A prediction model of speech transmission index based on

reverberation time in the non-native linguistic context

Da Yang 1 Key Laboratory of Cold Region Urban and Rural Human Settlement Environment Science and Technology, Ministry of Industry and Information Technology, School of Architecture, Harbin Institute of Technology. Qi Meng 2 Key Laboratory of Cold Region Urban and Rural Human Settlement Environment Science and Technology, Ministry of Industry and Information Technology, School of Architecture, Harbin Institute of Technology. Yue Wu 3 Key Laboratory of Cold Region Urban and Rural Human Settlement Environment Science and Technology, Ministry of Industry and Information Technology, School of Architecture, Harbin Institute of Technology. Fangfang Liu 4 Key Laboratory of Cold Region Urban and Rural Human Settlement Environment Science and Technology, Ministry of Industry and Information Technology, School of Architecture, Harbin Institute of Technology.

ABSTRACT High speech intelligibility is an essential requirement for classrooms, especially in relation to non-native students. The speech transmission index (STI) was proved as the most relevant acoustic parameter to assess speech intelligibility. In this paper, twenty- seven classrooms for non-native teaching purposes were selected for investigation. Physical acoustic measurements were conducted in these classrooms, and numerical simulation verification was determined by ODEON version 16. The relationships between STI values and RT values were fitted based on non-linear curve fitting regression models. In this paper, three primary forms of non-linear curve fitting regression models were employed for predicting curves. A logarithmic function was selected as the basic regression equation to describe the effects of RT values on STI

1 da.yang@hit.edu.cn

2 Corresponding author: mengq@hit.edu.cn

3 wuyuehit@hit.edu.cn

4 liufangfang@hit.edu.cn

worm 2022

values. The results showed that STI values increase with the decrease of RT values for all age groups. From the verified results, it was possible to propose the predictive equation that presents the best accuracy in predicting the experimental data for non- native teaching purposes. The prediction model is expected to estimate STI values by using RT values during the early design stage in a non-native linguistic context. 1. INTRODUCTION

Good acoustic quality is an essential requirement for classrooms to guarantee verbal communication between teachers and students [1]. Evidence shows that poor room acoustics, such as excessive noise and reverberation, reduce speech intelligibility in a classroom and interrupt verbal communication between teachers and students [2]. Speech intelligibility is highly correlated with classroom acoustic conditions. Previous studies revealed the effects of various acoustic descriptors (e.g., speech transmission index STI, clarity C80, definition D50, signal-to-noise ratio SNR, useful-to-detrimental energy ratio U50, etc.) on speech intelligibility [3-12]. Among these, RT and STI are the most representative acoustic parameters to assess acoustic properties in classrooms. STI is regarded as the most suitable parameter to evaluate speech intelligibility by only one index [13]. RT is well known and can be easily estimated with simple equations, especially in the early design stage for architects. In accordance with the formal analysis, it is useful to summarize prediction equations to estimate STI values in an early design stage. Previous studies proposed empirical equations to predict STI values based on reverberation times in the native linguistic context. Escobar and Morillas [14] conducted t acoustic measurements in 17 Spanish classrooms. The authors proposed a linear and a logarithmic regression equation to describe the relationships between RT and STI. Nowoswiat and Olechwska [15] investigated an experimental and numerical analysis to obtain a quick determination of the STI by knowing the RT of the room in Poland. The STI was described using a logarithmic function with the room RT. Galbrun and Kitapci [16] are investigating the accuracy of the speech transmission index as dependent on reverberation and signal-to-noise ratio. Leccese et al. compared the most significant equations proposed in the literature using the RT and STI values obtained from a measurement campaign at the School of Engineering of the University of Pisa on a sample of 11 educational rooms [17]. The acoustic properties of classrooms are essential to quickly assess their acoustic condition, especially speech intelligibility in the architecture design stage. The mentioned literature attempted to estimate STI, which is most correlated to speech intelligibility by RT values for its easy measurements in their native linguistic context. However, with the increased interaction between multilingual education, the importance of speech intelligibility in non-native language teaching draws more attention to architecture design. This paper takes classrooms in Hong Kong as case

worm 2022

studies to propose a prediction model of STI based on RT in non-native linguistic. It is due to English as the second language among local citizens being widely used in education. The results are expected to give suggestions to architects for designing non- native teaching classrooms. 2. EXPERIMENTAL METHOD

2.1 Classrooms for investigation In the current study, 27 classrooms were selected in a middle school (9 classrooms) and a university (18 classrooms) in Hong Kong as the objectivities. The selected middle school classrooms were not decorated with acoustical treatment (lime walls, cement floors, etc.). In comparison, the selected classrooms in the university were well decorated with acoustical treatment (sound absorptive panels, sound absorptive ceilings, floor isolation mat, etc.). All the classrooms were rectangular in shape, and the temperature in Hong Kong during the investigation was around 27 𝐶 ⬚

𝑜 , and the humidity was around 90%. The volumes of the 9 middle school classrooms were ranged from 151.81 to 157.84 𝑚 3 . While the volumes of selected 18 classrooms were in the university ranged from 109.03 to 1050.89 𝑚 3 . These classrooms were used for speech intelligibility investigation for non-native students [12].

2.2 Acoustic measurements The classroom impulse responses were measured by using an e-sweep signal generated from the internal DIRAC e-sweep source at the four listening positions with subjects in classrooms after the subjective questionnaire investigation. The e-sweep signal was generated from the same loudspeaker, which was placed at the same location as the subjective questionnaire tests. In order to reproduce the signal, which is similar to human’s mouth, the selected loudspeaker was Echo Speech Sound Source (B&K Type 4720). Acoustic parameters such as reverberation time ( T 30 ) and Speech transmission index (STI) were obtained during the measurement. In the meantime, the background noise level was measured by B&K 2270 sound analyzer for each listening position.

2.3 Numerical validation The numerical investigation studies were carried out in the program ODEON 16.08. Twenty-seven mentioned real rooms were modeled in the program. According to Chinese National Standard GB/T 36075.2-2018 and ISO Standard ISO 3382-2 2008 [18-19], the precision method was selected, which requires at least 12 source– microphone combinations for simulation. The main purpose of the calculations was to validate the relationship between reverberation time and speech transmission index (STI) at each investigated point. The selected simulated modeled classrooms set up in the ODEON program are given in Figure1.

worm 2022

Figure1: Models of three classrooms in the ODEON program The decorating materials of classrooms and sound absorption coefficient are given in Table 1. Table 1: Decorated materials of classrooms comparison

Sides Secondary School

University classrooms

classrooms

Floor Concrete floor Loop pile tufted carpet Sidewalls Painted concrete walls Painted concrete walls Ceiling Painted concrete walls Metal perforated plates Windows Double glazing windows Double glazing windows Door Solid wooden door Solid wooden door Front and rear walls Painted concrete walls Wooden perforated plates 3. RESULTS

worm 2022

3.1 Regression model

Several studies proposed equations in previous studies, which were obtained on an empirical basis (using the results of in situ measurement campaigns), with which it is possible to conduct a fast estimation of the STI based on RT values [15,17]. Two primary forms of regression models (linear and logarithmic function) were employed to fit classroom acoustics curves. The basic model functions of the mentioned two regression models were as follows:

𝑆𝑇𝐼= 𝑎−𝑏𝑅𝑇

𝑆𝑇𝐼= 𝑎−𝑏𝑙𝑛(𝑅𝑇)

where a, b, and c are the regression parameters generated from the fitting process. Figure 2 shows the fitting curves based on the two mentioned regression models for the description of STIs and RT values (500-1000Hz) in selected classrooms. The regression parameters and statistical characteristics are given in Table 2.

aw y &

worm 2022

Figure 2: Comparison results of two regression models in classrooms Table 2: Regression parameters of the two regression models and statistical characteri stics

𝒂 𝒃 Adj. 𝑹 𝟐

Linear 0.900 0.291 0.837

Logarithmic 0.594 0.198 0.886

The regression fitting curves were plotted with data collected from university classrooms in Fig 2. Besides, the regression parameters were given in Table 2. The adjusted R-square of logarithmic fitting curve (red line in Fig 2) is higher than the linear curve (blue line in Fig 2). Therefore, in the current study, the logarithmic regression fitting model is employed to describe the relationships between STIs and RTs. 3.2 Simulation results

The simulation results of STI values were plotted with the data from the measurement in Figure 3. A good agreement was found between the results collected in real and virtual classrooms. Compared to the measured values, the simulated STI values were very similar, generally within 1 JND (0.03) in most classrooms. The results revealed that the STI could be predicted accurately by acoustic simulation when there is a good

054 inear Regression Logarithmic Regression) 04 03 RT/s

agreement between the virtual models and the real rooms and that different RT may exert less significant influence on the simulated STI.

Figure 3: Measured and virtual STI in 27 classrooms 4. CONCLUSIONS

This paper employed three primary forms of non-linear curve fitting regression models to predict curves. A logarithmic function was selected as the basic regression equation to describe the effects of RT values on STI values. The results showed that STI values increase with the decrease of RT values for all age groups. The verified results made it possible to propose the predictive equation that presents the best accuracy in predicting the experimental data for non-native teaching purposes. 5. ACKNOWLEDGEMENTS

worm 2022

This work was supported by the National Natural Science Foundation of China (NSFC) [grant numbers 51878210, 51678180, and 51608147], the Open Projects Fund of Key Laboratory of Ecology and Energy-saving Study of Dense Habitat (Tongji University), Ministry of Education (2020030103), and Natural Science Foundation of Heilongjiang Province [YQ2019E022]. 6. REFERENCES

[1] C.M. Mak and Y.P. Lui. The effect of sound on office productivity. Build Serv Eng Res T 2012; 33; 339-45. [2] H.M. Wong, C.M. Mak and Y.F. Xu. A four-part setting on examining the anxiety- provoking capacity of the sound of dental equipment. Noise Health 2011; 13; 385-91. [3] D. Yang, C.M. Mak, “An investigation of speech intelligibility for second language students in classrooms,” Appl. Acoust 2018; 134; 54–59. [3] Bradley JS. Speech intelligibility studies in classrooms. J Acoust Soc Am 1986; 80: 846-54. [4] Yang W, Bradley JS. Effects of room acoustics on the intelligibility of speech in classrooms for young children. J Acoust Soc Am 2009; 125: 922-33.

[5] Bradley JS and Sato H. The intelligibility of speech in elementary school classrooms. J Acoust Soc Am 2008; 123; 2078-86. [6] Sato H and Bradley JS. Evaluation of acoustical conditions for speech communication in working elementary school classrooms. J Acoust Soc Am 2008; 123; 2064-77. [7] Astolfi A, Bottalico P and Barbato G. Subjective and objective speech intelligibility investigations in primary school classrooms. J Acoust Soc Am 2012; 131; 247-57. [8] Choi Y. Effect of occupancy on acoustical conditions in university classrooms. Appl Acoust 2016; 114; 36-43. [9] Peng JX, Yan NJ and Wang D. Chinese speech intelligibility and its relationship with the speech transmission index for children in elementary school classrooms. J Acoust Soc Am 2015; 137; 85-93. [10] Peng JX. Chinese speech intelligibility at different speech sound pressure levels and signal-to-noise ratios in simulated classrooms. Appl Acoust 2010; 71; 386-90. [11] Zhu P, Mo F, Kang J. Relationship between Chinese speech intelligibility and speech transmission index under reproduced general room conditions. Acta Acust United with Acust, 2014; 100; 880-887. [12] D. Yang, C.M. Mak, Effects of acoustical descriptors on speech intelligibility in Hong Kong classrooms, Appl. Acoust. 171 (2021) 107678. [13] H.J.M. Steeneken, T. Houtgast, Mutual dependence of the octave-band weights in predicting speech intelligibility, Speech Commun, 28 (1999), 109-123. [14] V. Gómez Escobar, J.M. Barrigón Morillas, Analysis of intelligibility and reverberation time recommendations in educational rooms, Appl Acoust, 96 (2015), 1- 10. [15] A. Nowoświat, M. Olechowska, Fast estimation of speech transmission index using the reverberation time, Appl Acoust, 102 (2016), 55-61. [16] L. Galbrun, K. Kitapci, Accuracy of speech transmission index predictions based on the reverberation time and signal-to-noise ratio, Appl Acoust, 81 (2014), 1-14. [17] F. Leccese, M. Rocca, G. Salvadori, Fast estimation of Speech Transmission Index using the Reverberation Time: Comparison between predictive equations for educational rooms of different sizes, Appl Acoust, 140 (2018), 143-149. [18] GB/T 36075.2-2018: Acoustics-Measurement of room acoustic parameters-Part 2: Reverberation time in ordinary rooms. International Organization for Standardization 1991. [19] ISO3382-2:2008: International standard ISO 3382-2:2008 Acoustics- Measurement of room acoustic parameters-Part 2: Reverberation time in ordinary rooms.

worm 2022