Acoustic characteristics of the clothes used for a wearable recording device

Use of wearable acoustic recording technology has been steadily increasing in research, clinical, educational, and home settings.^1–8 Modern technology is capable of capturing day-long audio samples in naturalistic, family environments from the auditory perspective of a user wearing the recording device. Critically, the naturalistic recording is analyzed (typically offline) using automatic speech processing (ASP) and/or automatic speech recognition (ASR) software to generate variables of interest to the user. One such system used to analyze children's speech and language characteristics is produced by the LENA Research Foundation (LRF; Boulder, CO). It consists of a wearable acoustic recording device and custom ASP software. The recorder is designed to be worn in a custom-fit pocket on the front of a shirt worn by a child. The system records and stores the full-day of audio for further analysis or playback. The goal of the system is to provide objective estimates of variables that are known to influence speech, language, and social development. Studies using this device have examined total word and conversational interaction exposures with adults,^1,2,8 electronic media exposure,^2,9 and the system has been demonstrated to be useful in the assessment of children with Autism spectrum disorder,^3,4 language delay,³ hearing loss,^5,9 and with preterm infants.⁷ 

Although wearable recording devices have been used extensively in a variety of settings, there are gaps in the scientific knowledge of the practical characteristics and interpretability of the audio recordings due to the way the recordings were obtained. In particular, acoustic transfer characteristics of the clothing used to collect the raw recordings is not known. Acoustic transmission loss (TL) has been examined generally for fabric positioned in front of a noise source^10–12 and in front of a voice,^13,14 but not for wearable recording systems such as the LENA. In general, denser material composition (i.e., weave density) and indirect microphone placement increase amplitude loss at frequencies greater than 1 kHz. Woven cotton has been shown not to influence TL up to 2.5 kHz, while other materials such as paper surgical masks have been shown to have increased TL above about 250 Hz.¹⁴ Other factors directly contributing to TL include the relationship between the noise or voice source and the polar response pattern of the microphone and signal interference from external noise or environmental sources.

There are no known reports of TL of a wearable recording device in situ. A better understanding of the acoustic properties of the transmission medium will increase our ability to effectively use and manage the signals collected from these devices. This report describes acoustic transfer characteristics of a bare-microphone compared with two types of clothing used with a specific, commonly-used recording system.

The LENA is a commercially available, body-worn audio recording system. The system includes a recorder weighing about 70 g and measuring about 1 × 5 × 8 cm designed to be a self-contained acoustic recording and storage device worn by a child for a whole day. The system requires peripheral clothing with a pocket to render the recorder wearable. One of the primary goals of securing the recorder inside the chest pocket is to fix the recorder's integrated microphone at a specific distance (10–15 cm) from the wearer's mouth. The LRF-designed standard clothing (SC) features a custom pocket manufactured into the product. Custom clothing (CC) reproductions were fabricated by the author based on the design patterns provided by the LRF. Both shirt types were comprised of plain cotton or cotton-blended fabric featuring a chest pocket internally backed with padded material. Three edges of the pockets were permanently sewn, and one edge had several snaps to secure the recorder in place. The LENA system used in the present experiments is designed so that the recorder remains snugly in place within the pocket, fixing its position inside the pocket and with respect to the wearer's body and mouth. The pockets were designed to be unobtrusive for the child wearing the shirt. Individual SC and CC shirts differed in color, size, and/or wear-and-tear, and CC shirts differed additionally in blend of fabric materials and/or stitching patterns. Overall, the selected shirts were soft and unadorned (e.g., without rhinestones or buttons) without extensive screen-printing or logos.

A total of 126 stimuli were constructed at each of nine frequencies [modulated (5%) pure-tones at 64, 125, 250, 500, 1000, 2000, 4000, 8000, and 16 000 Hz] and 14 amplitude levels [5 dB steps from 25–90 dB sound pressure level (SPL)]. Stimuli were presented from a near-field monitor loudspeaker (model 8030A, Genelec, Iisalmi, Finland). A calibrated measurement microphone (model XL2, NTi Audio, Schaan, Liechtenstein) was used to collect amplitude measurements in a bare baseline condition and inside the pocket of the clothing. No person occupied the room during testing. All measurements were collected inside a calibrated IAC sound-attenuated booth in the Speech and Language Lab at Washington State University.

For all recordings, the microphone was mounted on a boom stand and fitted with a foam windscreen to ensure consistent placement and no direct contact for the conditions inside the pocket. For all data collection, the head of the microphone was positioned about one-half inch from an acoustically-treated easel in the testing booth. The angle of the microphone and positions of all items in the recording room remained fixed throughout testing.

In the baseline condition, the microphone was bare with no object between it and the acoustic source. The purpose of the baseline condition was to establish the experimental setup so that the presentation levels of each stimulus were controlled. In the within pocket condition, each shirt was tacked to the easel simulating normal wear, and the microphone was placed inside the front pocket. The position of the measurement microphone into the pocket simulated the microphone position of the recorder in real world applications. Using this procedure, baseline response values were directly comparable to response values collected in the test conditions with the microphone inside the shirt pockets, thus allowing for direct assessment of the influence, if any, of the shirt fabric on the transmission of the acoustic signal.

Ten representative examples of each SC and CC clothing type were tested. Shirts were in regular-use rotation and were freshly laundered. All 126 stimuli were presented to each of 20 shirts. In total, 2520 amplitude observations were recorded. Within group comparisons used a one-sample t-test, and between-group comparisons used a independent-sample t-test. All statistical tests were two-tailed.

There were no within-group differences for either the SC or CC clothing types for any of the 126 stimuli [t(9) < 0.5, p > 0.2] or for observations pooled within each presentation level compared across each frequency [t(9) < 0.5, p > 0.3]. These results indicate similar response characteristics among the exemplars within each shirt type. There were no between-group differences for SC-CC, SC-baseline, or CC-baseline comparisons at individual stimuli or pooled by frequency [t(9) < 0.5, p > 0.2], indicating similar response characteristics between exemplars and the known baseline signal characteristics. Mean response amplitude values at each frequency and stimulus presentation value are displayed in Fig. 1, showing amplitude on the ordinate and log-scaled frequency on the abscissa. The stimuli reference values are shown by the dashed line, responses in the SC condition in circle markers, and responses in the CC condition in diamond markers. Table 1 shows response characteristics for each shirt type and the bare-microphone reference value. The difference between SC and CC shirt types at each of the 14 stimulus frequencies was not statistically significant [t(13) = 1.13, p > 0.2].

This report compares the amplitude response characteristics of a microphone placed in a bare condition and inside the fabric pockets of two shirt types commonly used for wearable audio recorders. Test stimuli comprised a range of amplitudes and frequencies important for speech production and perception. Reliable differences were not observed at any test frequency or level, within exemplars of either type of clothing, or among the shirt types and the bare microphone conditions. These findings suggest that the acoustic influence of the shirt materials tested here may have a negligible impact on recordings collected with wearable technology.

The results reported here are evidence that the acoustic recordings may be robust within the limits of the clothing design, and users may benefit from expanded options or applications for use with a recording device inside CC or with a bare-microphone. The possibility of expanding clothing options for children and adults with special needs, families with social, religious, or clinical restrictions, or in educational or clinical settings may be warranted. In addition, this result may reduce the overall resource expense of both time and financial burden to researchers, clinicians, educators, and parents if special clothing or bare-microphone conditions constitute reliable, less expensive alternatives.

A better understanding of the acoustic signal collected from a wearable recorder is useful to a variety of applications including ASP, ASR, and signal processing used in applications such as hearing aids and telephony. For these applied problems, the response characteristics of the system are important because acoustic details of the recording are (typically) the only input to the system. Current wearable recording systems such as the LENA technology have been shown to be useful for a variety of research questions,^{2–5,7–9,15} and there has been increasing interest in assessing details of the acoustic recordings.^3,6,15,16 An understanding of the characteristics of the collected signal may influence the nature of future inquiry using this or a similar application.

This work has several limitations. First, testing was conducted in the controlled laboratory environment with highly controlled stimuli and simulated-use environments. Ecological validity and a greater diversity of testing materials is needed, including comparing more varied fabrics and testing with a variety of users under various real-world situations. Second, statistical tests here showed non-significant effects which are difficult to directly interpret. Third, although statistical differences among acoustic response characteristics were not found, specific ASP or ASR algorithms may be sensitive to the acoustic noise or TL.

With increasing interest in wearable, passive technology in a wide variety of clinical, educational, and home settings, a better understanding of response characteristics is crucial to continued and extended applications. The work presented here will benefit researchers, clinicians, educators, and caregivers using wearable acoustic recording devices in a variety of contexts and may allow researchers to expand investigations in new directions in both theoretical and applied contexts.