Page 44 - January 2007

1980 standard with one additional
restriction. Whereas ANSI S3.4-1980 is
applicable to both 1/3-octave and 1/1-
octave bands, ISO 532 method A is lim-
ited to an octave-band analysis. This
additional restriction further limits the
suitability of method A for loudness cal-
culations. ISO 532 method B improves
method A in two important respects.
First, ISO 532 method B is applicable to
sounds with strong line spectra (e.g.,
Hellman and Zwicker, 1987). Second,
ISO 532 method B enables total loud-
ness to be calculated from a one-third-
octave band analysis. The new ANSI
S3.4-2005 standard provides distinct improvements that extend its suitability for calculating the loudness of steady sounds.
The origin of the new ANSI S3.4-2005 loudness standard has a long and significant history in psychoacoustics dating back to the pioneering research of Fletcher and Munson (1933). It is based on a model of loudness perception (Moore et al., 1997) that was developed from a model originally pro- posed by Zwicker and his co-workers (1958, 1965, 1984, 1999). Zwicker’s model is part of ISO 532:1975 method B. The new procedure for calculating the loudness of steady sounds is rooted in our current understanding of the human auditory system. The procedure is applicable to sounds presented in free field with frontal incidence, in a diffuse field, or via head- phones. It is available as a computer program that provides a loudness estimate in sones, and the corresponding loudness level in phons. The following features distinguish the new ANSI S3.4-2005 loudness standard from Zwicker’s procedure in ISO 532:1975 method B.
First, unlike ISO 532:1975 method B, ANSI S3.4-2005 is applicable to both monaural and binaural stimulus presenta- tions. The procedure is based on the simplifying assumption that the loudness from each ear is added to give the overall loudness when listening with two ears. For binaural presenta- tions with the same sound in both ears, the overall loudness is calculated as twice that for each ear separately (e.g., Fletcher and Munson, 1933; Hellman and Zwislocki, 1963; Marks, 1978). Second, although ISO 532 method B enables the loud- ness of sounds with strong line spectra to be determined (e.g., Hellman and Zwicker, 1987), its applicability is limited to mid- frequency tones combined with noise (e.g., Hellman and Zwicker, 1987; Hellman, 1991). When the added tone is locat- ed below 500 Hz, the modification by Moore et al. (1997) in ANSI S3.4-2005 provides a more accurate prediction of the total loudness of a tone-noise complex than loudness predic- tions based on ISO 532 method B (Hellman, 2002). The differ- ence at low frequencies is probably caused by the one-third- octave band analysis. In ISO 532 method B, one-third-octave band approximations are determined from Critical Bands (Frequenzgruppen). By comparison, in the new ANSI S3.4- 2005 standard, the one-third-octave-band approximations are determined from the Equivalent Rectangular Bandwidth (ERBN) of the auditory filter (Glasberg and Moore, 1990). At 1 kHz and above, the critical-band function and the ERBN func-
“The origin of the new ANSI S3.4-2005 loudness standard has a long and significant history in psychoacoustics dating back to the pioneering research of Fletcher and Munson (1933).”
tion closely agree (Patterson and Moore, 1986). However, below 500 Hz, the criti- cal bandwidth and the ERBN filter deviate markedly. In this low-frequency region, the critical bandwidth has a constant value whereas the ERBN value decreases as frequency decreases. At center fre- quencies below 500 Hz, the smaller ERBN values provide a more accurate represen- tation of the bandwidth of the auditory filter (Moore and Sek, 1995) than the crit- ical-band function. As noted in the analy- sis of old ANSI S3.4-1980, this improve- ment probably also accounts for the good agreement between the calculated equal-
loudness contours based on ANSI S3.4-2005 and the recently revised ISO 226:2003 contours at low frequencies. Third, like old ANSI S3.4-1980, the relation between loudness in sones and loudness level in phons in ISO 532 method B is based on an obsolete and withdrawn ISO standard (ISO/R131:1959). As seen in Fig. 1, this withdrawn ISO standard produces a loud- ness function relating loudness in sones to loudness level in phons that is described by a simple power function down to 20 phons. The accumulated empirical evidence clearly shows that below 40 phons a simple power function departs markedly as threshold levels are approached from the loudness function generated by the new ANSI S3.4-2005 loudness-calculation procedure. The departures are especially noteworthy as thresh- old is approached. Both experimental data and auditory theo- ry support these results (e.g., Hellman and Zwislocki, 1961, 1963; Hellman, 1976; Zwislocki, 1965; Buus et al., 1998). Instead of an incorrect value of zero loudness at threshold, within the new ANSI S3.4-2005 procedure, the absolute thresh- old of a sound is taken to correspond to the level at which the procedure gives a calculated loudness of 0.003 sones. This loud- ness value is in accord with the loudness model of Moore et al. (1997). It corresponds to a loudness level of 2.2 phons
1
(Glasberg and Moore, 2006). The procedure can therefore be
used to calculate the absolute threshold of a given sound by determining the input level that leads to a calculated loudness of 0.003 sones. Thus calculated, the derived absolute threshold for sinusoidal stimuli are very close to those specified in ISO 389-7:2005.
Another problem with ISO 532 method B is that the cal- culations are based on tables of values, which result in quan- tization effects; the calculated loudness level in phons may “jump” by around 2 dB when the input sound level is changed by only 0.1 dB. Loudness as calculated according to ANSI S3.4-2005 varies smoothly and continuously with input sound level
Summary and conclusions
The new ANSI S3.4-2005 standard for calculating the loudness of steady sounds is described and compared to the old ANSI S3.4-1980 standard. The new ANSI loudness stan- dard is also compared to the loudness standard in ISO 532. Taken together, the analysis indicates that loudness calculat- ed according to ANSI S3.4-2005 provides a distinct improve- ment over either old ANSI S3.4-1980 or ISO 532 method A
42 Acoustics Today, January 2007