1、 Rep. ITU-R BS.2103-1 1 REPORT ITU-R BS.2103-1 Short-term loudness metering (Question ITU-R 2/6) (2007-2008) Foreword This Report is in two parts. The first part discusses the need for different types of programme loudness meter for different contexts, discusses general psychoacoustic principles, sp
2、ecific psychoacoustic principles in the context of temporal loudness sensation, and discusses the incorporation of these principles into ballistic characteristics for short-term loudness meters. The second part of the Report deals with correlation of short-term meter characteristics with loudness li
3、stening tests. This correlation is examined for several possible equalization curves and for a range of ballistic rise and fall time constants. The purpose of a short-term loudness meter in broadcasting is to give a short-term indication that reflects the likely contribution to long-term loudness, r
4、ather than short-term loudness per se. While these quantities are related, recent research has shown that they are not the same thing. The references used for performance assessment of each short-term meter variant are therefore listening test results on long-term loudness. CONTENTS Page Part 1 1 In
5、troduction 2 2 Application specific loudness meters . 2 3 Human loudness perception . 3 4 Review of proposed ballistic types. 4 5 Implementation theory. 6 5.1 Algorithm topology 6 5.2 Output filter topology . 7 5.3 Time constant for instantaneous and fast response. 7 5.4 Audio block decimation 7 2 R
6、ep. ITU-R BS.2103-1 Page 6 Implementation Loudness meter comparison utility . 8 6.1 Introduction. 8 6.2 Calibration 8 6.3 Decimation 8 6.4 Gating . 10 7 Conclusions 10 Part 2 1 Purpose . 11 2 Method 11 3 Results 12 3.1 Meter types . 12 3.2 VU and PPM results . 13 3.3 Loudness meter results . 15 4 Co
7、nclusions 19 5 References and Bibliography 20 Appendix 1 Results in numerical form . 21 Part 1 1 Introduction Radiocommunication Study Group 6 has approved a number of new Recommendations for international standardization of loudness meters. This has occurred in response to a need by many member adm
8、inistrations for such an instrument or set of instruments. This need has arisen because of a number of recent developments in broadcasting. 2 Application specific loudness meters Loudness meters are needed for several applications, including: For emission, as a quality-control check. To confirm that
9、 loudness is within acceptable limits. For production, as a level meter to replace the VU meter. Rep. ITU-R BS.2103-1 3 In quality control, a slow (many seconds) averaging meter is needed so that the average programme loudness can be read easily. A numerical readout is an advantage in these applicat
10、ions as it removes any need to interpret needle swings and perform any visual average estimation. A numerical readout also allows corrective gain to be easily determined. In the production role, particularly if the loudness of complete programme mix is displayed by means of a single meter readout, i
11、t may be useful to have some indication of the perceived dynamic range as well as having an indicator fast enough to display individual loudnesses where these can be discriminated aurally. This gives rapid feedback to the programme maker on which to make decisions about individual levels in the mix
12、or overall mix level. This can be particularly important in productions “on location”, where for various reasons, audio monitoring may not be optimal. For these reasons, a moving-indicator with a short (less than one second) averaging time is needed in production for loudness monitoring. Such a mete
13、r could show instantaneous loudness or a slowed average of instantaneous loudness to simplify reading the instrument. 3 Human loudness perception While loudness models for arbitrary sound types and for arbitrary sound levels are very complicated, ITU-R work has shown that for typical broadcast mater
14、ial and a limited range of loudness, a simplified model can give surprisingly good correlation with perceived average loudness over a 10-15 s interval. There is, so far, no evidence to show that this model cannot also indicate instantaneous loudness under the same constraints. In fact, it is essenti
15、al to use the same model for instantaneous loudness assessment if results are to be obtained which are consistent with long-term average loudness measurements. The model is detailed in Recommendation ITU-R BS.1770. The main concern in this Report is adapting this model to indicate instantaneous or s
16、hort-term loudness. Human sensory perception follows Stevens Power Law Stevens, 1957. The general form of the law is: akII = )( where: I: magnitude of the physical stimulus : psychophysical function capturing sensation (the subjective size of the stimulus) a: exponent that depends on the type of sti
17、mulation k: proportionality constant that depends on the type of stimulation and the units used. This law is a consequence of the fact that biological processes, and perception in particular, can usually be modelled by first-order differential equations. These models generally apply to both the magn
18、itude of sensation and to the onset and disappearance of the sensation. In other words, they generally apply in both amplitude and time. In terms of temporal models of sensation, this means that sensations do not suddenly appear and disappear, but rather the sensation becomes gradually apparent when
19、 the stimulus is presented and the sensation gradually disappears after the stimulus is removed. This is illustrated by Zwickers data on temporal loudness effects for a 2 kHz toneburst stimulus (see Fig. 1) which indicate the relative loudness of a toneburst as a function of its duration. This can a
20、lso be interpreted as the onset of loudness perception at the start of a continuous tone. 4 Rep. ITU-R BS.2103-1 FIGURE 1 Zwickers data on temporal loudness sensation 4 Review of proposed ballistic types First order differential equations have solutions for impulse and step stimuli which are simple
21、exponential functions. Modelling temporal loudness perception can therefore be done with first-order analogue filters which have simple exponential rise and fall characteristics for such stimuli. This type of filtering can be replicated in the digital domain using a first-order IIR topology or an FI
22、R topology, although the IIR topology is far more computationally efficient. The FIR equivalent has an exponentially falling set of sample weights, corresponding to the impulse response of the first order analogue filter. It has been suggested in discussions on loudness metering that the long-term l
23、oudness algorithm, which uses FIR filtering with rectangular weighting, could be adapted to model instantaneous loudness perception by simply shortening the time window over which it is applied. This would provide a poor model of loudness perception, as illustrated in Figs. 2 and 3. Figure 2 shows t
24、he leading-edge step response of an FIR filter with rectangular weighting and a time window of 100 ms (the best fit for Zwickers data), compared with a first-order IIR filter with the same time constant. The IIR filter performance follows Zwickers data within 0.1 dB, while the rectangular FIR filter
25、 has a worst-case 2 dB error at 0.1 s. The agreement between the two is otherwise good in the early and late phases. Rep. ITU-R BS.2103-1 5 FIGURE 2 Leading edge step response comparison of two filter topologies Figure 3 shows a similar comparison for a step response at the trailing edge. Here the d
26、isagreement is more marked. While the IIR filter falls off gradually and continuously, the rectangular FIR filter falls gradually at first and then drops precipitously at T = 100 ms. This FIR filter is unlikely to be a good model of perception. It gives a very discontinuous representation which does
27、 not occur in actual sensation. In other words, it is “jumpy” or “jerky”. Since human perception generally follows Stevens Power Law, and this law represents the behaviour of biological systems which can generally be modelled by first-order differential equations, and since first-order IIR filters m
28、odel such equations well, and since rectangular-weighted FIR filters do not model such equations well, we should use a first-order IIR filter in any loudness meter that attempts to model instantaneous loudness perception. 6 Rep. ITU-R BS.2103-1 FIGURE 3 Trailing edge step response comparison of two
29、filter topologies 5 Implementation theory 5.1 Algorithm topology For consistency with the long-term loudness meter, the implementation of the short-term loudness meter followed the block diagram in Fig. 4. FIGURE 4 Modified loudness algorithm block diagram Rep. ITU-R BS.2103-1 7 5.2 Output filter to
30、pology The low-pass filter stage at the output was functionally equivalent to the first order IIR topology shown in Fig. 5. FIGURE 5 Signal flow diagram for a 1st order filter 5.3 Time constant for instantaneous and fast response The time constant of the filter should be a minimum of T = 100 ms to a
31、void exaggeration in the modelling of perceived instantaneous loudness. T = 100 ms will indicate the perceived dynamic range of the programme. A longer time constant could be used, at the discretion of the operator, to indicate the average loudness with reduced meter swing. Operational trials in Jap
32、an have shown a longer time constant to be preferred over the 100 ms time constant for short-term loudness measurement and indication. It is the view of Japanese experts that a ballistic that changes faster than the commonly used VU meter is not desirable for short-term loudness measurement and indi
33、cation. 5.4 Audio block decimation Decimation by averaging blocks of samples may be used to reduce the computation load on the output filter. The decimated sample rate is arbitrary but should be significantly higher than 10 sample/s. Tables 1 and 2 give examples of filter coefficients for a decimate
34、d sample rate of 320 samples/s. TABLE 1 Filter coefficients for first order low-pass filter, T = 100 ms Sample rate 320/s Gain 6.465674116e + 01 b01 a10.9690674172 b1TABLE 2 Filter coefficients for first order low-pass filter, T = 400 ms Sample rate 320/s Gain 2.556465999e + 02 b01 a10.9921767002 b1
35、8 Rep. ITU-R BS.2103-1 6 Implementation Loudness meter comparison utility 6.1 Introduction A software application called Loudness Meter Comparison Utility (LMCU V1.5) has been written by the Australian Broadcasting Corporation to allow comparison of VU, IEC Type II PPM and ITU-R loudness meter chara
36、cteristics (Recommendation ITU-R BS.1770 Algorithms to measure audio programme loudness and true-peak audio level). The software runs on a personal computer with the Microsoft Windows XP operating system and may be obtained from the downloads section of the SRG-3 Yahoo egroup (http:/ The LMCU softwa
37、re shows a waveform display (see Fig. 6) for a selected .wav file with an overlaid envelope for each of the three meters, plus a vertical bar graph display of the three meters side by side. Below each bar graph display is a statistical summary of the meter amplitude for the duration of the .wav file
38、, showing median, 75%, 95% and 100% levels. For the loudness meter, arithmetic mean1and geometric mean value are also provided. The software allows a first-order low-pass output filter to be selected from a choice of no filter, fast filter (T = 100 ms) and slow filter (T = 400 ms). The raw loudness
39、data can also be exported to an Excel spreadsheet for further analysis. The T = 100 ms filter (Fast) gives an indication of subjective dynamic range as it corresponds to the temporal integration characteristic of the ear at 2 kHz. This can be quite fatiguing to watch for long periods however, so a s
40、econd time constant of 400 ms is used to slow down the meter and reduce the size of the swings around the average value. As well as the ergonomic advantages of this approach, it is consistent with sound level metering practice in acoustics. The software has a play control that allows the .wav file t
41、o be played through the standard audio output device. A cursor scrolls across the screen as the item is played. 6.2 Calibration The levels are calibrated in dB, with the VU meter calibrated so that a sine wave peaking at clipping level reads 3 dB. The loudness meter is also calibrated to read 3 dB f
42、or a sine wave peaking at clipping level at 1 kHz.The meter time constants are shown in Table 3. 6.3 Decimation Audio sample decimation (sample rate reduction) was incorporated in the LMCU software. The decimation algorithm in the LMCU uses a simple arithmetic average for each block of samples. A si
43、n(x)/x weighting window could be used instead if desired. The block size was set to be significantly less than 100 ms, the time constant of the fast output filter. The block size chosen, 3.125 ms (150 samples at 48 kHz) was arbitrary and shorter blocks could also be used without interfering with the
44、 output filter characteristic. Decimation was incorporated for three reasons: to reduce the computation load on the output filter; to simplify graphic display of the meter envelope; to simplify animation of the bar graph display. The effect of decimation on the mean loudness value is negligible. As
45、it may be a useful tool for loudness meter builders, it is suggested that it should be an optional part of the instantaneous loudness meter specification. 1The arithmetic mean loudness is the measure specified in Recommendation. Rep. ITU-R BS.2103-1 9 FIGURE 6 Screen shot of LMCU software (showing t
46、rack 9 from EBU SQAM test CD) TABLE 3 Meter time constants Meter LP filter order Rise time Fall time PPM (IEC Type II) 1 2 0.5 dB from peak for 10 ms toneburst at 5 kHz 2.8 0.3 s for 24 dB fall VU 2 99% of final value in 300 ms, overshoot 1%-1.5% Symmetrical Loudness, no filter Instantaneous Instant
47、aneous Loudness, fast filter 1 First order low-pass filter, T = 100 ms Symmetrical Loudness, slow filter 1 First order low-pass filter, T = 400 ms Symmetrical 10 Rep. ITU-R BS.2103-1 6.4 Gating For instantaneous loudness indication, gating should not be used. Gating will give a false indication of b
48、ackground noise on wide-range meters during silences. 7 Conclusions Recent developments in broadcasting have created a need for a new type of level meter which can indicate perceived loudness more accurately than existing level meters. Existing metering systems were developed to indicate either peak
49、 level or a combination of peak level and an approximation of perceived loudness. The importance of peak level however has diminished with the introduction of digital audio production and emission methods with their expanded headroom and immunity to over-modulation. While it is sufficient to know a single-number long-term average loudness for quality control in emission, it is also necessary to have short-term loudness information in production work. This indicates that several types of loudness me
