Effect of speech intelligibility on task performance - an experimental laboratory study

The aim of this study was to examine the effects of speech varying in intelligibility on cognitive performance and subjective perceptions of sound environment disturbance. 37 subjects performed a series of tasks in three conditions in which speech transmission indexes were 0.10, 0.35 and 0.65. These correspond to cellular office, well-designed open office and unsatisfactory open office, respectively. The experiment was conducted in an office laboratory in which the direction of the speech source varied. The sound environments were presented at 48 dBA. Performance deteriorated in condition 0.65 compared to the other two conditions in a serial recall task (p < .05) and in a complex working memory task (p < .001). Proofreading performance did not differ in different conditions but the task was experienced as easier in the condition with lowest STI value than in the other two (p < .05). Questionnaire measures showed consistent, statistically significant differences between all three situations: the higher the STI value was, the more it was experienced to disturb performance and draw attention away from the task (p < .001). Similarly, the ease of habituation and the pleasantness of sound were rated higher as the STI value declined (p < .001). Self-rated work efficiency declined with ascending STI values (p < .05). Continuous background noise was not experienced to disturb performance in any condition. This supports the use of continuous masking sound in minimizing speech intelligibility, i.e. increasing speech privacy in open offices. INTRODUCTION Office surveys have shown that speech is experienced as the most distracting noise source in offices (e.g., Haapakangas et al. 2008). The problem is particularly great in open offices. Earlier laboratory experiments have shown that the disruptive effects of speech are not produced by the level of speech but speech intelligibility (Colle 1980; Ellermeier & Hellbrück 1998; Venetjoki et al. 2006; Schlittmeier et al. 2008). According to Hongisto et al. (2007), speech intelligibility in open offices varies greatly but it can be lowered by proper acoustical design. Research is needed on the effect of speech intelligibility on task performance in order to encourage investments in acoustic improvements. However, there are only a few published studies that have studied the effect of speech on cognitive performance with varying levels of speech intelligibility. Speech intelligibility can be evaluated with Speech Transmission Index (STI). For example, STI value 0.50 expresses roughly that 50 percent of syllables are correctly heard. STI can be easily determined between office workstations using acoustical measurements. The first experiment that used STI as a descriptor of irrelevant speech was published by Venetjoki et al. (2006). It was found that proofreading performance deteriorated significantly in STI 0.80 compared to STI 0.00. Later, similar experiments have been conducted by Schlittmeier et al. (2008). Performance: 9th International Congress on Noise as a Public Health Problem (ICBEN) 2008, Foxwoods, CT The aim of this experiment was to show how the level of speech intelligibility affects cognitive performance. Speech was expected to deteriorate performance more in higher levels of speech intelligibility. The study focused on the STI range from 0.100.65 because the acoustic conditions of offices are typically in this area. The study also aimed to validate the model of Hongisto et al. (2008) that predicts the deterioration of performance as a function of STI. METHODS Subjects and test setup A repeated measures design with three speech conditions was used. Altogether 37 students took part in the experiment in December 2007. The three speech conditions were STI 0.10, STI 0.35 and STI 0.65. A group of four participants was tested at a time. Experiments were conducted between 8.15 a.m. and 12.30 p.m. The experiment included a practice session and three test sessions lasting for about 55 minutes each. Each situation was followed by a 5-minute break. The order of speech conditions was counterbalanced across subjects, as was the presentation of different speech samples and test versions in different speech conditions. Subjects performed five tasks: a serial recall task, a complex working memory task, a proofreading task, a visual short-term memory task and a reading comprehension tasks. The latter two were mainly included as filler tasks to increase the length of test sessions and were being piloted for future experiments. The serial recall task was conducted following classic procedures in which digits from 1 to 9 are presented on a computer screen in random order. Subjects' task is to recall the digits in the same order. The number of digits recalled in correct serial positions is measured. The complex working memory task was modified from the operation span task developed by Turner and Engle (1989). In the task, subjects had to state whether simple arithmetic calculations, presented on a computer screen, were true or false. Each calculation was followed by a presentation of a word that the subject had to memorize. At the end of each equation-word pair list, subjects were asked to recall the presented words in the same order. The number of words to be remembered increased from 3 to 8. The total number of correctly recalled words was measured. The proofreading task was the same as in our earlier study (Venetjoki et al. 2006). It was a pen-and-paper task in which subjects looked for mistakes in the text. Half of the errors were spelling errors whereas half required semantic processing. Each speech condition was followed by a questionnaire that assessed subjective perceptions of the sound environment. A 5-point Likert scale was used in most questions. Disturbance of other environmental factors was included in the questionnaire that was presented after the last sound condition. Background information was gathered with a separate questionnaire before practice session. Speech conditions The investigations were carried out in an office laboratory (Figure 1). Three speech conditions were used: • STI=0.10 corresponds to a private office room with the door closed. Speech intelligibility is extremely low. Performance: 9th International Congress on Noise as a Public Health Problem (ICBEN) 2008, Foxwoods, CT • STI=0.35 corresponds to a very good open office or a private office room with opened doors. Speech intelligibility is reasonably low. • STI=0.65 corresponds to a typical open office without adequate acoustic design. Speech intelligibility is nearly perfect. STI values were obtained by changing the relative sound pressure levels of speech and masking as shown in Figure 2. However, the total sound pressure level was constant in all speech conditions, LA,eq=48 dB. Special effort was made to create an office-like environment. All factors of indoor environment were monitored during all experiments. Typical values are given in Figure 1.