1	A study of hearing damage caused by personal MP3 players Adriano Farina Liceo Ginnasio statale G.D. Romagnosi, Parma, Italy adriano@pcfarina.eng.unipr.it
2	Objectives Measuring noise exposition of teenager subjects with real-life volume settings and complying with the following: Following international and european standards about measurements techniques Using a test signal that is both standard and similar to real-life music
3	The “IEC” test signal
4	Generating the signal One minute of pink noise was generated It was equalized in order to obtain the desired spectrum The resulting sound had an average RMS value of -16dBFS instead of the standard -10
5	"In order to obtain the..." In order to obtain the correct RMS value we employed the Hard Limiting module of Adobe Audition, with a 6dB gain The Graphic Equalizer was again used to correct the minor distortion caused by the Hard Limiting The resulting sound was measured compliant with the IEC standard
6	"However" However, it was noted by Alastair Hardie, a Senior Electronic Engineer for Frontier Silicon, that the Crest Factor had a 3.1373 / 3.1372 ratio, instead of the 1.8/2.2 specified in section 5.1 of standard EN 50332-1:200
7	Crest Factor Problem While the IEC standard states that the programme simulation noise must have a crest factor ranging between 1.8 and 2.2, it was technically impossible to generate such a noise. In fact the standard is written considering an analog crest factor measurement.
8	"Analog measurements were performed using..." Analog measurements were performed using a quasi-peak detector, which systematically underestimated the result. It was thence attempted to emulate it employing the Statistical Analysis tool of Adobe Audition. Specifying a window width of 35 ms, this tool computes correctly the pseudo-peak value as maximum RMS, if you add 3dB to the result (or by specifying that 0 dB = FS sine wave). It was checked that with these settings one gets the readings specified in table A-II of the IEC standard, employing a 5kHz tone burst of 1ms length.
9
10
11
12	The “Music” test signal
13	Generating the signal First we had to measure an average spectrum from the collected music, which was done with the following system:
14
15	Generating the signal We then proceeded to generate the signal using the same method employed for the IEC one
16	Signal Comparison
17
18	File formats employed In order to perform the measurements the test signals were put on the digital audio players using the best codec available for each device (uncompressed wav when available). The formats employed are the following: Uncompressed WAV (44100 Hz, 16 bits, stereo) WMA Lossless WMA 192 kbps WMA 128 kbps MP3Pro 144 kbps MP3 192 kbps MP3 128 kbps Apple Lossless AAC 192 kbps AAC 128 kbps
19	Equipment and calibration
20	Ambassador dummy head This dummy head is compliant with IEC 60959, and is specifically manufactured for testing “internal” hearing aids The microphones have been calibrated, removing the pinna, and inserting over the capsule a Bruel & Kjaer type 4100 reference sound source, which provides a pure tone at the frequency of 1 kHz and with an RMS sound pressure level of 94 dB (1 Pa). The calibration signal was employed for setting the calibration of the SpectraRTA program, as shown here:
21	Ambassador dummy head However, it is also necessary to correct for the frequency response of this specific Head and Torso Simulator. The manufacturer does not provide a suitable free-field frequency response for the Ambassador dummy head. So it was necessary to employ the results of anechoic impulse response measurements which had been previously performed on this specific dummy head at the anechoic chamber of Winterthur (Switzerland), kindly made available by Rieter Automotive.
22	Frequency Response Correction SpectraRTA already provides the capability of correcting for the frequency response of the microphones employed, so the compensation of the frequency response of the Ambassador dummy head did not require any effort.
23	Spectra RTA SpectraRTA was configured for measuring a linearly-averaged spectrum in 1/3 octave bands, averaging the signal of both channels (ears), and computing an unweighted spectrum and an A-weighted wideband value. Each test signal was 60 s long, but the measurement time was set 30 s, leaving 20s at beginning for allowing the device to stabilize before starting the measurement.
24	Measurement procedure Each device was measured 5 times, dismounting and remounting the headphones each time, as recommended by standard EN 50332, in order to reduce the mounting error. The results were then averaged. For ensuring a better correlation between the results and in order to minimize the measurement time, we used the same headphone position with both test signals (IEC and MUSIC). The headphone was inserted, then the two signals were measured and then the headphones were removed and reinserted. The volume control of the player was left untouched since the last usage from the owner of the device. These results are thence not significant for discriminating "dangerous" devices from "safe" devices.
25
26
27	Results
28	SPL Measurements For each device under test it was possible to obtain two values of the “exposure sound pressure level”: the first based on the IEC programme test signal, the second on the MUSIC test signal. The following table shows the results, in terms of average SPL +/ the standard deviation.
29
30	Dispersion of the results Results are usually quite similar for the same device Huge differences between devices No difference between the signals
31
32	Frequency Response At 3150 Hz a strong ear duct resonance is present, due to the air trapped behind the ear bud. Looking at the free-field frequency response of the Ambassador dummy head, the peak in the frequency response was instead at 5 kHz, corresponding to the “dip” in the curves of the figure.
33
34
35	Frequency Response It is wrong to employ the free-field frequency response for correcting the recorded signals, as the free-field response does not take into account the modification of the ear duct resonance occurring when an ear bud is inserted in the pinna. This can be seen as a severe inconsistency of the current EN 50332-1 standard. It could be more advisable to employ a diffuse-field response (which is usually smoother)
36	Analysis of human exposure A significant number of the devices under test was found set for over 90 dB(A) playback levels. In terms of assessment of noise-induced health risk, usually a “safe” value is considered to be a daily exposure of 8 h at 80 dB(A). Whenever higher SPL values are present, the duration of the exposure should be reduced, in order of keeping the same daily “noise dose”. An energetic equivalence principle is assumed, which means that the exposure should be reduced at 4h for an SPL of 83 dB(A), to 2h for an SPL of 86 dB(A), and so on. For each of the devices under test, it was computed what is the maximum time allowed daily for employing it for listening to music, as shown in the following table:
37
38	Conclusions 5 of the 13 devices have a time limit below 1h Most users declared a 1 to 2 hours average usage A proper hearing risk assessment should consider a whole day, with all its activities. A lot of money has been spent for reducing the noise at the workplace, but this will be wasted if we allow our teenagers to be over-exposed for years due to the usage of personal digital audio players capable of dangerous levels.