Method and device for determining voice acquisition truncation time and electronic equipment

Abstract

The embodiment of the invention relates to the technical field of voice recognition, in particular to a method and device for determining voice acquisition truncation time and electronic equipment, and the method comprises the steps: in the process of acquiring voice data of a target user, based on the analysis of the acquired voice data, obtaining adjustment time under at least one analysis scene; the type of the analysis scene comprises at least one scene of a user voice habit scene, an equipment performance scene and a mood assistant scene, and the adjustment time is used for adjusting the truncation time of voice data acquisition; and adjusting the basic truncation time by adopting the adjustment time under the at least one analysis scene to obtain the subsequent truncation time for collecting the voice data of the target user. According to the method for determining the voice acquisition truncation time disclosed by the invention, the truncation time of the voice data can be more accurately determined by integrating various analysis scenes, redundant data and wrong recognition in a subsequent voice processing process are reduced, and the voice processing efficiency and accuracy are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of speech recognition technology, and in particular to a method, device and electronic device for determining a speech collection cutoff time. Background Art

[0002] With the continuous development of speech recognition technology, voice interaction is increasingly used in smart devices. In voice interaction, the voice collection and truncation method usually involves directly cutting off the unnecessary parts at the beginning and end of the recording to obtain the required voice fragments. For example, at the beginning of the recording, the system will wait for a period of time (such as 1 second) to ensure that the recording device is ready and starts recording, and then wait for a period of time (also such as 1 second) after the user stops reading or speaking to ensure that the voice is complete, and then truncate to remove the silence or irrelevant parts before and after.

[0003] However, the above traditional truncation method is prone to truncation. Truncation means that a certain word is not recorded in full at the beginning or end of the recording, resulting in incomplete sentences and easy to misjudge the semantics. For example, if the front or back end of a word is truncated, the meaning of the word may change, which in turn affects the subsequent speech recognition or natural language processing results, and easily leads to a decrease in speech recognition rate. Summary of the invention

[0004] The purpose of the present invention is to at least provide a method for determining the voice collection cutoff time, which can at least solve the problem that the traditional cutoff method is prone to sound cutting and affects the recognition accuracy, and at least can achieve the effect of improving the recognition accuracy.

[0005] To solve the above technical problem, at least one embodiment of the present application provides a method for determining a voice collection cutoff time, comprising:

[0006] In the process of collecting voice data of a target user, based on analyzing the collected voice data, an adjustment time for the target user in at least one analysis scenario is obtained; wherein the type of the analysis scenario includes at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust a cutoff time for voice data collection;

[0007] The basic cutoff time is adjusted using the adjustment time in the at least one analysis scenario to obtain a cutoff time subsequently used for collecting the voice data of the target user.

[0008] At least one embodiment of the present application further provides a device for determining a voice collection cutoff time, comprising:

[0009] An adjustment time calculation module is used to obtain an adjustment time for the target user in at least one analysis scenario based on analyzing the collected voice data during the process of collecting the voice data of the target user; wherein the type of the analysis scenario includes at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust the cutoff time of voice data collection;

[0010] The truncation time calculation module is used to adjust the basic truncation time by using the adjustment time under the at least one analysis scenario to obtain the truncation time used for collecting the voice data of the target user.

[0011] At least one embodiment of the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for determining the voice collection cutoff time.

[0012] At least one embodiment of the present application further provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned method for determining the voice collection cutoff time.

[0013] The method for determining the voice collection truncation time provided by the embodiment of the present application can more accurately identify the natural end point of the user's voice by considering the user's voice habit scenarios, such as speech speed, pause mode, etc., and avoid incomplete or redundant voice information caused by too early or too late truncation time; for frequently used modal particles or specific expressions, the method can identify and adjust the truncation time to reduce voice comprehension errors or user confusion caused by truncation of modal particles; by considering the device performance scenario, such as device processing capacity, storage space, etc., the truncation time can be dynamically adjusted to ensure that while maintaining the quality of voice data, the device resources are maximized, and unnecessary energy consumption and storage occupancy are reduced. In the case of limited device performance, the truncation time can also be adjusted in time to prevent problems such as overheating and freezing of the device due to continuous collection, thereby improving the stability and reliability of the device. By integrating multiple analysis scenarios, the method can more accurately determine the truncation time of voice data, thereby reducing redundant data and erroneous recognition in the subsequent voice processing process, and improving the efficiency and accuracy of voice processing.

[0014] In one embodiment, when the analysis scenario is the user voice habit scenario, obtaining the adjustment time for the target user in the user voice habit scenario based on analyzing the collected voice data includes:

[0015] Using a pre-trained user voice habit model to detect the collected voice data, determine a first group category to which the target user belongs among a plurality of group categories divided according to voice habits, each of the group categories being correspondingly provided with a voice habit parameter;

[0016] Based on the voice habit parameters corresponding to the first group category, the adjustment time of the target user in the user voice habit scenario is determined; the voice habit parameters include at least one of an average speaking speed and a pause time between words.

[0017] In this embodiment, since each user's speaking speed and pause time between words are unique, by identifying and adapting to these characteristics of the target user, it is possible to reduce the situation of false interruptions caused by mismatched voice characteristics, and more accurately determine when to interrupt voice collection, so that the collected voice data is more natural and smooth, in line with the user's voice expression habits, thereby improving the user experience and increasing the reliability and stability of voice collection.

[0018] In one embodiment, when the analysis scenario is the device performance scenario, obtaining the adjustment time for the target user in the device performance scenario based on analyzing the collected voice data includes:

[0019] Acquire performance parameters of a collection device currently collecting voice data, wherein the performance parameters include at least one of a central processing unit performance parameter, a memory parameter, and a microphone parameter;

[0020] Evaluate the performance parameters to obtain a device performance score;

[0021] Based on the device performance score, the adjustment time of the target user in the device performance scenario is determined.

[0022] In this embodiment, the performance parameters of the device, such as the performance of the central processing unit, the size of the memory, and the sensitivity of the microphone, directly determine the ability and efficiency of the device to process voice data. By obtaining these parameters, the actual performance of the device in collecting and processing voice data can be more accurately evaluated, so as to adjust the voice cutoff time according to the actual capabilities of the device. In this way, the problem of voice cutoff being too early or too late due to insufficient device performance can be avoided, and the collected voice data can be ensured to be complete and accurate.

[0023] In one embodiment, when the analysis scenario is the modal particle scenario, obtaining the adjustment time for the target user in the modal particle scenario based on analyzing the collected voice data includes:

[0024] Extracting modal particles appearing in the collected speech data, as well as the average duration and standard deviation of the modal particles appearing;

[0025] Based on the average duration and the standard deviation, the adjustment time of the target user in the context of the modal particle is determined.

[0026] In this embodiment, modal particles play the role of connecting sentences, modifying components and emphasizing the tone in speech. Their existence makes the speech more natural and fluent. If the speech truncation time is inappropriate, it may destroy the naturalness and fluency of the speech, making the recognition result appear stiff or abrupt. By correcting the speech truncation time, it can ensure that the modal particles are properly processed, thereby enhancing the naturalness and fluency of the speech and optimizing the user experience.

[0027] In one embodiment, determining the adjustment time of the target user in the context of the modal particle based on the average duration and the standard deviation includes:

[0028] The adjustment time is calculated using the following formula:

[0029] T_{filler}=T ave +τ*T dev

[0030] Wherein, T_{filler} is the adjustment time of the target user in the context of the modal particle, T ave is the average duration, T dev is the standard deviation, and τ is the adjustment coefficient.

[0031] In this embodiment, different users may have different pronunciation durations when using modal particles. By calculating the average duration and standard deviation, this individual difference is fully taken into account, so that the adjustment time is more in line with the actual usage of the target user. As a variable parameter, the adjustment coefficient can be flexibly adjusted according to the actual situation, which means that when the target user's modal particle usage habits change, the adjustment coefficient can be adjusted to adapt to this change, thereby maintaining the accuracy of the adjustment time.

[0032] In one embodiment, the adjusting the basic truncation time by using the adjustment time in the at least one analysis scenario to obtain the truncation time for subsequently collecting the voice data of the target user includes:

[0033] Performing weighted summation on the adjustment time under the at least one analysis scenario to obtain a comprehensive adjustment time;

[0034] The basic cutoff time and the comprehensive adjustment time are superimposed to obtain the subsequent cutoff time for collecting the target user's voice data.

[0035] In this embodiment, by performing weighted summation on the adjustment time in at least one analysis scenario to obtain the comprehensive adjustment time, the influence of various factors on the truncation time can be more comprehensively considered, thereby further improving the accuracy of the truncation time.

[0036] In one embodiment, it further includes:

[0037] The speech recognition effect evaluation result corresponding to the speech data collected based on the adjusted cutoff time by the target user is obtained, and the parameters in the process of obtaining the cutoff time are optimized.

[0038] In this embodiment, by obtaining the speech recognition effect evaluation results of the target user, we can intuitively understand the performance of the adjusted truncation time in actual applications, so as to carry out targeted optimization of the parameters in the process of obtaining the truncation time (such as average duration, standard deviation, adjustment coefficient, etc.), thereby improving the accuracy of speech recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] One or more embodiments are exemplarily described by the pictures in the corresponding drawings, and these exemplary descriptions do not constitute limitations on the embodiments.

[0040] Figure 1 This is a flow chart of a method for determining a voice collection cutoff time provided by an embodiment of the present application. Figure 1 ;

[0041] Figure 2 This is a flow chart of a method for determining a voice collection cutoff time provided by an embodiment of the present application. Figure 2 ;

[0042] Figure 3 It is a schematic diagram of a device for determining a voice collection cutoff time provided by an embodiment of the present application. DETAILED DESCRIPTION

[0043] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be construed as any limitation, and the various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0044] Embodiment 1:

[0045] The method for determining the voice collection cutoff time of this embodiment can be applied to electronic devices with communication, computing and data storage capabilities. The specific process can be as follows: Figure 1 As shown, including:

[0046] Step 110, in the process of collecting voice data of the target user, based on analyzing the collected voice data, an adjustment time for the target user in at least one analysis scenario is obtained; wherein the types of analysis scenarios include at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust the cutoff time of voice data collection.

[0047] It is understandable that the existing voice truncation method is usually a fixed time length, and the voice habits of different people are quite different. Therefore, the existing voice truncation time setting method is difficult to match the voice habits of different users, and it is easy to truncate the voice too early, resulting in incomplete collected voice information. Therefore, when collecting the voice data of the target user, the system of this embodiment is not just simply receiving the sound signal, but is constantly learning and adapting to the user's voice characteristics. By conducting in-depth analysis of the collected voice data, the system can identify the adjustments that need to be made in different analysis scenarios (including but not limited to user voice habit scenarios, device performance scenarios, and modal particle scenarios, etc.), thereby calculating the corresponding adjustment time to optimize the truncation time of voice data collection.

[0048] In the analysis of user speech habits, by analyzing the pronunciation habits of the target user, such as speaking speed, intonation, pauses, etc., the system can calculate the extra time that the user may need to complete a complete pronunciation or expression in different contexts. For example, if the target user is accustomed to a slower speaking speed, the system can increase the truncation time accordingly to ensure that every syllable of the user can be fully captured.

[0049] The voice collection device is also a key factor affecting the accuracy of voice collection. Different voice collection devices have differences in response speed and collection clarity, which affects the accuracy of voice truncation. In the analysis of device performance scenarios, different voice collection devices have significant differences in sampling rate, sound quality, noise suppression ability, etc. These differences directly affect the collection quality of voice data and the accuracy of subsequent processing. Therefore, adjusting the truncation time according to device performance can ensure relatively consistent voice recognition effects on different devices. For example, high-performance devices usually have stronger processing power and higher sampling rates, and can capture more voice details. In this case, appropriately increasing the truncation time can capture more useful voice information and improve recognition accuracy. Low-performance devices may require shorter truncation times to reduce processing burden and avoid resource overload.

[0050] Specifically, while acquiring the user's voice, relevant tools are called in parallel to monitor the device's operating status (such as battery power, network latency, memory operation, etc.) in real time, and the cutoff time is dynamically adjusted according to these statuses. For example, when the device is low on power, the system may reduce the sampling rate to extend the usage time. At this time, the cutoff time may need to be increased accordingly to ensure the accuracy of voice recognition.

[0051] In the analysis of modal particles, by analyzing the pronunciation characteristics of the target user when using modal particles (such as "hmm", "ah", "hey", etc.), the system can calculate the additional time required to identify these modal particles. For example, if the target user's pronunciation is long and dragging when using modal particles, the system can increase the truncation time accordingly to fully capture these modal particles. For example, natural language processing technology is used to convert voice data into text, identify modal particles in the text, and dynamically adjust the truncation time according to the frequency and location of these modal particles.

[0052] In some embodiments, in addition to the three analysis scenarios of user speech habits, device performance, and modal particles, the following objective factors may also be considered to improve the accuracy of determining the speech truncation time:

[0053] 1) Ambient noise level:

[0054] The user's ambient noise level is detected in real time through a built-in microphone or an external noise detection sensor. When the ambient noise is high, the system can automatically extend the cutoff time to ensure that the key information in the user's voice can be fully captured and reduce the impact of noise on speech recognition. This can be achieved by analyzing the signal-to-noise ratio (SNR) of the voice signal. When the SNR is below a certain threshold, the cutoff time is increased.

[0055] 2) Voice clarity:

[0056] By analyzing the spectral characteristics and clarity indicators of the speech signal (such as the speech clarity index, AI), the system can evaluate the clarity of the user's speech. For situations with low clarity, the system can extend the truncation time to capture more contextual information, which is helpful for subsequent speech recognition and understanding.

[0057] 3) User emotional state:

[0058] While the context of modal particles partially reflects the user's emotional state, a more comprehensive sentiment analysis can be achieved through features such as the pitch, volume, and rhythm of the speech. For example, when the target user is excited or nervous, their speech rate may speed up, their volume may increase, and their pitch may change. The system can dynamically adjust the truncation time by analyzing these features to adapt to changes in the user's emotional state. This can be achieved by applying machine learning algorithms such as emotion recognition models.

[0059] 4) Application scenario requirements:

[0060] Different application scenarios may have different requirements for voice truncation time. For example, in voice search scenarios, users usually expect to get results quickly, so the truncation time can be relatively short; while in voice recording or voice transcription scenarios, users may prefer to fully record their voice input, so the truncation time can be relatively long. The system can dynamically adjust the truncation time according to the current application scenario requirements. This can be achieved by configuring different parameter settings or applying different processing strategies.

[0061] By comprehensively considering the above objective factors and combining analysis scenarios such as user voice habits, device performance and modal particles, the system can more accurately determine the voice truncation time, thereby improving the accuracy of voice recognition and user experience.

[0062] Step 120: The basic cutoff time is adjusted by using the adjustment time in the at least one analysis scenario to obtain a cutoff time subsequently used for collecting the voice data of the target user.

[0063] In this embodiment, after obtaining the adjustment times under different analysis scenarios, the system applies these adjustment times to the basic cutoff time to calculate a cutoff time that is more suitable for the target user. By reasonably integrating the adjustment times under different analysis scenarios, it is ensured that the final cutoff time can not only meet the actual needs of the user, but also maintain the stability and efficiency of the system.

[0064] When adjusting, for example, a weighted average method can be applied: for the adjustment time in different analysis scenarios, a weighted average can be performed according to its importance and credibility to obtain the final adjustment time. For example, the adjustment time obtained in the user's voice habit scenario may be more important, so a higher weight can be given.

[0065] In some embodiments, the dynamic adjustment method is used to dynamically adjust the weight of the adjustment time in different analysis scenarios according to the current usage scenario and user needs. For example, in a scenario that requires a quick response, the weight of the adjustment time in the device performance scenario may be reduced to speed up the speech recognition.

[0066] In some embodiments, a machine learning algorithm is used to model and optimize the adjustment time and the basic cutoff time to find the optimal cutoff time setting, which can adapt to the changes and needs of target users through continuous learning and iteration.

[0067] By implementing the above steps and implementation methods, the system can capture the target user's voice data more accurately, so that the final calculated voice truncation time can meet the user's actual needs, while maintaining the stability and efficiency of the system, improving the accuracy and efficiency of voice recognition, thereby enhancing the user's voice interaction experience.

[0068] In summary, the method for determining the voice collection truncation time provided in this embodiment can more accurately identify the natural end point of the user's voice by considering the user's voice habit scenarios, such as speech speed, pause mode, etc., and avoid incomplete or redundant voice information caused by too early or too late truncation time; for frequently used modal particles or specific expressions, the method can identify and adjust the truncation time to reduce voice misunderstanding errors or user confusion caused by truncation of modal particles; by considering device performance scenarios, such as device processing capabilities, storage space, etc., the truncation time can be dynamically adjusted to ensure that while maintaining the quality of voice data, the device resources are maximized, and unnecessary energy consumption and storage occupancy are reduced. In the case of limited device performance, the truncation time can also be adjusted in time to prevent problems such as overheating and freezing of the device due to continuous collection, thereby improving the stability and reliability of the device. By integrating multiple analysis scenarios, the method can more accurately determine the truncation time of voice data, thereby reducing redundant data and erroneous recognition in the subsequent voice processing process, and improving the efficiency and accuracy of voice processing.

[0069] In one embodiment, when the analysis scenario is the user voice habit scenario, the adjustment time for the target user in the user voice habit scenario is obtained based on the analysis of the collected voice data, including: using a pre-trained user voice habit model to detect the collected voice data, and determining a first group category to which the target user belongs among multiple group categories divided by voice habits, each group category being provided with a corresponding voice habit parameter; based on the voice habit parameters corresponding to the first group category, determining the adjustment time of the target user in the user voice habit scenario; the voice habit parameters include at least one of an average speaking speed and a pause time between words.

[0070] In this embodiment, since each user's speaking speed and pause time between words are unique, by identifying and adapting to these characteristics of the target user, it is possible to reduce the situation of false interruptions caused by mismatched voice characteristics, and more accurately determine when to interrupt voice collection, so that the collected voice data is more natural and smooth, in line with the user's voice expression habits, thereby improving the user experience and increasing the reliability and stability of voice collection.

[0071] Specifically, the pre-trained user voice habit model can be trained by collecting a large amount of voice data from the user's daily voice interaction. For example, when using a voice assistant, user A will issue instructions such as "play music" and "set an alarm." In order to ensure the diversity and representativeness of the data, voice data of multiple users in various environments are collected. Then, the collected voice data are feature extracted and annotated, including the start time and end time of each word, as well as the pause time. Taking user A as an example, the average number of words spoken per second (average speaking speed) and the pause time between each word (inter-word pause time) are analyzed. Assume that the analysis results show that user A speaks 4 words per second on average, and the pause time between each word is 0.4 seconds. Then, the K-means clustering algorithm is applied to divide users into different groups. For example, users are divided into "fast speaking group" and "slow speaking group." User A is classified as the "fast speaking group", and the feature model of the "fast speaking group" records typical speaking habit parameters, namely the average speaking speed and inter-word pause time.

[0072] The average speaking speed can be calculated using the following formula:

[0073] Average speaking speed = total number of words / total speaking time. The total number of words refers to the total number of words spoken by the user in a voice interaction, and the total speaking time refers to the actual speaking time of the user (excluding pause time);

[0074] The pause time between words can be calculated using the following formula:

[0075] Average pause time = total pause time / total pause times. The total pause time refers to the total time of all pauses of the user in a voice interaction, and the total pause times refers to the number of times the user pauses in a voice interaction.

[0076] According to the user's average speaking speed and pause time between words, determine the adjustment time in the voice habit scenario (or voice habit adjustment time):

[0077] T_{user} = α × average speaking speed + β × average pause time. α and β are weight coefficients, which can be adjusted according to the actual application effect. The initial setting of α and β is 0.5, and the values ​​of α and β can be gradually optimized according to user feedback and actual application effect.

[0078] In one embodiment, when the analysis scenario is the device performance scenario, the adjustment time for the target user in the device performance scenario is obtained based on the analysis of the collected voice data, including: obtaining the performance parameters of the collection device currently collecting the voice data, wherein the performance parameters include at least one of the performance parameters of the central processing unit, memory parameters and microphone parameters; evaluating the performance parameters to obtain a device performance score; and determining the adjustment time of the target user in the device performance scenario based on the device performance score.

[0079] In this embodiment, the performance parameters of the voice collection device, such as the performance of the central processing unit, the size of the memory, and the sensitivity of the microphone, directly determine the ability and efficiency of the device to process voice data. By obtaining these parameters, the actual performance of the device in collecting and processing voice data can be more accurately evaluated, so as to adjust the voice cutoff time according to the actual capabilities of the device. In this way, the problem of voice cutoff being too early or too late due to insufficient device performance can be avoided, and the collected voice data can be ensured to be complete and accurate.

[0080] It is understandable that the processing power and microphone quality of voice collection devices may decline after years of use. In order to more accurately evaluate the current performance of the voice collection device, a dynamic performance evaluation is performed in real time. Specifically, the type of voice collection device being used by the target user is identified through the device's hardware identifier (such as device model, device ID, etc.). Use device APIs or system calls to obtain detailed information about the voice collection device, including model, operating system version, hardware configuration, etc.

[0081] After that, based on the detailed information of the voice collection device obtained, evaluate the device's processing capabilities and the quality of the device's microphone:

[0082] Processing power score = CPU performance score + memory score. The CPU performance score and memory score can be quantified based on actual performance test results.

[0083] Microphone quality score = sensitivity score + noise suppression score. The sensitivity is evaluated by calling the microphone test tool, and the noise suppression is evaluated by calling the noise test tool.

[0084] Finally, the comprehensive device performance score (i.e. the above device performance score) is calculated based on the processing power score and the microphone quality score:

[0085] Comprehensive device performance score = processing power score + microphone quality score. Based on the comprehensive device performance score, the adjustment time (or device performance adjustment time) T_{device} in the device performance scenario is calculated according to the following formula:

[0086] T_{device}=T_{base}+γ×comprehensive device performance score;

[0087] Among them, T_{base} is the basic cutoff time, and γ is the adjustment coefficient, which can be adjusted according to the actual application effect.

[0088] In some embodiments, the basic cutoff time parameter is set based on the acquired device performance. For example, the device that the user is using is identified as a model A smartphone through the device hardware identifier. Then the processing power and microphone quality of the device are evaluated. Assuming that model A smartphone has a high-performance processor and a high-quality microphone, the basic cutoff time can be set to 0.5 seconds. Assuming that model A smartphone has average processor and microphone performance, the corresponding basic cutoff time can be set to 0.8 seconds.

[0089] In one embodiment, when the analysis scenario is the modal particle scenario, the adjustment time for the target user in the modal particle scenario is obtained based on the analysis of the collected voice data, including: extracting the modal particles appearing in the collected voice data, as well as the average duration and standard deviation of the occurrence of the modal particles; based on the average duration and standard deviation, determining the adjustment time of the target user in the modal particle scenario.

[0090] In this embodiment, modal particles play the role of connecting sentences, modifying components and emphasizing the tone in speech. Their existence makes the speech more natural and fluent. If the speech truncation time is inappropriate, it may destroy the naturalness and fluency of the speech, making the recognition result appear stiff or abrupt. By correcting the speech truncation time, it can ensure that the modal particles are properly processed, thereby enhancing the naturalness and fluency of the speech and optimizing the user experience.

[0091] The average duration refers to the average duration of each modal particle said by the target user, and the standard deviation refers to the standard deviation of the duration of each modal particle said by the target user. In one embodiment, the adjustment time of the target user in the modal particle scenario is determined based on the average duration and the standard deviation, including: using the following calculation formula to calculate the adjustment time (or modal particle adjustment time) T_{filler} in the modal particle scenario:

[0092] T_{filler}=T ave +τ*T dev

[0093] Wherein, T_{filler} is the adjustment time of the target user in the context of the modal particle, T ave is the average duration, T devis the standard deviation, and τ is the adjustment coefficient.

[0094] In this embodiment, different users may have different pronunciation durations when using modal particles. By calculating the average duration and standard deviation, this individual difference is fully taken into account, so that the adjustment time is more in line with the actual usage of the target user. As a variable parameter, the adjustment coefficient can be flexibly adjusted according to the actual situation, which means that when the target user's modal particle usage habits change, the adjustment coefficient can be adjusted to adapt to this change, thereby maintaining the accuracy of the adjustment time.

[0095] In one embodiment, Figure 2 As shown, the basic truncation time is adjusted using the adjustment time under the at least one analysis scenario to obtain the subsequent truncation time for collecting the target user voice data, including: weighted summing the adjustment time under the at least one analysis scenario to obtain a comprehensive adjustment time; superimposing the basic truncation time and the comprehensive adjustment time to obtain the subsequent truncation time for collecting the target user voice data.

[0096] In this embodiment, by performing weighted summation on the adjustment time in at least one analysis scenario to obtain the comprehensive adjustment time, the influence of various factors on the truncation time can be more comprehensively considered, thereby further improving the accuracy of the truncation time.

[0097] For example, we calculate T_{user}, T_{device}, and T_{filler}, perform a weighted sum on these three items, and then add them to the basic cutoff time to get the final cutoff time T_{final}:

[0098] T_{final}=T_{base}+W_{user}*T_{user}+W_{device}*T_{device}+W_{filler}*T_{filler}.

[0100] For example, assuming that the base cutoff time T_{base} is 0.5 seconds, the user habit adjustment time T_{user} is 0.2 seconds, the device performance adjustment time T_{device} is 0.1 seconds, and the modal particle adjustment time T_{filler} is 0.3 seconds, the weight coefficients W_{user}, W_{device}, and W_{filler} are 0.5, 0.3, and 0.2 respectively. The final cutoff time is calculated as:

[0101] T_{final}=0.5+0.5*0.2+0.3*0.1+0.2*0.3=0.5+0.1+0.03+0.06=0.69s

[0102] According to the above calculation results, the cutoff time of voice collection is adjusted in real time. For example, the final cutoff time is calculated to be 0.69 seconds, and the system will wait for the user input to complete within this time.

[0103] In some embodiments, the analysis scenario required by the application scenario is further increased. For example, if it is detected that the user usually expects to get the result quickly in the current voice search scenario, the truncation time can be relatively short. Therefore, when calculating T_{final}, the adjustment time T_{scene} in the analysis scenario required by the application scenario is increased. Correspondingly, T_{final} is calculated according to the following formula:

[0104] T_{final}=T_{base}+W_{user}*T_{user}+W_{device}*T_{device}+W_{filler}*T_{filler}-W_{scene}*T_{scene}

[0105] Among them, W_{scene} is the weight coefficient of T_{scene}, and the weight coefficients W_{user}, W_{device}, W_{filler} and W_{scene} are 0.5, 0.2, 0.2 and 0.1 respectively.

[0106] Similarly, when it is detected that the user may prefer to fully record their voice input in the current voice recording or voice transcription scenario, the truncation time can be relatively long. In this case, when calculating T_{final}, T_{scene} is added to extend the truncation time so that the voice recording is more complete.

[0107] Similarly, in some embodiments, the adjustment of the voice cutoff time by the environmental noise level may also be considered, and the final cutoff time may be made more in line with actual usage requirements by integrating the adjustment time under multiple analysis scenarios.

[0108] In one embodiment, the method further includes: obtaining a speech recognition effect evaluation result of the target user corresponding to the speech data collected based on the adjusted cutoff time, and optimizing parameters in the process of obtaining the cutoff time.

[0109] In this embodiment, by obtaining the speech recognition effect evaluation results of the target user, we can intuitively understand the performance of the adjusted truncation time in actual applications, so as to carry out targeted optimization of the parameters in the process of obtaining the truncation time (such as average duration, standard deviation, adjustment coefficient, etc.), thereby improving the accuracy of speech recognition.

[0110] Collect feedback data from actual user use, including speech recognition accuracy and user satisfaction ratings. For example, after a voice interaction, the user rates the system, and the system records the recognition accuracy. Use feedback data to optimize the user habit model, device performance model, and modal particle detection model. For example, if it is found that the recognition accuracy of users in the "slow speech group" is low, the system automatically adjusts the model parameters. The system has the ability to continuously learn and iterate optimization, and improves the accuracy of cutoff time adjustment by continuously introducing new data and user feedback. For example, the system updates the model parameters once a month to ensure that it adapts to the needs of different users and devices.

[0111] Embodiment 2:

[0112] Another embodiment of the present application relates to a device for determining a voice collection cutoff time. The implementation details of the device for determining a voice collection cutoff time of the present embodiment are described in detail below. The following content is only provided for the convenience of understanding the implementation details and is not necessary for implementing the present solution. The schematic diagram of the device for determining a voice collection cutoff time of the present embodiment can be as follows: Figure 3 As shown, it includes an adjustment time calculation module 310 and a truncation time calculation module 320.

[0113] The adjustment time calculation module 310 is used to obtain the adjustment time for the target user in at least one analysis scenario based on analyzing the collected voice data during the process of collecting the voice data of the target user; wherein the type of the analysis scenario includes at least one of the user voice habit scenario, the device performance scenario and the modal particle scenario, and the adjustment time is used to adjust the cutoff time of voice data collection;

[0114] The truncation time calculation module 320 is used to adjust the basic truncation time by using the adjustment time under the at least one analysis scenario to obtain the truncation time used for collecting the target user's voice data subsequently.

[0115] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0116] In one embodiment, the adjustment time calculation module includes:

[0117] A group category detection unit, configured to detect the collected voice data using a pre-trained user voice habit model, and determine a first group category to which the target user belongs among a plurality of group categories divided according to voice habits, wherein each group category is correspondingly provided with a voice habit parameter;

[0118] The first adjustment time calculation unit is used to determine the adjustment time of the target user in the user voice habit scenario based on the voice habit parameter corresponding to the first group category; the voice habit parameter includes at least one of an average speaking speed and a pause time between words.

[0119] In one embodiment, the adjustment time calculation module includes:

[0120] A performance parameter acquisition unit, used to acquire performance parameters of a collection device currently collecting voice data, wherein the performance parameters include at least one of a performance parameter of a central processing unit, a memory parameter, and a microphone parameter;

[0121] A device performance score calculation unit, used to evaluate the performance parameters to obtain a device performance score;

[0122] A second adjustment time calculation unit is configured to determine the adjustment time of the target user in the device performance scenario based on the device performance score.

[0123] In one embodiment, the adjustment time calculation module includes:

[0124] A parameter extraction unit, used to extract the modal particles appearing in the collected speech data, as well as the average duration and standard deviation of the modal particles appearing;

[0125] The third adjustment time calculation unit is used to determine the adjustment time of the target user in the context of the modal particle based on the average duration and the standard deviation.

[0126] In one embodiment, the truncation time calculation module includes:

[0127] A comprehensive time calculation unit, configured to perform weighted summation of the adjustment time under the at least one analysis scenario to obtain a comprehensive adjustment time;

[0128] The cutoff time calculation unit is used to add the basic cutoff time and the comprehensive adjustment time to obtain the cutoff time subsequently used to collect the voice data of the target user.

[0129] In one embodiment, it further includes:

[0130] The optimization module is used to obtain the speech recognition effect evaluation result corresponding to the speech data collected based on the adjusted cutoff time by the target user, and optimize the parameters in the process of obtaining the cutoff time.

[0131] Embodiment three:

[0132] Another embodiment of the present application relates to an electronic device, comprising: at least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the following method for determining a voice collection cutoff time:

[0133] In the process of collecting voice data of a target user, based on analyzing the collected voice data, an adjustment time for the target user in at least one analysis scenario is obtained; wherein the type of the analysis scenario includes at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust a cutoff time for voice data collection;

[0134] The basic cutoff time is adjusted using the adjustment time in the at least one analysis scenario to obtain a cutoff time subsequently used for collecting the voice data of the target user.

[0135] In one embodiment, when the instructions are executed by at least one processor, the processor is capable of performing the following method:

[0136] When the analysis scenario is the user voice habit scenario, the collected voice data is detected using a pre-trained user voice habit model to determine a first group category to which the target user belongs among a plurality of group categories divided according to voice habits, each of the group categories being correspondingly provided with a voice habit parameter;

[0137] Based on the voice habit parameters corresponding to the first group category, the adjustment time of the target user in the user voice habit scenario is determined; the voice habit parameters include at least one of an average speaking speed and a pause time between words.

[0138] In one embodiment, when the instructions are executed by at least one processor, the processor is capable of performing the following method:

[0139] When the analysis scenario is the device performance scenario, obtaining performance parameters of the acquisition device currently acquiring the voice data, wherein the performance parameters include at least one of a performance parameter of a central processing unit, a memory parameter, and a microphone parameter;

[0140] Evaluate the performance parameters to obtain a device performance score;

[0141] Based on the device performance score, the adjustment time of the target user in the device performance scenario is determined.

[0142] In one embodiment, when the instructions are executed by at least one processor, the processor is capable of performing the following method:

[0143] When the analysis scenario is the modal particle scenario, extracting the modal particles appearing in the collected voice data, and the average duration and standard deviation of the modal particles appearing;

[0144] Based on the average duration and the standard deviation, the adjustment time of the target user in the context of the modal particle is determined.

[0145] In one embodiment, when the instructions are executed by at least one processor, the processor is capable of performing the following method:

[0146] Performing weighted summation on the adjustment time under the at least one analysis scenario to obtain a comprehensive adjustment time;

[0147] The basic cutoff time and the comprehensive adjustment time are superimposed to obtain the subsequent cutoff time for collecting the target user's voice data.

[0148] In one embodiment, when the instructions are executed by at least one processor, the processor is capable of performing the following method:

[0149] The speech recognition effect evaluation result corresponding to the speech data collected based on the adjusted cutoff time by the target user is obtained, and the parameters in the process of obtaining the cutoff time are optimized.

[0150] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0151] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0152] Embodiment 4:

[0153] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the following method is implemented:

[0154] In the process of collecting voice data of a target user, based on analyzing the collected voice data, an adjustment time for the target user in at least one analysis scenario is obtained; wherein the type of the analysis scenario includes at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust a cutoff time for voice data collection;

[0155] The basic cutoff time is adjusted using the adjustment time in the at least one analysis scenario to obtain a cutoff time subsequently used for collecting the voice data of the target user.

[0156] In one embodiment, when the computer program is executed by a processor, the following method is also implemented:

[0157] When the analysis scenario is the user voice habit scenario, the collected voice data is detected using a pre-trained user voice habit model to determine a first group category to which the target user belongs among a plurality of group categories divided according to voice habits, each of the group categories being correspondingly provided with a voice habit parameter;

[0158] Based on the voice habit parameters corresponding to the first group category, the adjustment time of the target user in the user voice habit scenario is determined; the voice habit parameters include at least one of an average speaking speed and a pause time between words.

[0159] In one embodiment, when the computer program is executed by a processor, the following method is also implemented:

[0160] When the analysis scenario is the device performance scenario, obtaining performance parameters of the acquisition device currently acquiring the voice data, wherein the performance parameters include at least one of a performance parameter of a central processing unit, a memory parameter, and a microphone parameter;

[0161] Evaluate the performance parameters to obtain a device performance score;

[0162] Based on the device performance score, the adjustment time of the target user in the device performance scenario is determined.

[0163] In one embodiment, when the computer program is executed by a processor, the following method is also implemented:

[0164] When the analysis scenario is the modal particle scenario, extracting the modal particles appearing in the collected voice data, and the average duration and standard deviation of the modal particles appearing;

[0165] Based on the average duration and the standard deviation, the adjustment time of the target user in the context of the modal particle is determined.

[0166] In one embodiment, when the computer program is executed by a processor, the following method is also implemented:

[0167] Performing weighted summation on the adjustment time under the at least one analysis scenario to obtain a comprehensive adjustment time;

[0168] The basic cutoff time and the comprehensive adjustment time are superimposed to obtain the subsequent cutoff time for collecting the target user's voice data.

[0169] In one embodiment, when the computer program is executed by a processor, the following method is also implemented:

[0170] The speech recognition effect evaluation result corresponding to the speech data collected based on the adjusted cutoff time by the target user is obtained, and the parameters in the process of obtaining the cutoff time are optimized.

[0171] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as: ROM), random access memory (Random Access Memory, referred to as: RAM), disk or optical disk and other media that can store program codes.

[0172] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A method for determining a speech collection cutoff time, characterized in that: include: In the process of collecting voice data of a target user, based on analyzing the collected voice data, an adjustment time for the target user in at least one analysis scenario is obtained; wherein the type of the analysis scenario includes at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust a cutoff time for voice data collection; The basic cutoff time is adjusted using the adjustment time in the at least one analysis scenario to obtain a cutoff time subsequently used for collecting the voice data of the target user.

2. The method for determining the voice collection cutoff time according to claim 1, characterized in that: When the analysis scenario is the user voice habit scenario, obtaining the adjustment time for the target user in the user voice habit scenario based on analyzing the collected voice data includes: Using a pre-trained user voice habit model to detect the collected voice data, determine a first group category to which the target user belongs among a plurality of group categories divided according to voice habits, each of the group categories being correspondingly provided with a voice habit parameter; Based on the voice habit parameters corresponding to the first group category, the adjustment time of the target user in the user voice habit scenario is determined; the voice habit parameters include at least one of an average speaking speed and a pause time between words.

3. The method for determining the voice collection cutoff time according to claim 1, characterized in that: When the analysis scenario is the device performance scenario, obtaining the adjustment time for the target user in the device performance scenario based on analyzing the collected voice data includes: Acquire performance parameters of a collection device currently collecting voice data, wherein the performance parameters include at least one of a central processing unit performance parameter, a memory parameter, and a microphone parameter; Evaluate the performance parameters to obtain a device performance score; Based on the device performance score, the adjustment time of the target user in the device performance scenario is determined.

4. The method for determining the voice collection cutoff time according to claim 1, characterized in that: When the analysis scenario is the modal particle scenario, obtaining the adjustment time for the target user in the modal particle scenario based on analyzing the collected voice data includes: Extracting modal particles appearing in the collected speech data, as well as the average duration and standard deviation of the modal particles appearing; Based on the average duration and the standard deviation, the adjustment time of the target user in the context of the modal particle is determined.

5. The method for determining the voice collection cutoff time according to claim 4, characterized in that: The determining, based on the average duration and the standard deviation, the adjustment time of the target user in the context of the modal particle comprises: The adjustment time is calculated using the following formula: T_{filler}=T ave +τ*T dev Wherein, T_{filler} is the adjustment time of the target user in the context of the modal particle, T ave is the average duration, T dev is the standard deviation, and τ is the adjustment coefficient.

6. The method for determining the voice collection cutoff time according to claim 1, characterized in that: The step of adjusting the basic cutoff time by using the adjustment time in the at least one analysis scenario to obtain a cutoff time subsequently used for collecting the voice data of the target user includes: Performing weighted summation on the adjustment time under the at least one analysis scenario to obtain a comprehensive adjustment time; The basic cutoff time and the comprehensive adjustment time are superimposed to obtain the subsequent cutoff time for collecting the target user's voice data.

7. The method for determining the speech collection cutoff time according to any one of claims 1 to 6, characterized in that: Also includes: The speech recognition effect evaluation result corresponding to the speech data collected based on the adjusted cutoff time by the target user is obtained, and the parameters in the process of obtaining the cutoff time are optimized.

8. A device for determining a speech collection cutoff time, characterized in that: include: An adjustment time calculation module is used to obtain an adjustment time for the target user in at least one analysis scenario based on analyzing the collected voice data during the process of collecting the voice data of the target user; wherein the type of the analysis scenario includes at least one of a user voice habit scenario, a device performance scenario, and an interjection scenario, and the adjustment time is used to adjust the cutoff time of voice data collection; The truncation time calculation module is used to adjust the basic truncation time by using the adjustment time under the at least one analysis scenario to obtain the truncation time used for collecting the voice data of the target user.

9. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for determining the voice collection cutoff time as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for determining the speech collection cutoff time according to any one of claims 1 to 7 is implemented.

Images