A power site satisfaction evaluation and feedback method based on speech recognition

Abstract

The present invention provides a method for evaluating and feedbacking electric power site satisfaction based on speech recognition. The method specifically collects audio data of users performing electric power site satisfaction evaluations, segments the collected audio data, obtains a number of audio slices, determines a target audio slice from the obtained audio slices, performs speed processing on the target audio slice, combines the speed-processed target audio slice with the remaining audio slices, transmits the combined audio data to a monitoring background, intercepts the target audio slice from the combined audio data, performs speech recognition on the target audio slice, and determines the satisfaction evaluation content based on the speech recognition content. The present invention can segment the audio data and select the target audio slice for speed processing, thereby ensuring the accuracy of speech recognition for satisfaction evaluation while reducing data processing and improving the efficiency of speech recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of power site monitoring, and in particular to a power site satisfaction evaluation and feedback method based on speech recognition. Background Art

[0002] With technological advancements, speech recognition technology is being applied to a growing number of fields to assist users in using equipment. Equipment deployed in power operations often incorporates voice evaluation recognition capabilities to capture user feedback on maintenance results and usage, thereby enabling improvements to subsequent equipment operation. The accuracy of these results also influences subsequent equipment commissioning. However, when performing voice evaluation, the equipment is affected by ambient audio and noise, such as the sounds of vehicles on the street, the sounds of people around, and electromagnetic noise, resulting in inaccurate results. Existing speech recognition processes often directly utilize common noise reduction algorithms to process the entire audio data. However, during the speech evaluation process, the evaluation results only represent a portion of the overall audio data. Denoising the entire audio data reduces overall speech recognition efficiency and increases the amount of data processing. Furthermore, when performing speech recognition on the entire, noise-reduced audio data, key information may be missed or misrecognized, reducing speech recognition accuracy. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a power site satisfaction evaluation feedback method based on speech recognition. By segmenting the audio data and selecting target audio slices therefrom, noise reduction processing is achieved through double-speed processing to solve the problems of only being able to reduce noise for the overall audio data, large data processing volume, low speech recognition efficiency and easy to cause information omission, and low speech recognition accuracy, so that the speech recognition results for the power site satisfaction evaluation are more accurate and the efficiency of speech recognition is further improved.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A power site satisfaction evaluation and feedback method based on speech recognition, comprising:

[0006] Collecting audio data of users' on-site power satisfaction evaluation, segmenting the collected audio data to obtain a number of audio slices, and determining a target audio slice from the obtained audio slices;

[0007] Process the target audio slice at a double speed, combine the target audio slice after the double speed processing with the remaining audio slices, and transmit the combined audio data to the monitoring background;

[0008] After receiving the combined audio data, the monitoring background intercepts the target audio slice from the combined audio data, performs speech recognition on the target audio slice, and determines the satisfaction evaluation content based on the speech recognition content.

[0009] Furthermore, when combining the target audio slice after the double-speed processing with the remaining audio slices, the start times of all the audio slices are retrieved, and the audio slices are combined according to the order of the start times.

[0010] Furthermore, the specific process of segmenting the collected audio data is: the audio data is segmented and intercepted into minimum time periods through the ffmpeg algorithm, and the remaining length of the audio data after segmentation is determined in real time. When the remaining length of the audio data after segmentation is less than the slice length of the segmentation with the minimum time period, the audio segmentation is completed and the remaining audio data is not processed.

[0011] Furthermore, when determining the target audio slice from the acquired audio slices, speech recognition is performed on each audio slice respectively, and audio slices containing keywords are screened out according to the speech recognition results, and the audio slices containing preset keywords are used as target audio slices.

[0012] Furthermore, when the target audio slice is speed-processed, a target speed value of the target audio slice is also selected. The specific process of selecting the target speed value of the target audio slice is as follows: performing speech recognition on the target audio slice, obtaining an initial speech recognition result of the target audio slice, taking the median of the current preset speed value range as the initial target speed value, speed-processing the target audio slice according to the initial target speed value, performing speech recognition on the speed-processed target audio slice, comparing the speech recognition result of the speed-processed target audio slice with the initial speech recognition result, adjusting the upper limit or lower limit of the preset speed value range according to the comparison result, and taking the adjusted value. The median value of the preset speed value range is used as the updated target speed value, the target audio slice is speed-processed according to the updated target speed value, and speech recognition is performed on the target audio slice after speed processing, and the above steps of comparing the speech recognition result of the target audio slice after speed processing with the initial speech recognition result are repeated, and the upper or lower limit of the preset speed value range is adjusted according to the comparison result, and the target speed value is updated until the lower limit of the preset speed value range is greater than or equal to the upper limit of the preset value range, and the median value of the preset speed value range when the lower limit of the preset speed value range is greater than or equal to the upper limit of the preset value range is used as the target speed value of the target audio slice.

[0013] Furthermore, when intercepting a target audio slice in the combined audio data, the target audio slice is determined according to the start time of the target audio slice and the target speed value.

[0014] Furthermore, after obtaining the satisfaction evaluation results, the monitoring background will match the satisfaction evaluation results with the power field operation work order, and automatically generate the corresponding text evaluation content of the power field operation work order based on the satisfaction evaluation content, and associate the corresponding power field operation work order with the audio data corresponding to the satisfaction evaluation results.

[0015] The beneficial effects of the present invention are:

[0016] It can segment the audio data and filter out the target audio slices from the audio slices obtained by segmentation. When achieving the purpose of noise reduction, only the target audio slices are processed. Speed ​​processing can achieve the purpose of noise reduction. Speed ​​processing can emphasize the human voice part in the target audio slice, thereby filtering out other noises. In order not to affect the information contained in the target audio slice, a corresponding target speed value is determined for each target audio slice, which can ensure that its audibility and comprehensibility are not affected while reducing noise, thereby ensuring the accuracy of speech recognition for satisfaction evaluation. And only speeding up the selected target audio slice can reduce the amount of data processing, thereby improving the efficiency of speech recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described below with reference to the accompanying drawings and examples.

[0019] Example:

[0020] A power site satisfaction evaluation feedback method based on speech recognition, such as Figure 1 As shown, including:

[0021] Collecting audio data of users' on-site power satisfaction evaluation, segmenting the collected audio data to obtain a number of audio slices, and determining a target audio slice from the obtained audio slices;

[0022] Process the target audio slice at a double speed, combine the target audio slice after the double speed processing with the remaining audio slices, and transmit the combined audio data to the monitoring background;

[0023] After receiving the combined audio data, the monitoring background intercepts the target audio slice from the combined audio data, performs speech recognition on the target audio slice, and determines the satisfaction evaluation content based on the speech recognition content.

[0024] When combining the target audio slice after speed processing with the remaining audio slices, the start times of all the audio slices are retrieved, and the audio slices are combined according to the order of the start times.

[0025] The specific process of segmenting the collected audio data is as follows: the audio data is segmented and intercepted into minimum time periods through the ffmpeg algorithm, and the remaining length of the audio data after segmentation is determined in real time. When the remaining length of the audio data after segmentation is less than the slice length of the segmentation with the minimum time period, the audio segmentation is completed and the remaining audio data is not processed.

[0026] Since each audio slice has the same duration, the audio slices can be distinguished based on the start time. The end time of the target audio slice after speed processing will change accordingly, but the start time will not change. The target audio slice can be one or more, so when combining, it is necessary to distinguish them by the start time of the target audio slice, so as to combine them in the original order to prevent subsequent speech recognition errors caused by combination errors.

[0027] Specifically, speech recognition in this embodiment is achieved using a speech recognition model based on the MFCC algorithm. The MFCC algorithm can extract features from audio data, further reducing the impact of other factors such as noise on the speech recognition results of the audio data. The speech recognition model can be a common machine learning model or a neural network model.

[0028] When determining the target audio slice from the acquired audio slices, speech recognition is performed on each audio slice respectively, and audio slices containing keywords are screened out according to the speech recognition results, and the audio slices containing preset keywords are used as target audio slices.

[0029] The preset keywords can be set accordingly according to the evaluation results, such as construction terms in the power site, and evaluation terms such as general, good, and poor.

[0030] When the target audio slice is speed-processed, the target speed value of the target audio slice is also selected. The specific process of selecting the target speed value of the target audio slice is as follows: performing speech recognition on the target audio slice, obtaining the initial speech recognition result of the target audio slice, taking the median value of the current preset speed value range as the initial target speed value, speed-processing the target audio slice according to the initial target speed value, and performing speech recognition on the target audio slice after speed processing, comparing the speech recognition result of the target audio slice after speed processing with the initial speech recognition result, adjusting the upper limit or lower limit of the preset speed value range according to the comparison result, and taking the adjusted preset speed value range as the initial target speed value. Set the median of the speed value range as the updated target speed value, speed-process the target audio slice according to the updated target speed value, and perform speech recognition on the target audio slice after speed processing, repeat the above steps of comparing the speech recognition result of the target audio slice after speed processing with the initial speech recognition result, adjust the upper or lower limit of the preset speed value range according to the comparison result, and update the target speed value until the lower limit of the preset speed value range is greater than or equal to the upper limit of the preset value range, and take the median of the preset speed value range when the lower limit of the preset speed value range is greater than or equal to the upper limit of the preset value range as the target speed value of the target audio slice.

[0031] When determining the upper and lower limits of the preset speed value range, the FFT algorithm is used to calculate the speech recognition model's ability to understand audio data at different playback speeds, and the upper and lower limits of the preset speed value range are determined based on the calculation results.

[0032] The FFT algorithm can be used to calculate the speech recognition model's ability to understand audio at different playback speeds, thereby setting the upper and lower limits of the speed value range to ensure that the sped-up audio data can be recognized by the speech recognition model and prevent information deviations within the target audio slice due to speed processing.

[0033] When intercepting a target audio slice in the combined audio data, the target audio slice is determined according to the start time of the target audio slice and the target speed value.

[0034] Because speed processing causes the duration of the target audio slice to change, when intercepting the target audio slice, it is necessary to determine the start and end times of the target audio slice after speed processing based on the original start time of the target audio slice and the target speed value. And because there may be multiple target audio slices, it is necessary to analyze the impact of each target audio slice to determine the new start and end times of each target audio slice in the combined audio data, so as to intercept the target audio slice.

[0035] After obtaining the satisfaction evaluation results, the monitoring background will also match the satisfaction evaluation results with the power field operation work order, and automatically generate the corresponding text evaluation content of the power field operation work order based on the satisfaction evaluation content, and associate the corresponding power field operation work order with the audio data corresponding to the satisfaction evaluation results.

[0036] The text content of the satisfaction evaluation content can be a satisfaction rating result, such as very good, good, fair, poor, etc. After associating the power field work order with the audio data corresponding to the satisfaction evaluation result, the corresponding audio data can be retrieved at any time to analyze the equipment operating status corresponding to the power field work order.

[0037] The embodiment described above is only a preferred solution of the present invention and does not limit the present invention in any form. Other variations and modifications are possible without exceeding the technical solution described in the claims.

Claims

1. A power site satisfaction evaluation and feedback method based on speech recognition, characterized in that: include: Collecting audio data of users' on-site power satisfaction evaluation, segmenting the collected audio data to obtain a number of audio slices, and determining a target audio slice from the obtained audio slices; Process the target audio slice at a double speed, combine the target audio slice after the double speed processing with the remaining audio slices, and transmit the combined audio data to the monitoring background; After receiving the combined audio data, the monitoring background intercepts the target audio slice from the combined audio data, performs speech recognition on the target audio slice, and determines the satisfaction evaluation content based on the speech recognition content; The specific process of segmenting the collected audio data is as follows: using the ffmpeg algorithm to segment the audio data into minimum time segments, and determining in real time the remaining length of the audio data after segmentation; when the remaining length of the audio data after segmentation is less than the length of the slice segmented by the minimum time segmentation, the audio segmentation is completed and the remaining audio data is not processed; When the target audio slice is speed-processed, the target speed value of the target audio slice is also selected. The specific process of selecting the target speed value of the target audio slice is as follows: performing speech recognition on the target audio slice, obtaining the initial speech recognition result of the target audio slice, taking the median value of the current preset speed value range as the initial target speed value, speed-processing the target audio slice according to the initial target speed value, and performing speech recognition on the target audio slice after speed processing, comparing the speech recognition result of the target audio slice after speed processing with the initial speech recognition result, adjusting the upper limit or lower limit of the preset speed value range according to the comparison result, and taking the adjusted preset speed value range as the initial target speed value. Set the median of the speed value range as the updated target speed value, speed-process the target audio slice according to the updated target speed value, and perform speech recognition on the target audio slice after speed processing, repeat the above steps of comparing the speech recognition result of the target audio slice after speed processing with the initial speech recognition result, adjust the upper or lower limit of the preset speed value range according to the comparison result, and update the target speed value until the lower limit of the preset speed value range is greater than or equal to the upper limit of the preset value range, and take the median of the preset speed value range when the lower limit of the preset speed value range is greater than or equal to the upper limit of the preset value range as the target speed value of the target audio slice.

2. The method for evaluating and feedback on electric power site satisfaction based on speech recognition according to claim 1, characterized in that: When combining the target audio slice after speed processing with the remaining audio slices, the start times of all the audio slices are retrieved, and the audio slices are combined according to the order of the start times.

3. The method for evaluating and feedback on electric power site satisfaction based on speech recognition according to claim 1, characterized in that: When determining the target audio slice from the acquired audio slices, speech recognition is performed on each audio slice respectively, and audio slices containing keywords are screened out according to the speech recognition results, and the audio slices containing preset keywords are used as target audio slices.

4. The method for evaluating and feedback on electric power site satisfaction based on speech recognition according to claim 1, characterized in that: When intercepting a target audio slice in the combined audio data, the target audio slice is determined according to the start time of the target audio slice and the target speed value.

5. The method for evaluating and feedback on electric power site satisfaction based on speech recognition according to claim 1, characterized in that: After obtaining the satisfaction evaluation results, the monitoring background will also match the satisfaction evaluation results with the power field operation work order, and automatically generate the corresponding text evaluation content of the power field operation work order based on the satisfaction evaluation content, and associate the corresponding power field operation work order with the audio data corresponding to the satisfaction evaluation results.

Images