Audiometric Validation of a Smart Watch Decibel Meter

Abstract

This diagnostic study assesses the performance of the Apple Watch Noise application in comparison with a class 1 sound level meter.

The Noise application on the Apple Watch measures the intensity of ambient sounds and is available on the Apple Watch SE and Apple Watch Series 4 and later.¹ One prior study using the Noise application in a critical-care setting recommended further studies to explore the role of wearable devices in measuring noise levels.² In addition, the promise of consumer wearables to help inform clinical decision-making has been emphasized as an important direction of future research in otolaryngology.³ To our knowledge, no studies have formally assessed the performance of the Noise application. Therefore, the aim of this diagnostic study was to assess the performance of the Apple Watch Noise application in comparison with a class 1 sound level meter.⁴

Methods

Both an Apple Watch Series 6 (Apple) and a Type 2250 Light handheld analyzer with the ZC-0032 microphone (Brüel & Kjær) were placed 50 cm from a speaker inside of an audiometric booth, and pure tones were played in half octaves from 125 to 8000 Hz in an amplitude range from 0.0001 to 1. A total of 21 amplitude measurements were performed within each half-octave frequency to allow sufficient statistical power in device comparison. The BZ-5503 Measurement Partner Suite software (Brüel & Kjær) was used with a D-75A amplifier (Crown) to produce the pure tones. Amplitude was provided as an input variable within this software. The measurement data for each amplitude across different frequencies were averaged for each device and compared with paired sample t test. SPSS software, version 17.0 (IBM) was used for statistical analysis, with P = .002 considered statistically significant after adjustment. This study was exempt from institutional review board approval because no human participants were involved.

Results

Device measurements demonstrated a range of sound from a minimum of 22.8 dB sound pressure level (A-weighted) to a maximum of 117.3 dB sound pressure level (A-weighted). Measurements between the 2 devices only displayed statistically significant differences between the 2 device outputs at 4 and 8 kHz (Table). In addition, correlation of dB measurements between the 2 devices at all frequencies was 0.961 or greater.

Table. Comparison of Measured Data by the 2 Devices for 21 Amplitude Measurements Across Different Frequencies.

Frequency, Hz	Sample size	Average measurement, mean (SD), dB		P valuea	95% CI	Effect size	Relative size
		Watch	Meter
125	21	64.05 (21.10)	66.52 (22.70)	.04	−4.86 to −0.08	0.47	Small
250	21	71.62 (22.20)	72.97 (22.99)	.26	−3.80 to 1.09	0.25	Small
500	21	78.81 (23.01)	79.60 (23.47)	.45	−2.97 to 1.38	0.17	Negligible
750	21	84.90 (23.22)	85.54 (23.64)	.51	−2.60 to 1.33	0.15	Negligible
1000	21	86.57 (23.67)	86.99 (24.01)	.67	−2.45 to 1.61	0.09	Negligible
1500	21	85.76 (23.53)	86.45 (23.43)	.47	−2.62 to 1.25	0.16	Negligible
2000	21	81.52 (24.16)	82.99 (24.19)	.14	−3.45 to 0.52	0.36	Small
3000	21	80.90 (22.50)	82.54 (23.55)	.12	−3.76 to 0.48	0.35	Small
4000	21	85.90 (22.49)	80.06 (22.66)	<.001b	3.87 to 7.81	−1.35	Large
6000	21	75.90 (23.11)	75.62 (24.84)	.77	−1.70 to 2.27	−0.07	Negligible
8000	21	59.76 (20.09)	64.30 (21.65)	.002b	−7.27 to −1.81	0.76	Moderate

^a P values were calculated by paired sample t test.

^b P value was below the threshold for statistical significance (P < .002).

Discussion

Very little is known about the noise measurement capabilities of the Noise application given the proprietary nature of the algorithms and modeling assumptions used to create it.² This diagnostic study is, to our knowledge, the first to formally quantify these capacities. Based on the present data, the smart watch application was as accurate as a class 1 sound level meter in the frequency range of 125 to 3000 Hz and at 6000 Hz, with minor (<5 dB) differences at 4000 and 8000 Hz. Prior work by Crossley et al in 2021 considered a correlation of 0.9 to represent acceptable accuracy for commercially available iPhone applications.⁵ By this metric, the Noise application can be considered accurate across all frequencies tested in this analysis, given correlation of at least 0.961.

The World Health Organization has estimated that 1.1 billion teenagers and young adults are at risk for unsafe sound exposure.⁶ Given the present results, this smart watch application may be useful in identifying harmful noise exposure environments where a class 1 sound level meter is unavailable.

This study did not assess performance outside of an audiometric booth and only studied pure-tone sound exposures. Specific variables at work in real-life settings that may limit sound measurement include interference with microphones by wind and blockage of sound by clothing. In addition, sound exposure at the wrist is not a perfect representation of sound exposure at an individual’s ear. Finally, harmful sound exposure in work environments is rarely composed of pure tones, and the accuracy of the smart watch application at measuring exposures comprising more complex sounds remains a future opportunity for research.

Supplement.

Data Sharing Statement