduction mechanism, so that the multi-sound-source interference is eliminated, and the recognition precision in a complex scene is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, server, and medium for identifying the gender of passengers in ride-hailing services. Background Technology

[0002] In ride-hailing scenarios, ensuring passenger safety, especially for female passengers, is one of the core industry requirements. Accurately identifying passenger gender is a key prerequisite for triggering targeted protection strategies. Currently, the ride-hailing industry lacks efficient and accurate passenger gender identification solutions.

[0003] In relevant scenarios, drivers need to verbally confirm the passenger's gender, but this method not only affects the travel experience but may also lead to misjudgment due to poor communication. Another solution attempts to directly identify gender through voice features, but the complex environment inside ride-hailing vehicles is easily affected by road noise and interference from in-vehicle equipment, resulting in low accuracy. Furthermore, the passenger's voice after boarding often contains irrelevant information; directly performing gender identification on the complete audio will further reduce accuracy as effective features are masked by redundant information. Summary of the Invention

[0004] This invention provides a method, server, and medium for passenger gender recognition in ride-hailing services. It can determine whether the submitted audio contains female voice and, in conjunction with the driver's known gender information, construct a reverse inference mechanism to indirectly infer the passenger's gender, thereby eliminating interference from multiple sound sources and improving recognition accuracy in complex scenarios.

[0005] In a first aspect, embodiments of the present invention provide a method for passenger gender identification in ride-hailing services, comprising: Obtain the target audio from inside a ride-hailing vehicle; Identify keywords in the target audio; and detect whether the audio following the position of the keywords in the target audio contains a preset number of consecutive numbers; If so, acoustic features are obtained from the target audio, and contextual information that affects gender determination is obtained. Using the contextual information and the acoustic features, it is determined whether the target audio contains a female voice. Based on the judgment result and the gender of the ride-hailing driver, determine whether to trigger the protection strategy for passengers.

[0006] The above method can collect target speech in the vehicle, filter whether the target speech contains specific numbers by using keywords in the target speech, and determine whether it contains female voices by combining acoustic features and contextual information. This can classify more accurately in noisy environments. When a female voice is identified in the current audio, it can be combined with the driver's fixed gender label to perform reverse reasoning, eliminate mixing interference, and improve the recognition accuracy in complex scenarios.

[0007] In one possible implementation, using the contextual information and the acoustic features to determine whether the target audio contains a female voice includes: The contextual information is mapped into the same dimensional space as the acoustic features through an embedding layer; By connecting the beginning and end of the signal, the mapped contextual information is concatenated with the acoustic features to obtain the fused features; The fused features are input into the recognition model, and the recognition model outputs whether the target audio contains a female voice.

[0008] In one possible implementation, determining keywords in the target audio includes: The target audio is input into the translation model, and the translation model outputs multiple words in the target audio and the sound start time of each word; Determine the order cancellation time; Words whose sound start time is less than the order placement time are selected as keywords in the target audio.

[0009] In one possible implementation, detecting whether the audio following the position of the keyword in the target audio contains a predetermined number of consecutive digits includes: The sound end time of the keyword is taken as the position of the keyword; wherein, the sound end time of the keyword is output by the translation model; A preset character following the sound end time of the keyword in the target audio is identified as a candidate character; Detect whether the candidate characters contain a preset number of consecutive digits.

[0010] In one possible implementation, obtaining acoustic features from the target audio includes: Extract the preset number of consecutive audio segments from the target audio; The extracted audio segments are copied multiple times to obtain processed audio; wherein the duration of the processed audio exceeds a preset duration. The processed audio is input into the extraction model, and the extraction model outputs acoustic features.

[0011] In one possible implementation, acquiring the target audio within the ride-hailing vehicle includes: Receive the initial audio from inside the ride-hailing vehicle; The initial audio from the ride-hailing vehicle is divided into frames of fixed frame length to obtain multiple segments; Identify segments from multiple clips where the probability of human voices exceeds a first threshold; Find out whether the distance between the voice of a segment exceeding the first threshold and the adjacent segment exceeds the second threshold; The two segments that exceed the second threshold are combined in chronological order to obtain the target audio.

[0012] Secondly, embodiments of the present invention provide a method for passenger gender identification in ride-hailing services, applied to a server, comprising: Receive at least one voice packet sent by the driver's terminal of the ride-hailing vehicle, and assemble the target audio according to the time sequence of the at least one voice packet; Identify keywords in the target audio; and detect whether the audio following the position of the keywords in the target audio contains a preset number of consecutive numbers; If so, acoustic features are obtained from the target audio, and contextual information that affects gender determination is obtained. Using the contextual information and the acoustic features, it is determined whether the target audio contains a female voice. Based on the judgment result and the gender of the ride-hailing driver, determine whether to trigger the protection strategy for passengers.

[0013] Thirdly, embodiments of the present invention provide a method for passenger gender identification in ride-hailing services, applied to a terminal, including: Collect initial audio from inside the ride-hailing vehicle; The initial audio within the ride-hailing vehicle is divided into segments of fixed frame length; For each segment, determine whether the probability of human voice in each segment exceeds a threshold; The segments exceeding the threshold are combined in chronological order to obtain at least one voice packet shorter than a preset length; at least one voice packet is sent to the server so that the server can use the voice packet to determine the passenger's gender.

[0014] Fourthly, embodiments of the present invention provide a server, comprising: processor; A processor for executing a computer program or instructions in the memory, such that the passenger gender identification method for ride-hailing services as described in any of the first aspects is executed.

[0015] Fifthly, embodiments of the present invention provide a computer-readable storage medium that, when instructions in the storage medium are executed by a processor, enables the processor to perform a passenger gender identification method for ride-hailing services as described in any of the first aspects.

[0016] In a sixth aspect, embodiments of the present invention provide a computer program product, the computer program product comprising: computer program code, which, when run on a computer, causes the computer to execute the passenger gender identification method for ride-hailing services as described in any of the first aspects.

[0017] Furthermore, the technical effects of any of the implementation methods in aspects two through six can be found in the technical effects of different implementation methods in aspect one, and will not be repeated here.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the interaction between a terminal and a server, provided as an embodiment of the present invention. Figure 2 A flowchart illustrating a method for passenger gender identification in ride-hailing services, provided as an embodiment of the present invention; Figure 3 A flowchart illustrating a method for determining target audio according to an embodiment of the present invention; Figure 4 A flowchart illustrating a method for determining keywords according to an embodiment of the present invention; Figure 5 A flowchart illustrating a method for detecting a preset number of consecutive digits in a target audio file, as provided in an embodiment of the present invention; Figure 6 A flowchart illustrating an acoustic feature extraction method provided in an embodiment of the present invention; Figure 7 A flowchart illustrating a method for identifying whether a target audio file contains a female voice, as provided in an embodiment of the present invention. Figure 8 A flowchart illustrating a passenger gender recognition method for ride-hailing vehicles applied to a terminal, provided as an embodiment of the present invention; Figure 9 A flowchart illustrating a method for passenger gender recognition in a ride-hailing service applied to a server, provided as an embodiment of the present invention; Figure 10 This is a structural diagram of a passenger gender recognition device for ride-hailing services provided in an embodiment of the present invention. Detailed Implementation

[0020] 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. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0021] The preferred embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for illustration and explanation of the present invention and are not intended to limit the present invention. Furthermore, the embodiments of the present invention and the features therein can be combined with each other without conflict.

[0022] The present invention provides a method for identifying the gender of ride-hailing passengers, applicable to gender identification scenarios in ride-hailing services, and combined with... Figure 1 As shown, the main entities performing the gender recognition scenario include multiple terminals 1~n and server 200.

[0023] Taking terminal 100 as an example, terminal 100 includes smart terminals for ride-hailing drivers (such as smartphones and in-vehicle terminals).

[0024] Server 200 can be an independent physical server, a server cluster composed of multiple physical servers, or a distributed system that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. Among them, the server can be connected to devices with display functions via wired or wireless means, and this invention does not limit the connection method.

[0025] Combination Figure 1 As shown, in a scenario where server 200 is the main execution entity, terminal 100 is only responsible for audio acquisition and sending it to server 200 through the communication network. Server 200 completes core steps such as target audio combination, feature extraction, and gender determination, and then feeds the results back to the terminal to trigger protection strategies (applicable to most conventional terminal devices).

[0026] The server used in the passenger gender recognition method for ride-hailing services provided in this embodiment of the invention will be described in detail below with reference to the accompanying drawings: Figure 2 This diagram illustrates the workflow of a passenger gender identification method for ride-hailing services provided by an embodiment of the present invention. Figure 2 As shown, the specific process of this method is as follows: S210: Obtain the target audio from inside the ride-hailing vehicle.

[0027] For example, when a ride-hailing driver arrives at the designated pick-up point (also known as the order's starting point), the driver clicks the "Arrived at Pick-up Point" button on the terminal. The terminal then automatically activates its recording function, sending the recorded audio to the server. The server then executes... Figure 2 The method.

[0028] S220: Determine the keywords in the target audio; and in the audio following the position of the keywords in the target audio, check whether it contains a preset number of consecutive numbers; if so, execute S230; otherwise, end.

[0029] For example, a preset number of consecutive digits are used as the last four digits of the phone number, and keywords such as "last four digits," "phone number," and "last four digits" are used to search for common terms related to the last four digits of the phone number.

[0030] In detail, the process involves identifying keywords such as "last four digits," "phone number," and "last four digits" from the target speech, determining the position of these keywords within the target speech, and then detecting whether subsequent audio contains four consecutive digits, such as the last four digits of a phone number.

[0031] The target audio is "Hello, please fasten your seatbelt and provide the last four digits of your phone number. The last four digits of your phone number are 1234." When executing step 220, first identify "phone number" and "last four digits" as keywords. Locate the audio following the keywords "phone number" and "last four digits." Check if the audio containing "phone number" contains a consecutive set number of digits. If "1234" is detected, proceed to step 430. If not, it means the audio does not contain the last four digits of a phone number, and the process can end.

[0032] S230: Obtain acoustic features from the target audio and obtain contextual information that affects gender judgment. Use the contextual information and acoustic features to determine whether the target audio contains a female voice.

[0033] For example, contextual information that influences gender judgment could be places women often visit, such as shopping malls. When the pick-up point of an order is known, it is determined whether the latitude and longitude of the pick-up point fall within a preset area. This preset area could be places women frequently visit, such as shopping malls, beauty salons, etc. If the latitude and longitude of the pick-up point fall within the preset area, then this preset area can be used as contextual information that influences gender judgment. If it does not fall within the preset area, then the contextual information that influences gender judgment is 0.

[0034] S240: Based on the judgment result and the gender of the ride-hailing driver, determine whether to trigger the protection strategy for passengers.

[0035] In detail, if the target audio contains a female voice and the ride-hailing driver is male, the female voice likely comes from a passenger, therefore the passenger is determined to be female. In this case, if both the driver and passenger are determined to be male, passenger protection strategies can be triggered. If the target audio contains a female voice and the driver is female, it's impossible to determine whether the voice originated from the driver or the passenger, requiring further investigation using other methods. If the target audio does not contain a female voice and the driver is male, the only voice in the vehicle is male, indicating the passenger is male, and therefore no female-specific protection strategy needs to be activated.

[0036] In some embodiments, combined with Figure 3 As shown, the specific implementation method for obtaining the target audio from the ride-hailing vehicle in step 210 is as follows: S310: Receives initial audio from inside the ride-hailing vehicle; S320: The initial audio from the ride-hailing vehicle is divided into frames according to a fixed frame length to obtain multiple segments; S330: Identify segments from multiple segments where the probability of human voices exceeds the first threshold; S340: Determine whether the distance between the voice of a segment exceeding the first threshold and the adjacent segment exceeds the second threshold; S350: Combine the pairs of segments that exceed the second threshold in chronological order to obtain the target audio.

[0037] For example, when the driver arrives at the pick-up point of the order, the trip begins recording audio. The original recording is segmented in real time into short audio segments of fixed length of 1 minute, such as audio_1.aac and audio_2.aac, and the format is uniformly 16kHz sampling rate and PCM 16-bit format to adapt to the input of the VAD model. The VAD model selected here is the silero-vad model.

[0038] A lightweight silento-vad model (approximately 1MB) is pre-integrated, supporting streaming processing. Regarding the key parameter configuration of the silento-vad model, the main considerations are human voice recall and sentence coherence. Therefore, the first threshold, which is the threshold for detecting whether a segment contains human voice, is set to 0.35 (a low threshold, improving the human voice detection rate); the second threshold, which is the threshold for whether two segments are the same sentence, or the longest silence interval (max_silence_duration), is set to 2 seconds. The 2-second setting allows for short pauses between segments, avoiding interruptions in dialogue; optionally, to filter transient noise, the minimum speech duration, i.e., the segment (min_speech_duration), can be set to 500ms (filtering transient noise).

[0039] The silero-vad model processes the initial audio from the ride-hailing vehicle by dividing it into frames of a fixed length, for example, 1536 frames, approximately 96ms / frame, resulting in multiple segments. It calculates the probability of human voices in each segment. If the probability exceeds 0.35, the segment is considered to contain human voices and can proceed to the next step; otherwise, it is considered not to contain human voices and will not proceed. Based on the start and end timestamps of each segment, it determines the adjacent segments for the segment proceeding to the next step and calculates the human voice interval between the segment and the adjacent segments. Of course, adjacent segments must also contain human voices. If the interval between two human voice segments is less than or equal to 2 seconds, they are merged into a single segment. The start and end timestamps of each segment described above, such as (start, end) and total duration (e.g., [start: 10.2s, end: 25.5s, duration: 16.3s], are used. The original duration was 15.3 seconds, but an extra second was added to insert a one-second fade-in / fade-out silence to avoid popping sounds. Segment information is temporarily stored locally in a structured format in data.json, and audio segments are temporarily stored locally as files. data.json is continuously updated. Of course, the merged segments in this application will also record start and end timestamps, in the same format as the start and end timestamps of the segments described above.

[0040] For example, if the time of the segment exceeding the second threshold is 10:00:00~10:00:16, the time of the second segment is 10:00:17~10:00:26, and the time of the third segment is 10:00:28~10:00:36, the segments can be combined in chronological order, starting with the second segment after the first segment ends, and so on.

[0041] In some embodiments, combined with Figure 4As shown, the specific implementation method for determining the keywords in the target audio in step 220 is as follows: S410: Input the target audio into the translation model, and the translation model outputs multiple words in the target audio and the start time of the sound of each word; S420: Determine the order cancellation time; S430: Select words whose sound start time is less than the order placement time as keywords in the target audio.

[0042] The translation model is ASR. After the target audio is input into ASR, ASR performs transcription and timestamp processing. ASR outputs multiple words with timestamps, the start time of the sound of each word, and the end time of the sound of each word. Output example: [{'word': 'last digit', 'start': 12.3, 'end': 12.8}, ...].

[0043] The order confirmation time is obtained, which is the time when the driver enters the last four digits of the passenger's phone number on their mobile phone and clicks confirm. Audio clips submitted for inspection are filtered based on this time point, primarily considering that the passenger provides the last four digits of their phone number first, the driver receives the last four digits, enters the information, and confirms the order. Therefore, only audio clips whose start time is before the order confirmation time are considered. If the start time of a word is before the order confirmation time, that word is considered a keyword; otherwise, it is filtered out.

[0044] In some embodiments, combined with Figure 5 As shown, the specific implementation of step 220, which involves detecting whether the audio following the keyword in the target audio contains a preset number of consecutive digits, is as follows: S510: Use the sound end time of the keyword as the position of the keyword; where the sound end time of the keyword is output by the translation model; S520: Determine the preset character after the sound end time of the keyword in the target audio as the candidate character; S530: Detect whether the candidate characters contain a preset number of consecutive digits.

[0045] For example, keywords such as "last four digits," "phone number," and "last four digits" are used. A phone number's last four digits are any four consecutive digits from 0 to 9. When the target audio contains these three keywords, "last four digits" can be processed first. The 20 characters following the last four digits of the target audio are considered as preset characters. The system checks if four consecutive digits exist within these preset characters. If four consecutive digits exist, the start and end times of the last four digits are obtained through timestamp mapping (e.g., "1234" corresponds to the time interval [15.2s, 16.0s]). If no four consecutive digits are found, the system checks if four consecutive digits exist within the 20 characters following "phone number." If so, the start and end times of the last four digits are obtained through timestamp mapping. If no four consecutive digits are found, the system checks if four consecutive digits exist within the 20 characters following "last four digits." If so, the start and end times of the last four digits are obtained through timestamp mapping.

[0046] In some embodiments, combined with Figure 6 As shown, the specific implementation method for obtaining acoustic features from the target audio in step 230 is as follows: S610: Extract a preset number of consecutive audio segments from the target audio; S620: Copy the extracted audio segments multiple times to obtain processed audio; wherein the duration of the processed audio exceeds the preset duration; S630: Input the processed audio into the extraction model, and extract the acoustic features output by the model.

[0047] For example, when the last digit of the mobile phone number is detected in step 530 above, the start and end times of the last digit of the mobile phone number in the target audio are recorded. Based on the start and end times of the last digit, the corresponding audio segment is extracted from the target audio. Since the audio here is generally short, about one second, the amount of information is too small and the model cannot recognize it well. Therefore, the audio is copied and the audio segment is copied to a new audio of no less than 5 seconds. This new audio is the processed audio and the preset duration can be 5 seconds.

[0048] The extraction model is the HuBERT-Large model, which has 768 hidden layers. The processed audio is input into the HuBERT-Large model, which processes the audio and extracts frame-level acoustic features.

[0049] In some embodiments, combined with Figure 7 As shown, the specific implementation of step 230, which uses contextual information and acoustic features to determine whether the target audio contains a female voice, is as follows: S710: Maps contextual information into the same dimensional space as acoustic features through an embedding layer; S720: By connecting the beginning and end of the sequence, the mapped contextual information and acoustic features are spliced ​​together to obtain the fused features; S730: Input the fused features into the recognition model, and the recognition model outputs whether the target audio contains a female voice.

[0050] For example, the recognition model can be a multimodal classification model. Acoustic features combined with common location information (such as stations, shopping malls, etc.) are used as input to the multimodal classification model. The location information is mapped to a fixed dimension (32-dimensional) through an embedding layer, and the length of the embedding space is the number of elements in the common location set. It is then concatenated with the acoustic features to form a joint representation (the location features (e.g., 32-dimensional) are directly concatenated with the acoustic features of each frame (768-dimensional), that is, the acoustic features and the mapped contextual information are linked end to end to form a fused feature representation (800-dimensional)).

[0051] The fused features are input into subsequent classification networks (such as fully connected layers and Transformer layers). The model automatically learns the association between the two features, such as "noise features of the station environment + specific acoustic features = real female voice". Classification objective: Do not directly identify gender, but only output a binary classification result of "is it a real female voice" or "no", avoiding the gender recognition problem in multi-person scenarios.

[0052] By fusing features in this way, the model can combine acoustic features and contextual information to more accurately determine whether the submitted audio contains a real female voice. Thus, by fusing location information, this embodiment of the invention significantly improves the model's classification accuracy in noisy environments.

[0053] It should be noted that, in addition to direct concatenation, multimodal classification models can also employ weighted fusion when concatenating features. This involves assigning different weights to acoustic features and contextual information (e.g., acoustic features weighted at 0.8, location features weighted at 0.2) to highlight the more important modality. Cross-fusion is also possible, using network layers such as Transformers and attention mechanisms to allow interaction between two types of features, such as the association between "station scene" and "specific female voice acoustic features," to uncover deeper connections.

[0054] It should be noted that feature mapping and alignment are required when processing multimodal classification models, namely, a unified feature format and temporal / logical alignment. A unified feature format maps acoustic features and contextual information to a unified dimensional space, as mentioned in this application. Temporal / logical alignment means that if there is a temporal correlation, such as each audio frame corresponding to a location, the modal features must be matched temporally / logically. For example, each frame of audio features in a scene is bound to its corresponding location. Since this application only processes one audio file and one location per order, no binding is required; it is sufficient to know that the two are a match.

[0055] Combination Figure 1 As shown, terminal 100 is responsible for audio acquisition. After preprocessing the speech, it is sent to server 200 via the communication network. Server 200 completes core steps such as target audio combination, feature extraction, and gender determination, and then feeds the results back to the terminal to trigger protection policies (applicable to most conventional terminal devices). Terminal 100 executes the following... Figure 8 The solution shown will cause server 200 to execute the following: Figure 9 The aforementioned solution.

[0056] Combination Figure 8 As shown, this embodiment of the invention provides a second method for passenger gender recognition in ride-hailing services, applied to a terminal, including: S810: Captures initial audio from inside ride-hailing vehicles; S820: Divide the initial audio from the ride-hailing vehicle into frames according to a fixed frame length to obtain multiple segments; S830: Identify segments from multiple segments where the probability of human voices exceeds a first threshold; S840: Determine whether the distance between the voice of a segment exceeding the first threshold and the adjacent segment exceeds the second threshold; S850: Combine the pairs of segments that exceed the second threshold in chronological order to obtain at least one voice packet with a length less than a preset length; send the at least one voice packet to the server so that the server can use the at least one voice packet to determine the passenger's gender.

[0057] Figure 8 Steps 810 to 840 and Figure 3 Steps 310 and 340 are the same; please refer to the following for details. Figure 3 The contents of steps 310 and 340.

[0058] For example, when assembling a voice packet, the preset length is 60 seconds. During the combination process, if the cumulative voice duration is >= 60 seconds (e.g., the first n-1 segments total 58 seconds, and the nth segment adds up to 63 seconds), then the first n-1 voice segments are merged to generate a voice packet, which is then uploaded to the cloud for subsequent server download. The merged segment data, including data.json and the audio segment file, is then cleared, and the nth segment is retained for the next round of accumulation until the initial audio ends.

[0059] Combination Figure 9 As shown, this embodiment of the invention provides a second method for passenger gender identification in ride-hailing services, applied to a server, including: S910: Receives at least one voice packet sent by the driver's terminal of the ride-hailing vehicle, and assembles the target audio according to the time sequence of the at least one voice packet.

[0060] For example, if the first voice packet ends at 10:00:00, the second voice packet ends at 10:00:53, and the third voice packet ends at 10:01:50, then when combining the three voice packets, the first voice packet can be combined with the second voice packet, and the second voice packet can be combined with the third voice packet.

[0061] S920: Determine the keywords in the target audio; and in the audio following the position of the keywords in the target audio, check whether it contains a preset number of consecutive numbers; if so, execute S930, otherwise end.

[0062] S930: Obtain acoustic features from the target audio and obtain contextual information that affects gender judgment. Use the contextual information and acoustic features to determine whether the target audio contains a female voice.

[0063] S940: Based on the judgment result and the gender of the ride-hailing driver, determine whether to trigger the protection strategy for passengers.

[0064] Optionally, using contextual information and the acoustic features, determining whether the target audio contains a female voice includes: The contextual information is mapped into the same dimensional space as the acoustic features through an embedding layer; By connecting the beginning and end of the signal, the mapped contextual information is concatenated with the acoustic features to obtain the fused features; The fused features are input into the recognition model, and the recognition model outputs whether the target audio contains a female voice.

[0065] Optionally, determining keywords in the target audio includes: The target audio is input into the translation model, and the translation model outputs multiple words in the target audio and the sound start time of each word; Determine the order cancellation time; Words whose sound start time is less than the order placement time are selected as keywords in the target audio.

[0066] Optionally, in the audio following the position of the keyword within the target audio, it is detected whether a preset number of consecutive digits are contained, including: The sound end time of the keyword is taken as the position of the keyword; wherein, the sound end time of the keyword is output by the translation model; A preset character following the sound end time of the keyword in the target audio is identified as a candidate character; Detect whether the candidate characters contain a preset number of consecutive digits.

[0067] Optionally, acoustic features are obtained from the target audio, including: Extract the preset number of consecutive audio segments from the target audio; The extracted audio segments are copied multiple times to obtain processed audio; wherein the duration of the processed audio exceeds a preset duration. The processed audio is input into the extraction model, and the extraction model outputs acoustic features.

[0068] For details, please refer to the above. Figures 2-7 Content (excluding) Figure 3 ).

[0069] like Figure 10 As shown, the present invention also provides a passenger gender recognition device for ride-hailing services, comprising: The acquisition module 1010 is used to acquire the target audio inside the ride-hailing vehicle after the ride-hailing vehicle arrives at the pick-up point of the order; The detection module 1020 is used to determine the keywords in the target audio; and to detect whether the audio following the position of the keywords in the target audio contains a preset number of consecutive numbers. The judgment module 1030 is used to, if yes, obtain acoustic features from the target audio and obtain contextual information that affects gender judgment, and use the contextual information and the acoustic features to determine whether the target audio contains a female voice; The determination module 1040 is used to determine whether to trigger a protection strategy for passengers based on the judgment result and the gender of the ride-hailing driver.

[0070] Optionally, the judgment module 1030 is specifically used for: The contextual information is mapped into the same dimensional space as the acoustic features through an embedding layer; By connecting the beginning and end of the signal, the mapped contextual information is concatenated with the acoustic features to obtain the fused features; The fused features are input into the recognition model, and the recognition model outputs whether the target audio contains a female voice.

[0071] Optional, the detection module 1020 is specifically used for: The target audio is input into the translation model, and the translation model outputs multiple words in the target audio and the sound start time of each word; Determine the order cancellation time; Words whose sound start time is less than the order placement time are selected as keywords in the target audio.

[0072] Optional, the detection module 1020 is specifically used for: The sound end time of the keyword is taken as the position of the keyword; wherein, the sound end time of the keyword is output by the translation model; A preset character following the sound end time of the keyword in the target audio is identified as a candidate character; Detect whether the candidate characters contain a preset number of consecutive digits.

[0073] Optionally, the judgment module 1030 is specifically used for: Extract the preset number of consecutive audio segments from the target audio; The extracted audio segments are copied multiple times to obtain processed audio; wherein the duration of the processed audio exceeds a preset duration. The processed audio is input into the extraction model, and the extraction model outputs acoustic features.

[0074] Optionally, module 1010 is used specifically for: Receive the initial audio from inside the ride-hailing vehicle; The initial audio from the ride-hailing vehicle is divided into frames of fixed frame length to obtain multiple segments; Identify segments from multiple clips where the probability of human voices exceeds a first threshold; Find out whether the distance between the voice of a segment exceeding the first threshold and the adjacent segment exceeds the second threshold; The two segments that exceed the second threshold are combined in chronological order to obtain the target audio.

[0075] This invention provides a passenger gender recognition device for ride-hailing services, applied to a terminal, comprising: The acquisition module is used to acquire the initial audio inside the ride-hailing vehicle; The judgment module is used to divide the initial audio in the ride-hailing vehicle into frames according to a fixed frame length to obtain multiple segments; for each segment, it is determined whether the probability of human voice in each segment exceeds a threshold. The combination module is used to combine the segments exceeding the threshold in chronological order to obtain at least one voice packet with a length less than a preset value; and to send the at least one voice packet to the server so that the server can use the voice packet to determine the gender of the passenger.

[0076] The present invention also provides a passenger gender recognition device for ride-hailing services, applied to a server, comprising: The receiving module is used to receive at least one voice packet sent by the driver's terminal of the ride-hailing vehicle, and to assemble the target audio according to the time sequence of the at least one voice packet; The detection module is used to determine the keywords in the target audio; and to detect whether the audio following the position of the keywords in the target audio contains a preset number of consecutive numbers. The judgment module is used to, if yes, obtain acoustic features from the target audio and obtain contextual information that affects gender judgment, and use the contextual information and the acoustic features to determine whether the target audio contains a female voice; The determination module is used to determine whether to trigger the protection strategy for passengers based on the judgment result and the gender of the ride-hailing driver.

[0077] Optional, a judgment module, specifically used for: The contextual information is mapped into the same dimensional space as the acoustic features through an embedding layer; By connecting the beginning and end of the signal, the mapped contextual information is concatenated with the acoustic features to obtain the fused features; The fused features are input into the recognition model, and the recognition model outputs whether the target audio contains a female voice.

[0078] Optional, detection module, specifically used for: The target audio is input into the translation model, and the translation model outputs multiple words in the target audio and the sound start time of each word; Determine the order cancellation time; Words whose sound start time is less than the order placement time are selected as keywords in the target audio.

[0079] Optional, detection module, specifically used for: The sound end time of the keyword is taken as the position of the keyword; wherein, the sound end time of the keyword is output by the translation model; A preset character following the sound end time of the keyword in the target audio is identified as a candidate character; Detect whether the candidate characters contain a preset number of consecutive digits.

[0080] Optional, a judgment module, specifically used for: Extract the preset number of consecutive audio segments from the target audio; The extracted audio segments are copied multiple times to obtain processed audio; wherein the duration of the processed audio exceeds a preset duration. The processed audio is input into the extraction model, and the extraction model outputs acoustic features.

[0081] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for passenger gender identification in ride-hailing services, characterized in that, include: Obtain the target audio from inside a ride-hailing vehicle; Identify the keywords in the target audio; And in the audio following the position of the keyword in the target audio, detect whether it contains a preset number of consecutive numbers; If so, acoustic features are obtained from the target audio, and contextual information that affects gender determination is obtained. Using the contextual information and the acoustic features, it is determined whether the target audio contains a female voice. Based on the judgment result and the gender of the ride-hailing driver, determine whether to trigger the protection strategy for passengers.

2. The method for passenger gender identification in ride-hailing services according to claim 1, characterized in that, Using the contextual information and the acoustic features, determining whether the target audio contains a female voice includes: The contextual information is mapped into the same dimensional space as the acoustic features through an embedding layer; By connecting the beginning and end of the signal, the mapped contextual information is concatenated with the acoustic features to obtain the fused features; The fused features are input into the recognition model, and the recognition model outputs whether the target audio contains a female voice.

3. The method for passenger gender identification in ride-hailing services according to claim 1, characterized in that, Determining keywords in the target audio includes: The target audio is input into the translation model, and the translation model outputs multiple words in the target audio and the sound start time of each word; Determine the order cancellation time; Words whose sound start time is less than the order placement time are selected as keywords in the target audio.

4. The method for passenger gender identification in ride-hailing services according to claim 3, characterized in that, In the audio following the location of the keyword within the target audio, detect whether it contains a preset number of consecutive digits, including: The sound end time of the keyword is taken as the position of the keyword; wherein, the sound end time of the keyword is output by the translation model; A preset character following the sound end time of the keyword in the target audio is identified as a candidate character; Detect whether the candidate characters contain a preset number of consecutive digits.

5. The method for passenger gender identification in ride-hailing services according to claim 1, characterized in that, Obtaining acoustic features from the target audio includes: Extract the preset number of consecutive audio segments from the target audio; The extracted audio segments are copied multiple times to obtain processed audio; wherein the duration of the processed audio exceeds a preset duration. The processed audio is input into the extraction model, and the extraction model outputs acoustic features.

6. The method for passenger gender identification in ride-hailing services according to claim 1, characterized in that, Obtain the target audio from within the ride-hailing vehicle, including: Receive initial audio from inside the ride-hailing vehicle; The initial audio from the ride-hailing vehicle is divided into frames of fixed frame length to obtain multiple segments; Identify segments from multiple clips where the probability of human voices exceeds a first threshold; Find out whether the distance between the voice of a segment exceeding the first threshold and the adjacent segment exceeds the second threshold; The two segments that exceed the second threshold are combined in chronological order to obtain the target audio.

7. A method for passenger gender identification in ride-hailing services, characterized in that, Applied to servers, including: Receive at least one voice packet sent by the driver's terminal of the ride-hailing vehicle, and assemble the target audio according to the time sequence of the at least one voice packet; Identify keywords in the target audio; and in the audio following the position of the keywords in the target audio, detect whether a preset number of consecutive numbers are contained; If so, acoustic features are obtained from the target audio, and contextual information that affects gender determination is obtained. Using the contextual information and the acoustic features, it is determined whether the target audio contains a female voice. Based on the judgment result and the gender of the ride-hailing driver, determine whether to trigger the protection strategy for passengers.

8. A method for passenger gender identification in ride-hailing services, characterized in that, Applied to terminals, including: Collect initial audio from inside the ride-hailing vehicle; The initial audio within the ride-hailing vehicle is divided into segments of fixed frame length; For each segment, determine whether the probability of human voice in each segment exceeds a threshold; The segments exceeding the threshold are combined in chronological order to obtain at least one voice packet shorter than a preset length; at least one voice packet is sent to the server so that the server can use the voice packet to determine the passenger's gender.

9. A server, characterized in that, include: Memory is used to store computer programs or instructions; A processor for executing a computer program or instructions in the memory, such that the passenger gender identification method for ride-hailing vehicles as described in any one of claims 1-6 is executed.

10. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor, the processor is able to perform the passenger gender identification method for ride-hailing vehicles as described in any one of claims 1-6.

Images