VR Technology-Assisted Intercultural Business Communication Innovation in College English

In the construction of the virtual scene of cross-cultural business communication, the software and hardware devices are combined so as to realize the construction of the virtual scene. The structure of the virtual scene construction device is shown in Figure 1, including virtual reality operation equipment, virtual scene construction equipment, 3D scanner and streaming box [23]. Among them, the virtual reality operation device consists of a virtual reality helmet, an interaction handle and a positioning base station, and the virtual scene construction device consists of a monitor and a computer host.

Figure 1.

Virtual scene building structure

The virtual reality operation equipment provides students with a virtual and realistic English cross-cultural business communication practical training scene, which can be used for business communication, voice control, English learning, and so on. Students wear virtual reality helmets to enter the virtual scene, use the interactive handle to operate the objects in the scene, and the positioning base station scans the helmet and handle sensor signals in real time and maps them to the virtual scene to realize positioning.

The virtual scene building device is connected to the virtual reality operating device through a streaming box. The image, audio and control signals are transmitted between the virtual scene building device and the streaming box through Display Port and USB ports, and the image, audio and control signals are transmitted between the streaming box and the virtual reality operating device through a customized cable. Students can create a virtual scene according to their teaching or training needs, and then share it on the virtual reality operating device.Students can correctly wear the headset and use the joystick to conduct cross-cultural business communication in the virtual environment.

The purpose of constructing cross-cultural business communication classroom based on VR technology is to provide a more realistic business communication environment to help students understand the key points to pay attention to in cross-cultural business communication, and to provide a reliable talent base for assisting the Belt and Road construction. In the development of the virtual scene, the prototype method is mainly used as the basis, and the virtual scene is designed according to the design process of the rapid prototype model. The rapid prototype model is shown in Figure 2. The first step is to establish a software prototype that can perform some of the functions, as a basis for listening to user feedback and evaluation, to assess whether the prototype can meet the needs of users.By modifying the prototype, the real needs of the users can be identified. After the requirements are determined, the overall development of the software is carried out to meet the real needs. The key to rapid prototyping is to build a prototype as soon as possible after determining the actual needs, so that it can be fully developed according to the actual needs.

Figure 2.

3D virtual scene development process

Based on VR technology, combined with the design ideas of rapid prototyping model, this paper designs a virtual scene applied to the cross-cultural business communication, and its specific development steps are shown in Figure 2. The production and development process of the virtual scene mainly includes the selection of production tools, the preparation and production of materials, the writing of scripts, the design of the learning interface, the design of interactive logical relationships, and the control and output of the program. Resources need to follow the development process when needed to visualize the production of virtual scenes.

In the virtual scenario of cross-cultural business communication, the intelligent recognition of students’ spoken English level can help to better understand students’ language changes in cross-cultural business communication. In this paper, the recognition of students’ spoken English speech is carried out by Hidden Markov Model (HMM), which mainly consists of two parts, represented by parameters A and B respectively [24]. Parameter A describes the stochastic process state jump occurrence probability is independent of the last round of jump position and only related to the real-time position. And parameter B describes the randomness of speech information features, corresponding to the output of the feature sequence. In the normal operation of the model, the judgment of parameter A is carried out first, and then the data is constrained by parameter B.

The construction of HMM model requires five parameters, i.e.: (4) λ={ A,B,C,M,N }

Where, N is the number of observable symbols, M is the number of states, C is the initial distribution, B is the probability distribution of observable symbols, and A is the set of state transfers.

In this study, the research object is English speech data, so the observable symbol set is English speech feature sequence, then there are: (5) A={ aij },aij=P[ qt+1=Sj|qt=St ] where 1≤i, j≤N, ∑j=1Naij=1 , ∀i, a_ij≥0.

where a_ij is the probability that a state at any moment jumps to a state at the next moment, S is the set of states for state number N, and O is the set of observable symbols M.

Assuming that O_k is the speech feature corresponding to moment t, determined by the unobservable internal state q_t = S_j, we have: (6) B={ bj(Ok),bj(Ok)=P{ Ok|qt=Sj } }1≤j≤N,1≤k≤M where B is the probability distribution of the speech feature sequence. If C is the initial state, then we have: (7) C={ Ci },Ci=P[ q1=Si ],1≤i≤N

After obtaining the results of the HMM characterizing students’ spoken English speech, the construction of the language model is carried out. The essence of language modeling is to find the joint probability of w₁,w₂,⋯,w_n in the speech data, which can be derived using Bayes’ formula: (8) P(W)=P(w1,w2,⋯,wı)=P(w1)P(w2|w1)P(w3|w1w2)⋯P(wt|w1⋯wt−1) where P(W) is the whole sentence sentence probability.

It can be seen from Eq. The joint probability can be expressed as the conditional probability of the next word after any given word, and the product of the resulting probabilities is the probability of the whole sentence.

Based on the joint probability of speech obtained from language modeling, this paper introduces a posteriori probabilistic algorithm for students’ spoken English speech evaluation [25].

In speech evaluation, P(W) is not the target of our examination, and the set of incorrect pronunciations w_i is infinitely many, which can be a sequence containing any non W, including noise mispronunciations. Obviously, there are an infinite number of cases for which this is not possible with computers. In rubrics, this is generally simplified by restricting the computation of posterior probabilities to the phoneme level, i.e., Phoneme Posterior Probability (PPP) and restricting the computational space. To wit: (9) PPP(pi)=P*(pi|O)=P(O|pi)∑j∈Replace(i)P(O|pj)=P(O|pi)P(O|pi)+∑j∈Replace(i)&j≠iP(O|pj) where p_i is the ind phoneme, and Replace(p_i) is the interphoneme list of all possible phonemes that the pronouncer could have pronounced when pronouncing phoneme p_i (i.e., the mispronunciation space containing phoneme i and it). We call Replace(p_i) the list of substitutions for phoneme p_i.

For English, the International Phonetic Alphabet (IPA) has only more than 40 phonemes, and even if the substitution list contains all the phonemes, the computational space is greatly reduced and covers most of the pronunciation error space of the pronouncer, so it not only can be computed efficiently, but also reflects the standard degree of pronunciation of the pronouncer better. The denominator part reflects the similarity with the standard pronunciation model p_i, i.e. the accuracy of pronunciation. The denominator describes the degree of similarity to phoneme p_j in the substitution list. Obviously, at p_j ≠ p_i, if the value of P(O|p_j) is smaller, i.e., the less similar to the wrong pronunciation, the value of PPP will be larger, so the denominator describes the degree of clarity of pronunciation.

GOP is also a widely used measure of good or bad pronunciation in assessment, which is a further simplification of the PPP algorithm. Note that the denominator part of the above equation, in general, if the pronunciation is wrong, its largest term will dominate in the denominator, so only the largest term can be considered. Therefore, after taking logarithms on both sides of the above equation and simplifying the denominator, the GOP formula for a particular phoneme p_i is: (10) GOP(pi)=log(P(O|pi)maxj∈ReplaceP(O|pj))=logP(O|pi)−log[ maxj∈ReplaceP(O|pj) ]

Notice that in HTK decoding, the output acoustic score is actually its logarithmic value, i.e., log P(O|p_i), and the acoustic score of speech recognition on the cut boundary, i.e., log[ maxj∈ReplaceP(O|pj) ] . Therefore, the above equation is also the acoustic score of the cut minus the acoustic score of the recognized acoustic score in the actual computation. This is only one subtraction operation, which greatly improves the computational efficiency with a small loss of performance.The GOP algorithm is expressed as: (11) GOP(pi)=logP(O|pi)−log[ maxj∈ReplaceP(O|pj) ]=Score(FA)−Score(Rec) where Score(FA) is the cut-off acoustic score, the Score(Rec) acoustic score of the substitution-list based recognition on the cut-off boundary.

After analyzing the advantages of VR technology applied to the course, based on the virtual scene of cross-cultural business communication designed in the previous article, the VR whole integration teaching is adopted to construct a VR college English multimodal teaching system as shown in Figure 3, which really improves the teaching effect of the course. The multimodal teaching system includes offline learning (i.e., classroom learning) and online learning (i.e., independent learning outside the classroom). In offline learning, the top layer is the three major advantages of VR education technology in language teaching, the middle layer is the session using VR courseware, and the bottom layer is the teaching activities related to teachers. Offline learning relies on immersive experience, virtual interaction, and in-context acquisition as the main components.

Figure 3.

VR multimodal teaching system framework

The purpose of arranging the online learning session is to promote the development of students’ independent learning ability and the formation of personalized learning methods, allow students to choose materials suitable for their needs at any time and any place, reflect the students’ subjective position in the teaching and learning process, and embody the practicality, culture and interestingness of English teaching, as well as the interactivity, achievability and ease of operation of English teaching and learning. Offline learning, on the other hand, emphasizes that students enter the VR virtual environment and experience the scenario where the topic is located as well as the natural language of the topic in the virtual environment created for cross-cultural business communication. Scripted interaction with the virtual person to achieve free communication in the scripted situation maximizes the advantages of VR technology, i.e., immersive experience, virtual interaction, and acquisition in context.

In order to verify the specific effect of the virtual environment on the teaching effect of the course based on VR technology, this paper chooses two classes of sophomore undergraduate students as the research object, and conducts a teaching experiment with the cross-cultural business communication course as an example. There were 80 students in the two classes, 40 of whom were randomly selected as the test group (TG) and the remaining 40 as the blank group (BG) after the order of students was disrupted.

The teaching experiment lasted for one semester, from September 2023 to January 2024. For the students in the test group, the teaching experiment was divided into three phases, with the first phase being the technical preparation phase, which started from the first week of the semester, and the students were trained in the technology related to the VR scene. The learning content mainly includes basic skills such as registering users, downloading and logging in the client, familiarizing with the operation interface, manipulating virtual avatars to move in the virtual world, and making friends and chatting. The second stage is the learning preparation stage, starting from the 2nd week, teachers lead students to carry out the online and offline interactive learning of the cross-cultural business communication course, collect multiple types of English teaching content from different countries, and design 3D scenes to guide students to carry out discussions and exchanges in English in the course. The third stage is mainly to guide students to carry out independent learning in the virtual scene at a later stage, selecting communication objects in different business communication scenarios, and requiring students to record the screen, recording each virtual classroom learning and oral communication activities in detail in the form of video.

For the students in the blank group, the traditional college English teaching mode is mainly adopted, in which the teacher lectures and the students listen to the lectures, without any extra related operations.

Before and after the start of the teaching experiment, the students’ performance in the cross-cultural business communication course was tested, and the independent samples t-test was used to compare the changes in the students’ performance before and after the teaching, so as to illustrate the effectiveness of the application of virtual scenarios in the cross-cultural business communication course under the technical support of VR.

In order to further analyze students’ satisfaction and experience with virtual scenarios in cross-cultural business communication courses supported by VR technology, this paper designed a questionnaire to obtain relevant data. For the students’ sense of experience, the survey was conducted mainly from the four dimensions of cognitive load, visual attention, communication presence and learning support. For the students’ satisfaction with the virtual scene teaching, the survey is mainly conducted from the three dimensions of scene design, information design and interaction design, and its main contents are shown in Table 1.

When doing the questionnaire analysis, the following values were assigned to each of the options for the objective questions:

“Very satisfied = 5 points”, “Somewhat satisfied = 4 points”, “Fair = 3 points”, “Dissatisfied = 2 points”, “Very dissatisfied = 1 point”. According to the results of the assignment, the Excel software tool is used to count the scores of each item and calculate the average score of each item. A total of 40 questionnaires were distributed and 40 were effectively recovered.

In order to compare the effectiveness of this paper’s algorithm in performing intelligent evaluation of English speech, speech phonemes are selected as the evaluation index, and multiple linear regression (A) and BP neural network (B) are additionally trained. The multivariate linear regression is fitted with a quadratic polynomial, the fitting function is obtained by solving the polynomial coefficients, and the predicted scores can be obtained by substituting the test data into the fitting function.The BP neural network, after testing, adopts the simplest 3-layer fully-connected network, and the activation function uses the Tanh function. The phoneme scoring results obtained from the models were statistically analyzed using the test set data after training the three models. Figure 4 shows the scoring accuracy of different models for different phonemes.

Figure 4.

Speech phoneme score experiment

The results of predicting the scores using the traditional polynomial regression method are very unsatisfactory, with one phoneme scored with an accuracy of even less than 5%, only one phoneme scored with an accuracy of about 90% or more, and about 23.33% of the phonemes scored with an accuracy of not reaching 60%. This shows that although the multiple linear regression model is able to judge pronunciation, it is not able to predict the score for the regression problem directly through polynomial regression. In addition, although using speech phonemes as training parameters, neural networks have a strong learning ability, and have a stronger characterization ability for non-linear relationships, so nearly 50% of the phoneme scoring accuracy is above 90%, there are also 53.33% of the phoneme scoring accuracy is less than 90%, compared with the polynomial regression method to improve the effect is significant. In this paper, after utilizing HMM for speech phoneme feature extraction, better scoring accuracy can be obtained by combining GOP value for speech phoneme scoring, only one tenth of speech phoneme scoring accuracy is lower than 92%, and the overall scoring accuracy is 94.54%, which is 38.99% and 5.44% higher than that of Algorithms A and B, respectively. Therefore, the intelligent algorithm designed in this paper can be utilized to score students’ speech communication data in the virtual scenario of cross-cultural business communication, providing a reliable reference for students to remedy the problems in cross-cultural communication.

To analyze the performance of the sentence scoring model proposed in this paper through comparative experiments, this paper additionally trained two scoring models, the BP neural network based on computational features and the LSTM model based on acoustic features, and both models use cross entropy as the loss function of the network. MSE, MAE, RMSE, and Pearson’s correlation coefficient (PCC) are selected as evaluation metrics, and the performance comparison results of speech sentence scoring are obtained as shown in Fig. 5.

Figure 5.

Performance comparison results of speech sentence score

The results show that the MSE, MAE and RMSE of the BP neural network are slightly lower than that of the LSTM network structure, and the MSE, MAE and RMSE have decreased by 10.25%, 7.13%, and 8.04%, respectively, but the Pearson’s correlation coefficient has also risen by 4.35%. Since the LSTM model is trained directly on acoustic features, the amount of data available does not guarantee the performance of the end-to-end model and the training results are prone to bias, in contrast the BP neural network is trained using computational features, which can achieve better results with less data. Compared with the BP neural network, the MSE, MAE and RMSE decreased by 50.00%, 28.84% and 26.63% respectively, and the Pearson’s correlation coefficient increased by 19.65%.The HMM model can efficiently obtain the speech signal of end-to-end cross-cultural business communication, and discover the link between the acoustic features and the ratings. Meanwhile, it combines with the GOP algorithm to better compute the correlation between speech features, so the effect is better than using acoustic features or computational features alone to train the model. Not only the MSE, MAE and RMSE are much lower than the two models, the Pearson correlation coefficients also show a strong correlation, and the model’s grading criteria for scoring are very similar to those of manual. Therefore, under the virtual scenario of cross-cultural business communication in college English, using the HMM model in combination with the GOP algorithm can realize the intelligent analysis of students’ communicative speech and provide technical support for students to find out the drawbacks in cross-cultural business communication, in order to promote the optimization of students’ utterances related to business communication.

Before using the cross-cultural business communication virtual scene to carry out teaching, the test group and the blank group of students for cross-cultural business communication course performance test, found that there is no significant difference between the two groups of students course performance, that is, it can reflect the effectiveness of the virtual scene teaching. At the end of the teaching experiment, the students in the two groups were tested for their course performance, and the data obtained were entered into SPSS software to conduct an independent sample t-test. Table 2 shows the results of the independent sample t-test of the course results, where ***,** indicate significant differences at the 1% and 5% levels, respectively.

As can be seen from the table, after the teaching of intercultural business communication virtual scenario, the total score of intercultural business communication of the test group students is 13.84±0.742, which is 4.07 points higher than that of the blank group, and the t-test result is 3.995, with a Sig (two-tailed) value of 0.003<0.01. In addition, in the two modules of Literacy and Application, the t-test results are all at the 1% level with a significant difference of 1% and 5%. Results are extremely significant at the 1% level, and the comprehension module is also significantly different at the 5% level. This shows that there is a significant difference of 1% in the performance of the students in the intercultural business communication course before and after the teaching experiment, which confirms that the virtual scene of intercultural business communication supported by VR technology can help to improve the performance of the students, help them to better master the relevant skills of intercultural business communication, and improve their intercultural communicative competence.

When teaching the virtual scene of cross-cultural business communication with VR technology support, students’ satisfaction is an important index to verify the feasibility of teaching the virtual scene. In this regard, this paper designs questionnaires from four dimensions: cognitive load, visual attention, communication presence and learning support, so as to obtain students’ satisfaction. The acquired data were entered into SPSS software to carry out the one-way difference test, and Table 3 shows the test results.

As can be seen from the table, the Sig (two-tailed) value of the test result under cognitive load is 0.943 > 0.05, so there is no significant difference between the test group and the blank group in terms of cognitive load level. Again, the mean value shows that there is only a difference of 0.38 points between the cognitive load levels of the two groups. This also indicates that there is no advantage of the test group resources over the blank group resources in reducing the cognitive load level. The possible reason for this result is that the group conducted the intercultural business communication course without practical teaching, but still used the teacher to discuss relevant knowledge. On the contrary, in the test group, in the virtual environment, the full-subtitle method of updating the relevant business communication scene resources faster may make it difficult for the learners to keep up with the reading speed, resulting in an increase in the level of psychological anxiety. In addition, the frequent switching of screen information in some of the test group’s resources takes up a large amount of attention resources, interferes with the comprehension of linguistic information, and also causes an increase in cognitive load. Business communication sense of presence showed an extremely significant difference at the 1% level between the test group and the blank group, and the score of business communication sense of presence of the test group was 7.25 points higher than that of the blank group. Constructing real business communication scenarios with virtual scenarios can create a better context for students, which allows them to realize the interactive experience between virtual scenarios and their thinking. In addition, in terms of visual attention and learning support, there is an extremely significant difference of 1% between the test group and the blank group, which further indicates that cross-cultural business communication carried out in the virtual scene can better attract students’ attention, and a variety of different types of learning tools can assist students in better realizing cross-cultural communication, and help them improve their cross-cultural communication competence.

For the satisfaction of students in the teaching process of the virtual scene of intercultural business communication supported by VR technology, this paper sets up a questionnaire from three dimensions of scene, information and interaction, and inputs the obtained student satisfaction data into the statistical software to carry out descriptive statistical analysis. Figure 6 shows the results of descriptive statistical analysis.

Figure 6.

Student satisfaction descriptive statistics

From the viewpoint of students’ satisfaction with the virtual scene teaching mode, the overall experience of students in the teaching scene is good, with an evaluation between 4.03 and 4.71 points, and the mean value of the overall satisfaction is 4.394 points, which means that the satisfaction is between relatively satisfied and very satisfied. Combined with the completion of the teaching guide, it can be found that the lack of reasonable settings of the teleportation and rotation keys is the main reason for the slightly lower evaluation of the interaction design of the teaching scene by the testers. Students in the test group generally thought that the teaching scene performed better in terms of 3D business communication scene display, teaching guidance, difficulty setting and immersion, etc. Students’ subjective evaluations included that it was like the newbie teaching part of the game, intuitive and easy to learn, and did not have too much pressure on learning.However, there are still shortcomings such as the virtual scene UI interface being not clear enough, and there is no real force feedback in the virtual space. It is pointed out that the UI interface is not clear mainly due to the fact that the students wear the Oculus headset incorrectly, which leads to pressure on the eyes, and the experimenters helped them adjust the headset according to the feedback from the testers, and the above situation has been significantly improved. Analog force feedback can not be supported only by Oculus Qusest2 hardware device, you need to use other hardware devices such as force feedback gloves. Overall, students showed high satisfaction with the VR-supported virtual scene teaching mode of college English cross-cultural business communication, which effectively improved students’ mastery of cross-cultural business communication skills and promoted their cross-cultural communicative competence.