Object knowledge is derived from the interaction of multiple senses, yet our understanding about the variables that influence human object processing stems predominantly from studies of visual words and pictures. Far less is known about the variables that influence the processing of environmental sounds, and to date there have been only a small number of studies investigating the variables that may influence the successful recognition and identification of environmental sounds that can be used as normative data. In Table 1 we have provided a summary of these studies and the variables that were measured. Although multiple different variables have been considered, only the study by Marcell, Borella, Greene, Kerr, and Rogers [14] provided a comprehensive set of measurements of naturally occurring sounds across a range of conceptual categories. However, because these sounds were of variable temporal duration, they are less suitable for use in well-controlled paradigms in the cognitive or neuroimaging domains.

When designing this study, we considered the following variables to be of primary value across multiple domains for empirical research into object-level natural sound processing: Response latencies, identification, categorization, familiarity, confidence, representativeness, affective ratings and imageability. In order to maximize the number of respondents, we utilized both an on-line questionnaire format for qualitative responses (Study 1), and a laboratory-based study for both qualitative and response times variables (Study 2). This enabled the efficient collection of a large number of responses across multiple different variables. We therefore obtained the following data:

Given the limited number of studies currently available that provide comprehensive normative data for environmental sounds, our aim was to provide data on a large set of environmental sound stimuli designed for researchers investigating auditory processing across cognitive (neuro)psychological, psychophysiological and cognitive neuroscience domains. We obtained response time and identification data for a set of 110 common objects, natural events, human actions and animals, along with a comprehensive set of ratings of object categorization, familiarity, pleasantness, arousal and imageability. All environmental sounds have been made freely available online, along with a detailed table of the response data to assist researchers in the selection of the most suitable auditory objects for their own parameters of interest.

The study complied with the Australian National Statement on Ethical Conduct in Human Research and was approved by The University of Queensland Medical Research Ethics Committee (Project #2007001910), including the process of providing consent online. In Study 1 participants provided their informed consent by clicking a checkbox prior to proceeding with the sound questionnaire. For Study 2, participants provided their written informed consent to the researcher prior to commencing.

Sounds had been collected as part of an ongoing database for stimuli used in fMRI studies by the first author [e.g., 36], primarily downloaded from the website www.sounddogs.com and www.freesounds.com. From this database, the same 110 natural sounds were used in Studies 1 and 2, with equal numbers of living and manmade items from 9 conceptual categories. All sounds were normalized to 1sec duration (16-bit 44,100Hz) using Audacity 1.2.5 (http://audacity.sourceforge.net) and presented in mp3 format in Study 1 (due to restrictions with the online software used) and wav format in Study 2. All files have been made available (in wav format) for download at http://www.imaging.org.au/Nessti. Also provided online are measures of concept frequency based on the Hyperspace Analogue to Language frequency norms (HAL) [39] for all available concepts, as well as the number of phonemes, syllables, and the Harmonics-to-Noise (HNR) Ratio.

There were 162 participants (85 male, 77 female), mean age 27.7 years (SD: 8.90), mean years of education 16 (SD: 3.27). See Table 2 for a summary of the participant demographics. One group of questionnaire respondents (73) were undergraduate students from the University of Queensland School of Health and Rehabilitation Sciences who were given course credit for participation. The second group of respondents (89) was recruited through on-line advertising to staff and students across the University of Queensland. These participants were encouraged to complete either one or both questionnaires by being entered into a draw to win a Macintosh iPad (both questionnaires completed meant 2 entries into the draw). In total we obtained 123 datasets for questionnaire 1 and 123 datasets for questionnaire 2, with 84 people completing both questionnaires. Although 5 participants reported a history of hearing impairment, their accuracy did not significantly differ from the remaining participants (p = .68).

An online questionnaire was developed and administered using the platform Questchain (www.questchain.com). Participants were given a link to the Internet page on which either part 1 or part 2 of the questionnaire could be accessed. The format of the self-paced questionnaire was as follows: Page 1 consisted of an information and consent form, which required the participant to give their informed consent to take part before being able to move to the next page. Page 2 described the study and gave the participant detailed instructions on what was required of them. At this point, the respondent was able to play a test sound to ensure the volume was set appropriately. Page 3 requested demographic and other personal information. The actual normative questionnaire started on the page 4, with one page per sound. The participant would first play the sound until they decided on their response, and then answer a series of questions. All questions had to be answered in order to advance to the next page. A screenshot of the page used to probe the participants’ knowledge regarding environmental sounds has been included in Supporting Information (Questionnaire S1).

The normative data collected were as follows:

In Study 1, we received a total of 13515 correct responses out of a maximum of 13530. The proportion for all sounds correctly identified is listed in Table S2 and the frequency distribution is shown in Figure 1A. Five items were identified correctly by all participants (cat, horse, laugh, phone, sneeze), and 19 items were identified with a success rate of over 95%. The percentage of the sounds identified accurately across all participants was 65.31%. This accuracy was not dependent on whether the items were from a living (70.5% correct) or manmade (60.1% correct) category (t[108] = −1.87, p = .065). The scoring criteria for a correct name included multiple responses (i.e., acceptable synonyms, see Table S1), which meant that it was possible for the modal name to differ from the concept name; for example the modal name for grasshopper was insect. The modal name for 20% of sounds did not correspond to the expected label given to each sound by the experimenters, so for those cases we have provided the alternative modal name in parentheses in Table S2. It is also of note that some modal names were verbs rather than nouns (e.g., sweeping was the most common response for broom). The corresponding percentage of responses for each modal name is also given. The H-value was also calculated as a measure of name agreement on correct responses for each item and ranged from 0–4.90 (Table S2).

Level of identification was highly correlated with the familiarity rating for that item (r[110] = −.796, p<.0005), where the more likely a response was correct, the more familiar an item was rated to be. The measure of representativeness of a sound correlated with identification rate (r[110] = −.795, p<.0005) as did the rating of pleasantness (r[110] = −.335, p<.0005). There was no correlation between identification and arousal (r[110] = −0.143, p = .068) or frequency (r[110] = 0.046, p = .326). Figure 2A–C shows scatterplots for the significant correlations.

On a scale of 1 - 6, where 1 = most familiar, mean familiarity rating ranged from 1.09–3.86 (M:2.33, SD:0.73). Sneeze was rated the most familiar and skiing was least familiar. There was a significant difference between the familiarity of living and manmade items (t[108] = 3.04, p = 0.003) with living things rated as more familiar than manmade sounds (see Table 3). Familiarity was highly correlated with representativeness (r[110] = .981, p<.0005). The high r-value suggests that these two ratings are indexing the same cognitive measure (Figure 2D). Familiarity also correlated significantly with pleasantness (r[110] = .458, p<.0005), which appeared to reflect the fact that ratings tended to be narrowly distributed about the mean on both scales. There was no clear relationship between familiarity and arousal (r[110] = .149, p = .06). Items are listed in Table S3 and the frequency distribution for familiarity ratings is shown in Figure 1C.

Representativeness was measured on a scale of 1–5 where 1 = highly representative (M:2.16, SD:0.64), and ranged from a horse which was judged to be a highly representative sound with a score of 1.08 to the sound of someone skiing given 3.45. Living and manmade items significantly differed on their ratings (t[108] = 3.18, p = .002), with living things judged to be more representative than manmade sounds (see Table 3). Ratings are listed in Table S3 and the frequency distribution is shown in Figure 1D. Unexpectedly, the rating for how representative a sound was of its source correlated with the rating for arousal (r[110] = .203, p = .017). The more representative an item was, the higher the rating of calmness (see Figure 2G for scatterplot).

Ratings for pleasure were on a scale of 1 (most pleasant) - 9 (most unpleasant) with a range of 2.32–6.95 (M:4.74, SD:0.89). The sound rated most pleasant was laugh and most unpleasant was machine gun. Pleasure negatively correlated with arousal (r[110] = −.401, p<.0005). Arousal rating ranged from 3.21 (fire alarm) –6.76 (yawn) where 1 = most excited and 9 = most calm. (M:5.22, SD:0.69). There was no significant difference between living and manmade items on either pleasantness (t[108] = 1.04, p = 0.3) or arousal (t[108] = −0.81, p = .42) (see Table 3). Affective ratings are listed in Table S3 and the frequency distribution can be seen in Figures 1E–F. A scatterplot of the relationship between pleasantness and arousal is provided in Figure 2H.

Participants classified each sound as belonging to one of 9 categories. The modal category response is provided against each sound in Table S2, as well as the proportion of respondents who selected that category. Only 9 sounds were categorized differently from those categories agreed by the raters, of which 4 were from the Recreational category and were alternatively categorized as being from the Household category (pinball machine, basketball, book and skiing). These 4 items also had low identification rates (30.89%, 29.27%, 13.01%, 1.63% respectively). For the remaining 5 sounds, whistle and fire truck had high identification rates (96.75% and 80.49% respectively) and were classified by the experimenters as Alarm and Transport whereas respondents classified them as Recreational and Alarm. The remaining 3 (cicada, rockfall, washing machine) had correspondingly low rates of correct identification (30.89%, 12.2%, 3.25% respectively). Table S4 summarizes the responses by category.

There were 58 participants in total (22 male, 36 female), mean age 31.72 years (SD: 10.9), mean years of education 16.82 (SD: 3.33). Participants were recruited through advertising to staff and students across the University of Queensland. No volunteers who had participated in Study 1 took part in Study 2. 5 participants were excluded from the analysis due to technical difficulties with the recording equipment. Although 2 participants reported a history of hearing impairment, their accuracy did not significantly differ from the remaining participants (p = .72). See Table 2 for a summary of demographics.

This computer-based experiment was programmed in Cogent (www.ucl.ac.uk/vislab) and implemented in Matlab (Mathworks, Sherborne, MA, USA). Before beginning the experiment, participants were told that they would be hearing a large number of environmental sounds and their task was to verbally identify the object, animal or action depicted by the sound as quickly and clearly as possible. They would then be asked to rate how confident they were that their response was correct, and how imageable the sound was. Participants were asked not to describe or mimic the sound and were encouraged to use a single word response only. They were also asked not to use an article before the noun e.g. responding “dog” not “a dog”, or “ball” not “the ball”. A microphone was used for recording responses and reaction times, and sounds were played through headphones at each participant’s preferred volume. A brief practice session was run (using a different set of 6 sounds) and was repeated when necessary, with feedback from the experimenter, to correct any deviation from these instructions. The sound threshold for recording was adjusted for each individual during the practice trials in order to maximise recording sensitivity to their voice and minimise sensitivity to other extraneous noise. The time allowed to make a response was five seconds from the onset of the sound. After the 5 seconds, participants were then cued to provide their rating of 1) confidence in their response, and 2) the imageability of the sound. The normative data collected were as follows:

In Study 2, there were a total of 5272 correct responses, from a maximum of 5830. Compared to Study 1 only two sounds were identified 100% correctly (dog, horse) and 13 sounds were identified with a success rate of over 95%. Across all participants, the mean percentage of sounds accurately identified was 56.91%. The proportion for all sounds correctly identified is listed in Table S5 and the frequency distribution is shown in Figure 3A. The mean identification percentage for manmade sounds was 51.68%, compared to 62.14% for living sounds, but this difference was not significant (t[108] = −1.83, p = .07). Identification highly correlated with mean reaction times (r[110] = −.67, p<.0005), where the faster a response, the more likely it was identified correctly. Identification correlated with ratings of confidence (r[110] = .75, p<.0005) and imageability (r[110] = .75, p<.0005). See Figures 4A–C for scatterplots of correlations.

Mean reaction times (RT) ranged from 1264–3476 msec (M: 2222, SD:400). Sneeze was identified the fastest (1264 msec), and rock fall required the longest time to be identified (3476 msec). RT also correlated with confidence (r[110] = −.75, p<.0005) and imageability (r[110] = −.75, p<.0005). See Figures 4D–E. RTs for manmade and living sounds were not significantly different (t[104] = 1.37, p = .17) (see Table 3). The frequency distribution for reaction time is shown in Figure 4B.

Ratings for confidence were on a scale of 1 (low confidence) - 7 (high confidence) with a range of 2.87–6.83 (M: 5.26, SD: 1.05). The sound identified with most confidence was cat and with least confidence was book. The mean confidence ratings for manmade sounds versus living sounds was significantly different (t[108] = −3.34, p = .001), with participants more confident at identifying living sounds. Ratings for imageability were on a scale of 1 (no imagery) –9 (high imagery). Mean imageability of the sounds was 6.36 (SD: 1.38), and the concepts reflected those for confidence: cat was rated the highest for imageability (8.51) and sound of a book the least imageable (3.06). This was reflected in the high correlation between confidence and imageability ratings (r[110] = .98, p<.0005), see Figure 4F. The mean imageability ratings for manmade versus living sounds were significantly different (t[108] = −3.09, p = .003), with living things rated as more highly imageable than manmade sounds. Living versus manmade ratings are provided in Table 3 and the frequency distributions are shown in Figures 3C–D.

An independent t-test was computed to evaluate if identification accuracy differed between the two types of studies. This showed a significant difference between studies, where participants were more accurate on the questionnaire in Study 1 compared to the lab-based design for Study 2 (t[218] = −2.08, p = .04). This result should be viewed with a note of caution however: although the sounds were the same, there was a difference in the number of times the sound could be heard between the two studies, as well as the type of identification question that was asked.

The scoring criteria for a correct name included multiple responses (i.e., acceptable synonyms, see Table S1 & Table S5, for Study 1 and Study 2, respectively). This meant that it was possible for the modal name to differ from the target name, for example the modal name for grasshopper was insect. In Study 1, 20% and in Study 2, 31%, of the modal names for sounds did not correspond to the expected label given to each sound by the experimenters, so for those cases we have provided the alternative modal name in parentheses in Table S2 & Table S5. The corresponding percentage of responses for each modal name is also provided.

There are many aspects of sound processing that have not been the focus of the present study, but would be fruitful in enhancing our understanding of the different ways in which sounds are perceived, encoded and processed. These include the many low-level acoustic variables that are known to modulate behavioral perception as well as neuronal responses (e.g., [45], [46], [49]). Collecting the same data from older adults may also provide useful neuropsychological control data for clinical populations where naming and recognition deficits occur, which may also necessitate investigating how longer versions of these environmental sounds influence identification and ratings, as suggested by the effect for length reported by Marcell et al. [14]. The different ways in which objects are categorized would also provide valuable information, and this would be particularly relevant cross-culturally.

In summary, our aim was to provide a comprehensive set of ratings for natural living and manmade sounds that are also equated by a length suitable for use in a wide range of behavioral and neuroimaging settings. We have made these sounds freely available, and have provided a table of responses in order that the researcher has as much assistance in selection of appropriate and relevant stimuli as possible. Although there are many dimensions of environmental sound processing that still require attention, we hope that the data collected for these 110 unique sounds provides a level of description that will assist the researcher investigating environmental sound processing at multiple levels and in multiple research domains.