Emotion perception and guest room management method based on voiceprint recognition

Abstract

The invention provides an emotion perception and guest room management method based on voiceprint recognition, and the method comprises the steps: generating a service configuration adjustment scheme matched with the emotion tendency of a user according to a demand prediction result, carrying out the fine adjustment of service contents according to the real-time change captured by dynamic data, and obtaining adjusted service parameters; if there is a deviation between the optimized emotional periodic change rule and the actual user feedback, updating the voiceprint feature matrix and the emotional tendency distribution by re-collecting the long-term data record, and determining a new periodic fluctuation feature; and according to the new periodic fluctuation characteristics, the construction of an emotion portrait model and the personalized demand prediction process are executed circularly, the pertinence of service configuration adjustment is continuously improved, and a final optimized service quality improvement scheme is obtained.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an emotion perception and guest room management method based on voiceprint recognition. Background Art

[0002] As a key branch of artificial intelligence and affective computing, voiceprint recognition technology holds significant value in enhancing personalized services and user experience, showing significant potential in hotel room management. Analyzing a person's voice characteristics to perceive their emotional state not only helps the service industry more accurately understand customer needs but also provides data support for optimizing service processes. However, current approaches still have significant shortcomings, primarily in their ability to dynamically capture and analyze user emotional data over time. Many solutions focus solely on collecting sound data from a single scenario or a short period of time, ignoring the complex dynamics of user emotions across time and environment. This results in a lack of continuity and targeted service adjustments.

[0003] Against this backdrop, the core challenges facing this field are gradually becoming apparent. The cyclical patterns of emotional states are difficult to effectively identify. Because emotions are influenced by multiple factors, such as time and context, relying solely on fragmented data records cannot construct a complete user emotional profile, which in turn limits the accuracy of service predictions. Furthermore, the inadequate identification of these cyclical changes further complicates the prediction of personalized needs. The inability to extract user emotional tendencies within specific time periods from long-term data means that service configurations fail to truly meet user expectations and may even lead to user dissatisfaction. These two challenges, layered together, constitute a bottleneck in current technology applications.

[0004] Therefore, how to record and analyze users' voiceprint emotional data over a long period of time, identify their periodic changes, and accurately predict their emotional needs in a specific time period has become a key issue in improving the quality of room management services. Summary of the Invention

[0005] The present invention provides an emotion perception and guest room management method based on voiceprint recognition, which mainly includes:

[0006] The user's voice data in the room management scenario is recorded in real time through voiceprint collection equipment. The sound signals are preliminarily classified according to different time periods and environmental conditions to obtain the original voiceprint dataset for subsequent emotional state analysis and feature extraction, and a preliminarily classified voiceprint signal set is obtained;

[0007] For the voiceprint signal set after preliminary classification, voiceprint feature extraction technology is used to deeply process each sound signal, from which the acoustic feature parameters related to emotion are separated. A voiceprint feature matrix containing multi-dimensional features such as pitch, volume and speaking speed is constructed to determine the basic data of emotional characteristics.

[0008] Based on the voiceprint feature matrix, we use time series analysis methods to mine the regularity of long-term recorded emotional feature basic data, identify the cyclical changes in users' emotions in different time periods, and determine the time nodes and change trends of cyclical fluctuations;

[0009] If the identified periodic fluctuation time nodes and change trends show significant patterns, a user emotion profile model is constructed based on these patterns. This model combines long-term data records with scenario-influencing factors to obtain the user's emotional tendency distribution in a specific time period.

[0010] Through the user emotion profiling model, we dynamically analyze the distribution of emotional tendencies in a specific time period. By combining time period division and scene influencing factors, we can predict the user's personalized needs in the future time period and determine the demand forecast results.

[0011] Based on demand forecast results, a service configuration adjustment plan is generated to match user sentiment trends. Based on real-time changes captured by dynamic data, service content is fine-tuned to obtain adjusted service parameters.

[0012] The service process of the guest room management system is updated in real time through the adjusted service parameters. Combined with the feedback data from the service configuration adjustment, the user emotion profile model is continuously optimized to obtain a more accurate emotional cycle change pattern.

[0013] If there is a deviation between the optimized emotion cycle change pattern and the actual user feedback, long-term data records will be collected again to update the voiceprint feature matrix and emotion tendency distribution to determine the new periodic fluctuation characteristics;

[0014] Based on the new periodic fluctuation characteristics, the construction of the emotion portrait model and the personalized demand prediction process are cyclically executed to continuously improve the targeted service configuration adjustment and obtain the final optimized service quality improvement plan.

[0015] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0016] The present invention discloses an emotion perception and room management method based on voiceprint recognition. By collecting and classifying user voice data in real time, extracting a voiceprint feature matrix, analyzing the changing patterns of emotion cycles, and constructing a user emotion profile model, the system can predict the user's personalized needs during a specific time period and dynamically adjust service configuration accordingly, achieving intelligent and personalized room management services. Through technologies such as voiceprint collection, feature extraction, and time series analysis, the present invention achieves accurate perception and prediction of the user's emotional state, proactively adjusting service content to meet user needs, and effectively improving the quality of room management services and user satisfaction. Furthermore, the present invention also has a self-optimization function that can continuously improve model accuracy based on actual feedback, achieving continuous improvement in service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of an emotion perception and room management method based on voiceprint recognition according to the present invention.

[0018] Figure 2 This is a schematic diagram of an emotion perception and room management method based on voiceprint recognition of the present invention Figure 1 .

[0019] Figure 3 This is a schematic diagram of an emotion perception and room management method based on voiceprint recognition of the present invention Figure 2 . DETAILED DESCRIPTION

[0020] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] like Figure 1-3 This embodiment provides an emotion perception and room management method based on voiceprint recognition, including:

[0022] Step S101: Use a voiceprint acquisition device to record the user's voice data in the room management scenario in real time, perform preliminary classification of the sound signals under different time periods and environmental conditions, obtain the original voiceprint dataset for subsequent emotional state analysis and feature extraction, and obtain a preliminary classified voiceprint signal set.

[0023] A voiceprint acquisition device is used to record user voice signals in a guest room management scenario, and a first voiceprint dataset is obtained using microphone array technology. Based on the first voiceprint dataset, a filter is used to process ambient noise to obtain a second voiceprint dataset. If the signal-to-noise ratio of the second voiceprint dataset is higher than a preset threshold, the second voiceprint dataset is classified according to time period and environmental conditions using a classification algorithm to obtain a first classified signal set. Frequency and amplitude features are extracted from the first classified signal set, and the first classified signal set is processed using fast Fourier transform technology to obtain a first feature set and store it in a database. The first feature set is matched with a preset emotional state template to determine the emotion category corresponding to the first feature set. Based on the emotion category and the first feature set, a classification report is generated and stored in a database to determine the final voiceprint signal analysis results.

[0024] Specifically, in the guest room management scenario, the user's voice data is recorded in real time through a voiceprint collection device. A high-sensitivity microphone array device can be used, with a sampling rate of 44.1kHz and a bit depth of 16bit to ensure the capture of high-quality sound signals and cover the user's voice data at different locations in the guest room. Next, a preliminary classification of sound signals under different time periods and environmental conditions can be performed. The time domain analysis algorithm can be used to divide a day into three peak periods: 6:00-9:00 in the morning, 11:00-14:00 in the afternoon, and 18:00-22:00 in the evening. Combined with the environmental noise detection module, the background noise intensity is identified (for example, below 30dB is a quiet environment, and above 50dB is a noisy environment). Short-time energy analysis and zero-crossing rate calculation are used to preliminarily divide the sound signal into speech signals and non-speech signals, resulting in the original voiceprint dataset, of which speech signals account for approximately 60%. Subsequently, for subsequent emotional state analysis and feature extraction, the Mel-Frequency Cepstral Coefficient (MFCC) algorithm is used to extract voiceprint features. The frame length is set to 25ms and the frame shift is 10ms. 13th-order MFCC coefficients are calculated, and principal component analysis (PCA) dimensionality reduction is combined to retain 95% of the feature variance to ensure feature validity. During this process, the initially classified voiceprint signal set is further subdivided into signal categories using a support vector machine (SVM) classifier. The kernel function is set to the radial basis function (RBF) and the parameter C is optimized to 1.0, achieving a classification accuracy of over 85%, forming the classified voiceprint signal set. These steps are closely linked through algorithms and data processing. The classification results of the previous step directly influence the accuracy of the feature extraction in the subsequent step. Furthermore, the system integrates with guest room management scenarios, such as adjusting the classification threshold based on time period and ambient noise to ensure a speech recognition rate of at least 80% in noisy environments, thus providing reliable data support for subsequent emotional analysis.

[0025] In step S102, for the voiceprint signal set after preliminary classification, voiceprint feature extraction technology is used to deeply process each sound signal, separate the acoustic feature parameters related to emotions, construct a voiceprint feature matrix containing multi-dimensional features such as pitch, volume and speaking speed, and determine the basic data of emotional characteristics.

[0026] Initial data is obtained from the voiceprint signal, and acoustic features related to emotions are separated using feature extraction tools to obtain a preliminary acoustic feature set. Based on the preliminary acoustic feature set, quantification processing is performed on pitch parameters, volume changes, and speech rhythm to determine the corresponding multidimensional feature data. Through the multidimensional feature data, a feature matrix containing pitch parameters, volume changes, and speech rhythm is constructed to obtain a comprehensive feature matrix structure. If the eigenvalue of a certain dimension in the comprehensive feature matrix structure deviates from the preset threshold range, the dimensional data is standardized to obtain an adjusted feature matrix. Based on the adjusted feature matrix, the support vector machine algorithm is used to classify the emotion data to determine the distribution of features related to emotions. For the classified emotion data, the mapping relationship corresponding to the basic data is obtained to determine the final emotion feature basic data set. The final emotion feature basic data set is verified in combination with the pre-established acoustic feature model to obtain verified emotion feature data.

[0027] Specifically, for the initially classified voiceprint signal set, each segment is preprocessed using digital signal processing techniques. Assuming the input audio signal sampling rate is 16kHz, the signal is framed using a Hanning window, with each frame length of 25 milliseconds and a frame shift of 10 milliseconds to reduce spectral leakage. Subsequently, the spectral features of each frame are calculated using the Short-Time Fourier Transform (STFT). The fundamental frequency (F0) is extracted as the pitch feature, calculated between 75Hz and 300Hz. The first 13 coefficients of the Mel-Frequency Cepstral Coefficients (MFCC) are combined to capture timbre variations in the sound, forming a preliminary feature set. Next, for volume features, the short-term energy of each frame is calculated and normalized to a range between 0 and 1 using a logarithmic transformation. For example, an energy value of 0.85 for a frame reflects sound intensity. Speech rate is quantified by measuring the ratio of pause time to effective speech time. For example, assuming a 3-second audio segment has 2.1 seconds of effective speech time, the speech rate ratio is 0.7, reflecting the rhythm of speech. Furthermore, the extracted multi-dimensional features such as pitch, volume and speaking speed are integrated into a voiceprint feature matrix. Assuming the matrix dimension is N×15, where N is the number of frames and 15 is the feature dimension (including F0, 13 MFCC coefficients and short-time energy), the matrix is ​​subjected to principal component analysis (PCA) to reduce the dimension to a dimension with a main contribution rate of 95%, such as 10 dimensions, to reduce redundant information. On this basis, combined with the statistical analysis of emotion-related acoustic features, the mean and variance of each dimension are calculated. For example, the mean of pitch is 120Hz and the variance is 25Hz. 2 , serving as the basic data for emotional features. This data can be used for subsequent emotion classification model training, forming a complete logical chain from signal processing to feature construction, ensuring that the feature matrix effectively represents emotional fluctuations.

[0028] Step S103: Based on the voiceprint feature matrix, a time series analysis method is used to mine the regularity of the long-term recorded emotional feature basic data, identify the periodic change pattern of the user's emotions in different time periods, and determine the time nodes and change trends of the periodic fluctuations.

[0029] Voiceprint feature data is acquired and an initial feature matrix is ​​constructed. This initial feature matrix contains long-term recorded voiceprint feature data. A normalization method is used to normalize the initial feature matrix to obtain a processed feature matrix. Based on this processed feature matrix, a time series analysis method is applied to decompose the feature distributions for different time periods to obtain a fluctuation sequence of emotional features. This fluctuation sequence, combined with a preset threshold range, is used to determine the periodic variation patterns of emotional features. If the fluctuation sequence exceeds the threshold range within a certain time period, the time period is marked as a significant fluctuation interval, resulting in a set of significant fluctuation time nodes. Based on this set of significant fluctuation time nodes, a sliding window method is used to analyze the time intervals and fluctuation amplitudes between adjacent nodes, extracting local variation trends and determining the direction of emotional variation within each time period. Based on these local variation trends and the historical distribution of the long-term recorded data, an emotional cyclical fluctuation model is constructed to identify the primary time nodes of cyclical fluctuations. Based on these primary time nodes, the emotional feature data is classified using the K-means clustering method to obtain a category distribution of emotional variation within different time periods. Based on this category distribution and the continuity characteristics of the fluctuation sequence, the transition patterns between categories are analyzed to determine the trend of the user's emotional cyclical variation.

[0030] Specifically, the emotional data mining process for the voiceprint feature matrix involves preprocessing the long-term voiceprint data. For example, we collect daily voiceprint data from a user for six consecutive months. Each day, we extract feature values, including pitch mean, volume standard deviation, and speech rate change, to form a feature matrix containing 540 time points (6 months × 30 days × 3 times / day). Next, we use time series analysis to identify patterns in this data. We employ an autoregressive integrated moving average (ARIMA) model with parameters set to (p = 1, d = 1, q = 1). By performing differential processing on the pitch mean series to eliminate trend fluctuations, we calculate the autocorrelation coefficient of the residual series and find a significant period of 7 days, indicating that user emotions may exhibit a weekly cycle. We then identify the time points of periodic fluctuations and perform frequency domain analysis on the feature matrix using a Fourier transform. The period corresponding to the main frequency extracted is 7.2 days. Combined with the timestamp data, we determine that Mondays at 9:00 AM represent a low point in emotional trough, and Fridays at 5:00 PM represent a high point in emotional trough, with an error margin of ±2 hours. Further analysis of the changing trends, using a sliding window method (with a 14-day window size) to calculate the slope of the change in the mean pitch, revealed a positive slope of 0.15 in the first week of each month, indicating an upward trend in sentiment, while the slope was negative 0.12 in the fourth week, indicating a downward trend in sentiment. Finally, integrating cyclical patterns with business operations, the system automatically generates a mood fluctuation report. Based on the identified 7-day cycle and key nodes, the system predicts the lowest mood periods for the coming week and connects with the user's schedule data. If the low mood period overlaps with an important meeting, the meeting reminder frequency is automatically adjusted from the default 2 hours to 4 hours, ensuring that users are adequately prepared and forming a data-driven, intelligent, and assisted decision-making closed loop.

[0031] Step S104: If there are significant patterns in the identified periodic fluctuation time nodes and change trends, a user emotion profile model is constructed based on these patterns, and the long-term data records are combined with scene influencing factors to obtain the user's emotional tendency distribution in a specific time period.

[0032] Data related to periodic fluctuations and time nodes are extracted from historical records to construct an initial dataset. Preprocessing methods are used to remove noise from the initial dataset to obtain a cleaned fluctuation dataset. Based on the cleaned fluctuation dataset, the relationship between time nodes and specific times is analyzed to determine whether there is a significant periodic pattern. If the fluctuation data exhibits a repeating pattern within a specific time period, it is determined to be periodic data. A feature matrix is ​​constructed for the periodic data, integrating the correlation between influencing factors and changing trends to obtain a characterized emotion-related dataset. Using this characterized emotion-related dataset, a support vector machine algorithm is applied to train a user model, deriving a classification result for emotion tendencies and obtaining a preliminary emotion tendency distribution. Based on this preliminary emotion tendency distribution, the distribution result is corrected for time, combined with data from specific times and the periodic fluctuations. If the corrected distribution deviation exceeds a preset threshold, the feature matrix is ​​readjusted to obtain corrected emotion distribution data. This corrected emotion distribution data is mapped to the user model to construct a complete emotion profile. A determination is made as to whether the emotion profile covers all specific time periods to obtain the final user emotion profile. According to the final user emotion portrait result, the distribution derivation and change trend are associated to generate emotion tendency prediction data within a specific time period to complete the dynamic monitoring of user emotions.

[0033] Specifically, an initial dataset is constructed by extracting data related to periodic fluctuations and time nodes from historical records. Preprocessing is then used to remove noise, resulting in a cleaned fluctuation dataset. Based on this cleaned fluctuation dataset, the relationship between time nodes and specific time periods is analyzed to determine whether significant periodic patterns exist. If the fluctuation data exhibits a repeating pattern within a specific time period, it is identified as periodic data. For this identified periodic data, a feature matrix is ​​constructed by combining long-term data and scenario factors. This matrix integrates the correlation between influencing factors and changing trends to obtain a characterized emotion-related dataset. Using this characterized emotion-related dataset, a support vector machine algorithm is applied to train a user model, deriving emotional tendency classification results and obtaining a preliminary emotional tendency distribution. Based on this preliminary emotional tendency distribution, the distribution is corrected for time, combined with data from specific time periods and periodic fluctuations. If the corrected distribution deviation exceeds a preset threshold, the feature matrix is ​​readjusted to obtain the corrected emotional distribution data. This corrected emotional distribution data is then mapped onto the user model to construct a complete emotional profile. The final user emotional profile is then determined to ensure that the emotional profile covers all specific time periods. Based on the final user emotion portrait results, associated distribution derivation and change trends, the emotion tendency prediction data within a specific time period is generated to complete the dynamic monitoring of user emotions.

[0034] By analyzing user social media data, we identify the timing and trends of cyclical fluctuations. Assume that daily user text data containing emotional keywords (e.g., "happy" appears 500 times, "anxious" appears 300 times) is collected from January 2024 to March 2025. First, we use a fast Fourier transform (FFT) algorithm to perform a time-series analysis of the frequency of emotional keywords and calculate cyclical fluctuations. We find that the keyword "happy" peaks in frequency on Mondays and Fridays (averaging 50 times on Mondays and 45 times on Fridays), while "anxious" peaks on Wednesdays (averaging 35 times), with a 7-day period. A significance test p-value of < 0.05 confirms the existence of a pattern. Next, a user sentiment profiling model was constructed. Using the K-means clustering algorithm (K=3), sentiment was categorized into three categories: positive, neutral, and negative. Based on word frequency vectors (1000-dimensional, including keywords and contextual features), clustering results showed that 60% of the data was positive, 25% was neutral, and 15% was negative, with Euclidean distances from the center points of 0.8, 0.6, and 0.9, respectively. Combining long-term data with contextual factors, the model used user location (70% in City A), weather (60% in sunny days), and holidays (a 20% increase in positive sentiment during the Spring Festival) as input features. A random forest algorithm was used to predict sentiment, achieving an accuracy of 85% on the training set and 80% on the test set. Finally, the sentiment distribution for a specific time period (e.g., Monday from 9:00 AM to 11:00 AM) was calculated, yielding a 65% probability of positive sentiment, 20% of neutral sentiment, and 15% of negative sentiment. A heat map was then used to visualize sentiment trends across time and context, ensuring that the model's predictions were consistent with actual scenarios (e.g., holiday effects), completing a closed-loop validation process.

[0035] Step S105, through the user emotion portrait model, dynamically analyze the emotional tendency distribution in a specific time period, combine the time period division and scene influencing factors, predict the user's personalized needs in the future time period, and determine the demand prediction result.

[0036] User interaction data and scenario factor data are obtained from a multi-source data platform, and data cleaning techniques are used to remove noise data to obtain a structured data set. Using the structured data set, a pre-established portrait model is used to extract user emotional features and determine the user emotional distribution. If the change in the user emotional distribution exceeds a preset threshold within a specified time period, the time period is dynamically divided to obtain a time division result. Based on the time division results and scenario factor data, a random forest algorithm is used to analyze the changing trend of the emotional distribution to obtain the emotional change trend. Based on the emotional change trend and scenario factor data, a user personalized demand prediction model is constructed, and a logistic regression algorithm is used to predict user demand to obtain the demand prediction result. If the confidence level of the demand prediction result is lower than the preset threshold, the Bayesian network is used to adjust the prediction model based on historical data and scenario factors to obtain an optimized prediction result. Based on the optimized prediction results, a user personalized demand report is generated to determine the final demand prediction result.

[0037] Specifically, a user sentiment profiling model dynamically analyzes the distribution of sentiment over a specific time period. This involves first collecting social media data and app usage records from users over the past 30 days. Assume that daily sentiment data consists of three categories: positive, negative, and neutral, accounting for 40%, 30%, and 30%, respectively. Using sentiment analysis algorithms such as support vector machines (SVMs), the text content is scored for sentiment polarity. Combined with timestamp data, sentiment is broken down by hourly timeframes. For example, during work hours (8:00 AM - 5:00 PM), negative sentiment rises to 35%, while during evening hours (6:00 PM - 10:00 PM), positive sentiment increases to 45%. Next, combining timeframes with contextual factors, such as weekday and weekend differences, reveals that negative sentiment peaks at 38% on Mondays, while positive sentiment rises to 50% on weekends, presumably due to work stress and leisure activities. Furthermore, using time series analysis algorithms such as ARIMA models to predict user sentiment over the next seven days, the team concludes that negative sentiment is likely to remain at 36% during work hours on Mondays, while positive sentiment is likely to drop to 42% during evening hours. Based on this, we predict personalized user needs. For example, by pushing stress-relieving content on weekdays, we can expect a 20% increase in meditation app downloads, while by pushing entertainment activities on weekends, we expect a 15% increase in participation. Ultimately, the demand forecast results are verified using machine learning models (such as random forests). Combined with historical data, the error rate is controlled within 5%. It is determined that users will be more inclined towards stress-relieving and entertainment services in the future. The system automatically generates personalized recommendation lists to ensure that the click-through rate of recommended content increases by more than 10%. Through the above analysis, a complete logical chain is formed from data collection to sentiment prediction to demand recommendation, relying on algorithms and numerical analysis to ensure the accuracy of the results.

[0038] Step S106: Generate a service configuration adjustment plan that matches the user's emotional tendency based on the demand forecast result, fine-tune the service content according to the real-time changes captured by the dynamic data, and obtain adjusted service parameters.

[0039] Receive the user emotional tendency data captured in real time, compare it with the pre-established emotional classification rule library, extract features for at least one emotional category, obtain the emotional intensity value and tendency label, and obtain a preliminary emotional judgment result. According to the tendency label in the emotional judgment result, combined with the service usage record in the dynamic data, a preset threshold range is used for matching. If the emotional intensity value exceeds the threshold range, the configuration adjustment logic is triggered to determine the direction of service parameter adjustment. According to the service parameter adjustment direction, obtain a configuration scheme that matches the emotional tendency from the preset service content template library, fine-tune the real-time changing data, determine whether the adjusted parameters meet the preset balance standards, and obtain the final parameter configuration combination. Through the parameter configuration combination, use a data mapping tool to apply the adjusted parameters to the service content distribution process, obtain real-time feedback data, determine whether the parameter value needs to be further fine-tuned, and determine the final presentation form of the service content.

[0040] Specifically, during the implementation of service configuration adjustments, user data is first analyzed using a demand forecasting model. Assuming the system has collected user service request data from the past 30 days and found that the peak request volume is 5,000 per day, it is predicted that the peak volume may increase to 5,500 within the next 7 days. Using a time series-based ARIMA model for forecasting, the forecast error range is calculated to be ±5%, thus ensuring a 95% confidence level for the forecast value. Next, combined with user sentiment analysis, natural language processing technology is used to assign sentiment scores to user feedback text. Assuming a score range of -1 to 1, the current average user sentiment score is 0.3, indicating a positive sentiment. However, some negative feedback focuses on long response times. The system automatically extracts the keywords "wait" and "slow," which account for 20% of the total. Correlation analysis between sentiment and keywords indicates that response time needs to be optimized. Subsequently, the system monitored the service response time in real time based on the real-time changes captured by dynamic data, and found that the current average response time was 3.2 seconds, 2.5 seconds higher than the target value. Combined with the predicted increase in request volume, the system automatically triggered the fine-tuning mechanism, calling the load balancing algorithm to increase the server resource allocation ratio from the original 60% to 75%, while increasing the cache hit rate target from 80% to 85%. Through simulation testing, the adjusted response time is expected to drop to 2.4 seconds. Finally, after obtaining the adjusted service parameters, the system generates a new configuration plan, including increasing the number of server cluster nodes from 10 to 12 and increasing the bandwidth limit from 100Mbps to 120Mbps. The new configuration plan is automatically pushed to the server through the API interface to complete the configuration update. The entire process forms a closed-loop logic, ensuring seamless integration of prediction, sentiment analysis, and real-time adjustments to enhance the user experience.

[0041] In step S107, the service process of the guest room management system is updated in real time through the adjusted service parameters. The user emotion portrait model is continuously optimized in combination with the feedback data of the service configuration adjustment to obtain a more accurate emotion cycle change pattern.

[0042] By acquiring service parameter adjustment records from the guest room management system and analyzing their impact on the service process, a preliminary trend of service response changes is derived. Based on these service response trends and combined with the configuration feedback data stream, a comparison is performed using preset thresholds to determine whether abnormal fluctuations exist and identify service components requiring optimization. If abnormal fluctuations are detected in a service component, relevant emotional expression features are extracted from the user's emotional feedback data to determine the distribution pattern of emotional fluctuations. This emotional fluctuation distribution pattern is analyzed and input into a pre-established support vector machine model to derive potential patterns of user emotional cycles. Based on the derived emotional cycle patterns, the parameters of the user emotional profile model are adjusted to obtain more realistic emotional profile updates. The updated emotional profile results are compared with historical data to analyze their consistency and deviation, determining the direction and extent of model optimization. Based on the direction and extent of model optimization, the service parameter configuration strategy is dynamically adjusted to generate a service process improvement plan for the guest room management system.

[0043] Specifically, the service process of the guest room management system is updated in real time through the adjusted service parameters. The system first automatically adjusts the service response time and resource allocation based on historical data and real-time user feedback.

[0044] For example, the system detected a 30% increase in room service requests during peak hours (6:00 PM to 9:00 PM daily). Using dynamic scheduling algorithms (such as weighted round-robin), it adjusted the waiter allocation ratio from 1:10 to 1:8, shortening the response time from 5 minutes to 3.5 minutes. The algorithm calculated the optimal path based on the waiter's location and task saturation (the saturation threshold is set at 0.8), reducing ineffective movement distance by 15%. In conjunction with feedback data on service configuration adjustments, the system collects user ratings (1-5 points) and text reviews, and uses natural language processing (BERT model) to analyze sentiment trends. Scores above 4.2 indicate positive sentiment, while scores below 3.0 indicate negative sentiment, with an accuracy rate of 92%. The system receives approximately 500 pieces of feedback data daily, and fine-tunes the sentiment analysis model every hour through an incremental update mechanism, optimizing the F1 score from 0.85 to 0.89. The system continuously optimizes its user emotion profile model, using time series analysis (ARIMA model, p=2, d=1, q=1) and combining user length of stay (average 3.2 days) and spending behavior (average daily spending of 500 yuan) to predict emotional cycles. The system found that the probability of negative emotions increased by 20% after 6:00 PM on the second day of their stay. Based on this, service strategies were adjusted, such as increasing the frequency of proactive greetings during this period (from once every four hours to once every two hours), which improved user satisfaction by 12%. Through these steps, the system forms a closed-loop optimization logic, with service process adjustments and emotion model updates mutually feedback-generating. The error in emotion cycle prediction is kept within ±5%, ensuring precise service.

[0045] Step S108: If there is a deviation between the optimized emotion cycle variation pattern and the actual user feedback, long-term data records are collected again, the voiceprint feature matrix and emotion tendency distribution are updated, and new periodic fluctuation characteristics are determined.

[0046] The user's long-term voice recordings are obtained through the data acquisition module to construct an initial voiceprint feature data set. Key voiceprint parameters are extracted from the voiceprint feature data set to construct a voiceprint feature matrix and store it in the preset database. Based on the voiceprint feature matrix, the support vector machine algorithm is used to classify emotional tendencies to obtain preliminary emotional tendency distribution results. If there is a deviation between the emotional tendency distribution results and the user feedback records, the long-term data acquisition mechanism is triggered, the voiceprint feature matrix is ​​updated, and the emotional tendency distribution is recalculated. Based on the updated emotional tendency distribution, the periodic fluctuation characteristics are analyzed to determine the changing trend of the fluctuation characteristics. Based on the changing trend of the fluctuation characteristics and combined with historical data records, a new emotional cycle pattern is derived and stored in the system database. After obtaining the new emotional cycle pattern, the parameters of the pattern optimization module are adjusted to determine the optimized emotional cycle characteristics.

[0047] Specifically, after optimizing the emotional cycle change pattern, if deviations are found from actual user feedback, adjustments and updates can be made through information technology. First, the system will automatically trigger the long-term data collection mechanism, using intelligent voice devices to record user voice data 24 hours a day for 30 consecutive days with user authorization, generating an audio data set with a total length of more than 720 hours, and removing background noise through preprocessing algorithms to ensure that the signal-to-noise ratio reaches more than 20dB. Next, the system will call deep learning models such as convolutional neural networks (CNN) to extract voiceprint features, segment the audio data into segments of 5 seconds, calculate the Mel-frequency cepstral coefficients (MFCC) of each segment, form a feature matrix with a dimension of 13×100, and perform principal component analysis (PCA) dimensionality reduction on the matrix, retaining 95% of the variance information, and generate a new voiceprint feature matrix. Subsequently, the system will analyze the distribution of emotional tendencies based on the updated feature matrix and the Hidden Markov Model (HMM), calculate the probability of the user's daily emotional state in the past 30 days, for example, positive emotions account for 60%, neutral emotions account for 30%, and negative emotions account for 10%, and predict the trend of emotional fluctuations through time series analysis (such as the ARIMA model). It is concluded that the periodic fluctuation characteristics are an emotional peak every 7 days, and the peak emotional intensity is 0.85 (the full score is 1.0). Finally, the system will compare the newly calculated periodic fluctuation characteristics with historical data. If the deviation exceeds 10%, the emotional cycle model in the database will be automatically updated, and through associated business modules, such as the intelligent recommendation system, the emotional tone of the pushed content will be adjusted, such as increasing the push ratio of positive emotional content to 70% to match the user's current emotional cycle and ensure the accuracy of the recommendation. Through the above-mentioned full-process automated processing, the system can dynamically adapt to changes in user emotions and form a data-driven closed-loop optimization logic.

[0048] Step S109, based on the new periodic fluctuation characteristics, the emotional portrait model construction and personalized demand prediction process are cyclically executed to continuously improve the pertinence of service configuration adjustment and obtain the final optimized service quality improvement plan.

[0049] Using a periodic fluctuation feature extraction algorithm, periodic fluctuation features are extracted from user interaction data to determine the fluctuation period and amplitude pattern. Feature analysis techniques are used to classify and cluster the acquired periodic fluctuation features to obtain an initial feature set for the sentiment profile. If the classification accuracy of the initial feature set falls below a preset threshold, the features are re-screened using a random forest algorithm to obtain an optimized sentiment profile feature set. Based on the optimized sentiment profile feature set, a sentiment profile model is constructed and trained using a support vector machine algorithm to obtain a sentiment profile classification model. The sentiment profile classification model is combined with historical user behavior data to predict personalized needs and obtain personalized demand prediction results. Using the personalized demand prediction results, service configuration parameters are adjusted to generate a targeted, optimized service configuration plan. If the deviation between the feedback data of the service configuration plan and the expected quality improvement target exceeds a preset threshold, the sentiment profile model and demand prediction process are iteratively updated to obtain the optimal service quality improvement plan.

[0050] Specifically, based on the new cyclical fluctuation characteristics, we first extract the cyclical characteristics of sentiment profiles through time series analysis. The Fast Fourier Transform (FFT) algorithm is used to process user interaction data. Assume that the dataset contains 10,000 user sentiment ratings (ranging from 0 to 10) with a time span of 30 days. FFT analysis reveals that the main cycles are 7 and 14 days, reflecting the weekly and biweekly fluctuations of user sentiment. Next, we construct a sentiment profile model, using a long short-term memory network (LSTM) to process the time series data. The input is a standardized sequence of sentiment ratings, the hidden layer is set to 128 units, and training is performed for 100 epochs. The optimizer is Adam, the learning rate is 0.001, and the loss function is the mean squared error (MSE). The model outputs a dynamic profile of user sentiment with a prediction accuracy of 85%. Subsequently, personalized demand prediction is performed based on a collaborative filtering algorithm combined with user historical behavior.

[0051] For example, 1,000 service interaction records were used to calculate user preference vectors. Cosine similarity was used to assess inter-user similarity with a threshold of 0.8 to predict potential user demand for services (e.g., 60% preferred a response time < 2 seconds). To continuously improve service configuration, a reinforcement learning algorithm (Q-learning) was used, with a service quality score (0-100) as the reward function. Service parameters were adjusted (e.g., server response time was optimized from 2.5 seconds to 1.8 seconds). After 1,000 iterations, the Q value converged and the service quality score improved by 15%. The final optimization solution was validated through A / B testing, with 1,000 users divided into two groups. The optimized group had a response time of 1.8 seconds, while the control group had a response time of 2.5 seconds. The results showed that user satisfaction in the optimized group increased by 20%, and the service interruption rate decreased to 0.5%. Through periodic analysis, model building, demand forecasting, and configuration adjustment, a closed-loop optimization logic was formed to continuously improve service quality.

[0052] The above disclosure is only a preferred embodiment of the present invention, and it is certainly not intended to limit the scope of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for emotion perception and room management based on voiceprint recognition, characterized in that: The method comprises: The user's voice data in the room management scenario is recorded in real time through voiceprint collection equipment. The sound signals are preliminarily classified according to different time periods and environmental conditions to obtain the original voiceprint dataset for subsequent emotional state analysis and feature extraction, and a preliminarily classified voiceprint signal set is obtained; For the voiceprint signal set after preliminary classification, voiceprint feature extraction technology is used to deeply process each sound signal, from which the acoustic feature parameters related to emotion are separated. A voiceprint feature matrix containing multi-dimensional features such as pitch, volume and speaking speed is constructed to determine the basic data of emotional characteristics. Based on the voiceprint feature matrix, we use time series analysis methods to mine the regularity of long-term recorded emotional feature basic data, identify the cyclical changes in users' emotions in different time periods, and determine the time nodes and change trends of cyclical fluctuations; If the identified periodic fluctuation time nodes and change trends show significant patterns, a user emotion profile model is constructed based on these patterns. This model combines long-term data records with scenario-influencing factors to obtain the user's emotional tendency distribution in a specific time period. Through the user emotion profiling model, the emotional tendency distribution in a specific time period is dynamically analyzed. Combined with the time period division and scene influencing factors, the personalized needs of users in the future time period are predicted and the demand forecast results are determined; Based on demand forecast results, a service configuration adjustment plan is generated to match user sentiment trends. Based on real-time changes captured by dynamic data, service content is fine-tuned to obtain adjusted service parameters. The service process of the guest room management system is updated in real time through the adjusted service parameters. Combined with the feedback data from the service configuration adjustment, the user emotion profile model is continuously optimized to obtain a more accurate emotional cycle change pattern. If there is a deviation between the optimized emotion cycle change pattern and the actual user feedback, long-term data records are collected again, the voiceprint feature matrix and emotion tendency distribution are updated, and new periodic fluctuation characteristics are determined; Based on the new periodic fluctuation characteristics, the construction of the emotion portrait model and the personalized demand forecasting process are cyclically executed to continuously improve the targeted service configuration adjustment and obtain the final optimized service quality improvement plan.

2. The emotion perception and room management method based on voiceprint recognition according to claim 1 is characterized in that: The voice data of the user in the room management scenario is recorded in real time by the voiceprint collection device, and the sound signals are preliminarily classified under different time periods and environmental conditions to obtain the original voiceprint data set for subsequent emotional state analysis and feature extraction. The preliminarily classified voiceprint signal set is obtained, including: The user's voice signal is recorded in the guest room management scenario through a voiceprint collection device, and the first voiceprint dataset is obtained using microphone array technology; According to the first voiceprint dataset, the environmental noise is processed using a filter to obtain a second voiceprint dataset; If the signal-to-noise ratio of the second voiceprint dataset is higher than a preset threshold, classifying the second voiceprint dataset according to time period and environmental conditions using a classification algorithm to obtain a first classification signal set; Extracting frequency and amplitude features from the first classified signal set, processing the first classified signal set using a fast Fourier transform technique, obtaining a first feature set, and storing the first feature set in a database; Matching the first feature set with a preset emotional state template to determine the emotion category corresponding to the first feature set; According to the emotion category and the first feature set, a classification report is generated and stored in a database to determine the final voiceprint signal analysis result.

3. The emotion perception and room management method based on voiceprint recognition according to claim 1 is characterized in that: The voiceprint signal set after preliminary classification is processed in depth using voiceprint feature extraction technology to separate acoustic feature parameters related to emotions, construct a voiceprint feature matrix containing multi-dimensional features such as pitch, volume, and speaking speed, and determine the basic data of emotional features, including: Initial data is obtained from the voiceprint signal, and feature extraction tools are used to separate the acoustic features related to emotions to obtain a preliminary acoustic feature set; Based on the preliminary acoustic feature set, quantify the pitch parameters, volume changes, and speech rate rhythm to determine corresponding multi-dimensional feature data; Constructing a feature matrix including pitch parameters, volume changes, and speech speed rhythm through the multidimensional feature data to obtain a comprehensive feature matrix structure; If the eigenvalue of a certain dimension in the comprehensive feature matrix structure deviates from a preset threshold range, the dimensional data is normalized to obtain an adjusted feature matrix; Based on the adjusted feature matrix, a support vector machine algorithm is used to classify the emotion data to determine the distribution of features related to the emotion; For the emotion data after the classification process, obtaining a mapping relationship corresponding to the basic data to determine a final emotion feature basic data set; The final emotion feature basic data set is combined with a pre-established acoustic feature model for verification to obtain verified emotion feature data.

4. The method of emotion perception and room management based on voiceprint recognition according to claim 1, characterized in that: The method uses a time series analysis method based on the voiceprint feature matrix to conduct regularity mining on the long-term recorded emotional feature basic data, identify the periodic change pattern of the user's emotions in different time periods, and determine the time nodes and change trends of periodic fluctuations, including: Acquire voiceprint feature data and construct an initial feature matrix, wherein the initial feature matrix includes voiceprint feature data recorded over a long period of time; Normalizing the initial feature matrix using a normalization method to obtain a processed feature matrix; Decomposing the feature distribution of different time periods using a time series analysis method based on the processed feature matrix to obtain a fluctuation sequence of emotional features; By combining the fluctuation sequence with a preset threshold range, the periodic variation pattern of the emotional characteristics is determined. If the fluctuation sequence exceeds the threshold range within a certain time period, the time period is marked as a significant fluctuation interval, and a set of significant fluctuation time nodes is obtained. Based on the set of significant fluctuation time nodes, a sliding window method is used to analyze the time intervals and fluctuation amplitudes between adjacent nodes, extract local change trends, and obtain the direction of sentiment change in each time period; By combining the local change trend with the historical distribution of the long-term recorded data, a sentiment cycle fluctuation model is constructed to determine the main time nodes of the periodic fluctuation; According to the main time nodes, the emotional feature data is classified using the K-means clustering method to obtain the category distribution of emotional changes in different time periods; By analyzing the category distribution and the continuity characteristics of the fluctuation sequence, the conversion rules between categories are analyzed to obtain the trend of the user's emotional cycle changes.

5. The emotion perception and room management method based on voiceprint recognition according to claim 1 is characterized in that: If the identified periodic fluctuation time nodes and change trends have significant patterns, a user emotion profile model is constructed based on these patterns. The long-term data records are combined with the scene influencing factors to obtain the user's emotional tendency distribution in a specific time period, including: Extract relevant data on periodic fluctuations and time nodes from historical records to construct an initial data set; Using preprocessing means to remove noise data in the initial data set to obtain a cleaned fluctuation data set; Based on the cleaned fluctuation data set, the relationship between the time nodes and the specific time is analyzed to determine whether there is a significant periodic pattern. If the fluctuation data shows a repetitive pattern within a specific time period, it is determined to be periodic regular data; For the periodic regular data, a feature matrix is ​​constructed, and the correlation between influencing factors and changing trends is integrated to obtain a characterized emotion-related data set; Using the characterized emotion-related data set, a support vector machine algorithm is applied to train the user model, derive the classification results of the emotion tendency, and obtain a preliminary emotion tendency distribution; Based on the preliminary emotional tendency distribution, combined with the data of the specific time and the periodic fluctuation, the distribution result is corrected in the time dimension. If the distribution deviation after correction exceeds a preset threshold, the feature matrix is ​​readjusted to obtain corrected emotional distribution data; The corrected emotion distribution data is mapped to the user model to construct a complete emotion profile, and it is determined whether the emotion profile covers all specific time periods to obtain the final user emotion profile result; According to the final user emotion portrait result, the distribution derivation and change trend are associated to generate emotion tendency prediction data within a specific time period to complete the dynamic monitoring of user emotions.

6. The emotion perception and room management method based on voiceprint recognition according to claim 1 is characterized in that: The user emotion profiling model dynamically analyzes the distribution of emotion tendencies in a specific time period, combines time period division and scenario influencing factors, predicts the user's personalized needs in the future time period, and determines the demand prediction results, including: Obtain user interaction data and scenario factor data from a multi-source data platform, use data cleaning technology to remove noise data, and obtain a structured data set; Through structured data sets, we use pre-established portrait models to extract user emotional features and determine user emotional distribution; If the change in the user's emotion distribution within a specified time period exceeds a preset threshold, the time period is dynamically divided to obtain a time division result; Based on the time division results and scene factor data, the random forest algorithm is used to analyze the changing trend of emotion distribution and obtain the emotion change trend; By using emotion change trends and scenario factor data, a user personalized demand prediction model is constructed, and a logistic regression algorithm is used to predict user needs to obtain demand prediction results; If the confidence level of the demand forecast result is lower than the preset threshold, the Bayesian network is used to adjust the forecast model based on historical data and scenario factors to obtain an optimized forecast result; Through the optimized forecast results, a user personalized demand report is generated to determine the final demand forecast results.

7. The method of emotion perception and room management based on voiceprint recognition according to claim 1, characterized in that: The above method generates a service configuration adjustment plan that matches the user's emotional tendency based on the demand forecast results, fine-tunes the service content according to the real-time changes captured by the dynamic data, and obtains the adjusted service parameters, including: Receive the user's emotional tendency data captured in real time, compare it with the pre-established emotional classification rule library, extract features for at least one emotional category, obtain the emotional intensity value and tendency label, and obtain a preliminary emotional judgment result; Based on the tendency label in the emotion judgment result, combined with the service usage record in the dynamic data, a preset threshold range is used for matching. If the emotion intensity value exceeds the threshold range, the configuration adjustment logic is triggered to determine the direction of service parameter adjustment; According to the service parameter adjustment direction, a configuration scheme that matches the emotional tendency is obtained from a preset service content template library, and fine-tuned according to the real-time change data. It is determined whether the adjusted parameters meet the preset balance standard to obtain the final parameter configuration combination; Through the parameter configuration combination, the adjusted parameters are applied to the service content distribution process using a data mapping tool to obtain real-time feedback data, determine whether further fine-tuning of parameter values ​​is required, and determine the final presentation form of the service content.

8. The method of emotion perception and room management based on voiceprint recognition according to claim 1, characterized in that: The adjusted service parameters are used to update the service process of the guest room management system in real time. Combined with the feedback data from the service configuration adjustment, the user emotion profile model is continuously optimized to obtain more accurate emotion cycle change patterns, including: By obtaining the adjustment records of service parameters from the guest room management system, we can analyze the impact of these adjustments on the service process and obtain preliminary service response change trends. Based on the service response change trend and the configuration feedback data stream, we use preset thresholds for comparison to determine whether there are abnormal fluctuations and identify the service links that need optimization. If abnormal fluctuations in the service process are detected, relevant emotional expression features are extracted from the user's emotional feedback data to obtain the distribution pattern of emotional fluctuations; By analyzing the distribution pattern of emotional fluctuations and inputting it into a pre-established support vector machine model, the potential change patterns of the user's emotional cycle are derived; According to the derived emotion cycle change rules, the parameter configuration of the user emotion profile model is adjusted to obtain more realistic emotion profile update results; By comparing the updated sentiment profile results with historical data, we analyze their consistency and deviation levels to determine the direction and magnitude of model optimization. According to the direction and magnitude of model optimization, the configuration strategy of service parameters is dynamically adjusted to generate a service process improvement plan for the guest room management system.

9. The method of emotion perception and room management based on voiceprint recognition according to claim 1, characterized in that: If the optimized emotion cycle variation pattern deviates from actual user feedback, long-term data records are collected again, the voiceprint feature matrix and emotion tendency distribution are updated, and new periodic fluctuation characteristics are determined, including: The data collection module obtains the user's long-term voice records and constructs the initial voiceprint feature dataset; Extract key voiceprint parameters from the voiceprint feature dataset, construct a voiceprint feature matrix, and store it in a preset database; Based on the voiceprint feature matrix, the support vector machine algorithm is used to classify emotional tendencies and obtain preliminary emotional tendency distribution results; If there is a deviation between the emotional tendency distribution result and the user feedback record, the long-term data collection mechanism will be triggered, the voiceprint feature matrix will be updated, and the emotional tendency distribution will be recalculated; Based on the updated emotional tendency distribution, analyze the periodic fluctuation characteristics and determine the changing trend of the fluctuation characteristics; By combining the changing trend of fluctuation characteristics with historical data records, new emotional cycle patterns are derived and stored in the system database; After obtaining the new emotional cycle pattern, the parameters of the pattern optimization module are adjusted to determine the optimized emotional cycle characteristics.

10. The emotion perception and room management method based on voiceprint recognition according to claim 1 is characterized in that: Based on the new periodic fluctuation characteristics, the emotional portrait model construction and personalized demand forecasting process are cyclically executed to continuously improve the pertinence of service configuration adjustment and obtain the final optimized service quality improvement plan, including: Through the periodic fluctuation feature extraction algorithm, the periodic fluctuation features are obtained from the user interaction data to determine the fluctuation period and amplitude pattern; Using feature analysis technology, the acquired periodic fluctuation features are classified and clustered to obtain the initial feature set of the emotional portrait; If the classification accuracy of the initial feature set is lower than the preset threshold, the features are re-screened using the random forest algorithm to obtain an optimized emotional profile feature set; Based on the optimized emotion portrait feature set, an emotion portrait model is constructed and trained using the support vector machine algorithm to obtain an emotion portrait classification model; Through the emotional portrait classification model, combined with the user's historical behavior data, personalized needs are predicted and personalized demand prediction results are obtained; Use personalized demand forecast results to adjust service configuration parameters and generate targeted optimized service configuration plans; If the deviation between the feedback data of the service configuration plan and the expected quality improvement target exceeds the preset threshold, the sentiment profile model and demand forecasting process are iteratively updated to obtain the optimal service quality improvement plan.

Images