User manual | null User manual

Implementing Methods
for Equal Loudness
in Radio Broadcasting
Matti Zemack
Supervisors
Royal Institute of Technology: Professor Sten Ternström
Swedish Radio: Technical Strategist Lars Jonsson
Date of approval: 12th June 2007 • Approved by: Professor Sten Ternström
Master of Science Thesis
KTH - Skolan för Datavetenskap och kommunikation (CSC)
Avdelningen för Tal, musik och hörsel
100 44 Stockholm
Table of Contents
Implementing methods for equal loudness in radio broadcasting
Abstract in English
Abstract in Swedish
Recommendations for Swedish Radio at implementing better loudness control
1
What is loudness?....................................................................................................1
1.1
1.2
1.2.1
1.2.2
1.2.3
1.2.4
1.3
2
Spectral effects on loudness..............................................................................................2
Phon scale .........................................................................................................................3
Sone scale .........................................................................................................................5
Temporal aspects of loudness ...........................................................................................6
Approach to the problem ...................................................................................... 6
Recent research .......................................................................................................7
2.1
2.1.1
2.1.2
2.1.3
2.1.4
2.1.5
2.1.6
2.1.7
2.1.8
2.2
2.2.1
3
‘Perceived loudness’ or just ‘loudness’ ................................................................ 2
How does the ear interpret loudness? .................................................................. 2
Different models ..................................................................................................... 7
Leq (Linear, A-, B-, C-, D-, M-, RLB-, R2LB-weighted) ................................................7
PPM ..................................................................................................................................9
Zwicker (SI++, ISO 532-B)..............................................................................................9
CBS Loudness Indicator ...................................................................................................9
Moore & Glasberg ..........................................................................................................10
TC LARM.......................................................................................................................10
TC HEIMDAL................................................................................................................10
Replay gain .....................................................................................................................10
Comparison of models ......................................................................................... 12
Conclusions – comparing methods .................................................................................13
Loudness at the Swedish Radio ............................................................................15
3.1
Measured loudness. Comparing before and after the Swedish Radio final
dynamic processors. .......................................................................................................... 15
3.1.1
3.1.2
3.1.3
3.2
Method............................................................................................................................15
Different software measuring systems............................................................................17
Measurements.................................................................................................................19
Results of measurements ..................................................................................... 20
3.2.1
Results – short recordings...............................................................................................20
3.2.1.1
P1 speech channel .................................................................................................20
3.2.1.2
P3 pop music / speech channel..............................................................................22
3.2.2
Results – long recordings................................................................................................24
3.2.2.1
P1 speech channel .................................................................................................25
3.2.3
Comparing Replay gain with Leq(R2LB).......................................................................27
3.3
4
Workflows and levels at the Swedish Radio.........................................................31
4.1
4.1.1
4.2
4.2.1
4.2.2
4.3
4.3.1
4.3.2
4.3.3
4.3.4
5
Conclusions of measurements ............................................................................. 30
The pre-digital era ............................................................................................... 31
Methods and workflow – pre digital era .........................................................................32
Digital era.............................................................................................................. 33
Methods and workflow – pre-produced in the digital era ...............................................33
Methods and workflow – live radio in the digital era .....................................................34
The future – method and workflows .................................................................. 35
Production monitor sound levels?...................................................................................35
How to implement fully controlled loudness levels in an automatic broadcast..............36
Manually calculating parts of a show .............................................................................37
Full sound file levelling ..................................................................................................37
Real life meter usage.............................................................................................39
6 Discussion concerning a new meter.....................................................................41
7
Other comments closely related to loudness in broadcasting..............................43
7.1
7.2
7.3
7.4
Balance speech/music........................................................................................... 43
Classical music...................................................................................................... 43
Different meter usage between channels............................................................ 43
Different dynamics for different usages ............................................................. 45
8
Acknowledgment ...................................................................................................47
9
Glossary .................................................................................................................49
10 Production Flowchart for Swedish Radio............................................................51
11 References .............................................................................................................53
Implementing methods for equal loudness in radio
broadcasting
Abstract
Sound levels are perceived as a growing problem in radio and TV. Quite often, great
variations in perceived sound level exist inside a single program or between adjacent
programs. Today the broadcaster uses a plethora of media platforms, all with different
listener groups. They all have one thing in common; they all want even perceived
sound levels. How can a broadcasting company accomplish this?
This objective can be achieved by intentional work in consecutive steps. The first step
is to assimilate the latest research in this area. The second step is to choose the best
measurement method. The third step is to implement this single measurement method
in all steps of production. Training and support for all programme producing staff is a
must. The fourth step is to implement an automatic gain measurement and correction
feature in the metadata of the play out system. The fifth step is that the broadcast
company must itself try to control as much as possible of the final dynamic
processing.
In this paper, the above steps are examined, and some recommendations, large and
small, are proposed for Swedish Radio about how their broadcast chain may be
improved so that better perceived sound levels are achieved.
The methods of measurement that are tested in this report are both Leq(R2LB) and
Replay gain. I have also compared the final dynamic processing systems at Swedish
radio. Both of these measure methods and the final processing systems, Factum
Cadenza together with Orban 8200 work very well.
With the use of these tools, Swedish Radio can achieve more even perceived sound
levels, which is important to keep and obtain new listeners.
Automatiska metoder för jämn hörnivå i rundradio
Sammanfattning
Ljudnivåer uppfattas som ett allt större problem inom radio och tv. Inom program och
mellan program märks stora hopp i ljudnivåerna. Idag använder lyssnaren en mängd
olika mediaplattformar, och de har alla olika lyssnargrupper. Men en sak har de
gemensamt, de vill alla ha jämna ljudnivåer. Hur kan ett broadcastingföretag
åstadkomma uppfattat jämna lyssningsnivåer?
Målet kan uppnås genom medvetet arbete i flera steg. Första steget innebär att
tillgodogöra sig den senaste forskningen på området. Andra steget är att i denna
forskning hitta en tillförlitlig mätmetod. Tredje steget innebär att all produktion måste
följa denna enda mätstandard. En utbildning av alla programproducerande
medarbetare måste genomföras. Fjärde steget är att en automatisk korrigering av de
färdiga programmen måste göras vid eller inför programläggning till utsändaren.
Femte steget går ut på att företaget självt ska ta ansvaret för sin egen avgående signal
med hjälp av slutprocessorer för att i alla distributionskanaler kunna kontrollera
utsänd dynamik och nivåer.
I detta examensarbete utreds alla de ovanstående stegen. Rapporten lämnar även
rekommendationer, stora som små, till SR om hur just deras sändningskedja ska
kunna nå jämnare ljudnivåer, vilket leder till bättre hörbarhet.
Mätmetoderna som främst analyseras är Leq(R2LB) samt Replay gain för att mäta
uppfattade ljudnivåer i talad radio. Dessa jämförs med den slutprocess som redan idag
finns hos Sveriges Radio. Båda metoderna ger mätmässigt ifrån sig ett bra resultat,
liksom slutprocessen, Factum Cadenza.
Sveriges Radio kan uppnå bättre hörbarhet vad gäller uppfattade ljudnivåer, vilket är
viktigt för att hålla kvar gamla lyssnare och för att rekrytera nya.
Recommendations for Swedish Radio at implementing
better loudness control
Below are the recommendations to Swedish Radio regarding the implementation of
loudness at Swedish Radio.
•
Define a metering standard, the same for all channels (including web channels,
pod casts, etc.), preferably using a new metering model. The recommendation
is Leq(R2LB) as proposed by ITU BS.1770 (ITU-R, 2006).
•
The meter is recommended to be similar to the BBC meter, which resembles
the old-style VU meter. This type is easier to comprehend in the corner of the
field of sight. The meter must also be adjusted so that the stipulated Loudness
Unit (LU) is with the needle pointing straight up. The meter should also give
big response at small level changes, the most interesting metering area is
centred around a 12 dB range.
•
Educate all co-workers as to the usage of this new meter.
•
Incorporate automatic measurement into the system, such that when a finished
sound file is submitted to the broadcast intake, a sound level measurement and
metadata adjustment is carried out automatically.
•
Document and study the main processing units. Try to use them more
offensively. Use the dynamic processors harder (or maybe softer for some
program types). More experiments and discussions must be introduced.
•
Equip the voice tracking studios (where the main play out level is set) with
different listening devices, such as big speakers, small speakers, computer
speakers and headphones. Encourage the producers to use them all, so that
they can understand the differences.
•
Post process the web feeds and pod feeds to a much higher extent. Let the
dynamic processors work harder.
•
Design a few examples of sound excerpts where correct levels are set, to
establish a norm for producers and engineers. These examples can also show
how loud we usually should mix music or sound effects behind speech, or how
loud pop music can be relative to the presenter. Distribute these sound files
both through Swedish Radio’s internal distribution systems, and to external
production studios.
1 What is loudness?
Loudness is a subjective entity. Every individual person’s subjective impression of
sound intensity is unique. Can loudness even be measured, can any fragment of sound
be given an exact value so that two different sound fragments can be presented in
sequence without the user reaching for the volume knob on their radio or TV? And
how exactly can this measurement be done?
The research concerning loudness evaluation has made great progress recently. This
may be as a result of television and radio starting digital distribution. The sound
chain, from the producer handling the interview or the ad, to the consumer in front of
their screen or radio, is implemented entirely in the digital domain. Globalisation in
media production has also led to problems where one TV channel controls their
outgoing loudness with the type of meter used in that country, and the other country,
adding the commercials controls loudness according to a different standard and meter.
And as we know from commercial radio, a ”louder channel is a better channel” which
leads to louder and more processed signal levels in a digital signal chain, with little
control at the receiving point.
Producers of commercials have always been using the technique of hard limited levels
in order to outperform their competitors. The sound nowadays is also multi-band
compressed in a further attempt to become louder, and as an easy way of levelling out
the frequency content in various neighbouring sound signals.
All this has led to an increased interest in measuring loudness, as a necessary means
to quantify perceived sound levels.
1
1.1 ‘Perceived loudness’ or just ‘loudness’
Loudness is defined as a subjective entity. Research papers in this area often use the
term “perceived loudness”. Is the wording “perceived loudness” a tautology? In this
paper perceived loudness will be named only “loudness”.
1.2 How does the ear interpret loudness?
1.2.1 Spectral effects on loudness
Our sense of hearing assesses loudness by how the cilia and corresponding auditory
nerve fibres are excited in the basilar membrane in the inner ear [Bonello, 2007]. This
excitation is distributed by frequency bands on the membrane, forming a kind of
biological spectrum analyzer. Each frequency excites a certain zone on the basilar
membrane. Each excited zone adds up to the total loudness.
If two sounds arrive into the ear with similar frequency content they both compete
trying to excite the same hair cells and the same nerves. These nerves have a
maximum rate of firing, and this is thought to be the reason why doubling the sound
intensity does not double the perceived sound level. See fig. 1.
Fig. 1. The figure shows why doubling the sound intensity at nearly the same
frequencies does not double the loudness. [Hyperphysics loud]
2
On the other hand, if the different sounds contain different frequencies they do not
occupy the same hair cells and therefore not the same nerves, hence adding two
equally loud sounds doubles the loudness, see fig. 2.
Fig. 2.
1) The figure shows why doubling the sound intensity at differing frequencies doubles
the loudness. [Hyperphysics loud]
2) Showing the placement of the basilar membrane in the inner ear. [Hyperphysics
place]
The maximum frequency distance for these differing sounds to fire the same nerves
are called critical bands. One way of measuring these bands has been proposed by
Zwicker et al. [Zwicker et al., 1957]. These critical bands are narrower at low
frequencies (90Hz wide critical bands for sounds under 200Hz) and wider at higher
frequencies (900Hz wide critical bands for sounds around 500Hz) [Backus, 1977].
1.2.2 Phon scale
The ear does not have a straight frequency response. The ear has its own built in
equaliser. Fletcher/Munson tested many recruits. Their findings were later
standardised by ISO, see fig. 3.
3
Fig. 3. Shows the ears frequency response at differing levels of sound intensity. By
Fletcher/Munson 1933. Revised by ISO in several steps since then. [Hyperphysics
(eqloud)]
These measurements were done with sinusoidal tones from the front. As can be seen
the curves do not look the same at all test levels. At lower listening levels the lowest
frequencies are perceived quieter than the mid frequencies. Therefore home stereo
systems usually have a loudness button that enhances these low frequencies at low
listening levels.
Using the Fletcher/Munson-ISO curves we can find the loudness level of a sinusoidal
tone. This is measured in phon. The phon value is defined by that a 1 kHz sinusoidal
tone measured in dB gives name to the whole phon curve. For example the 40 phon
curve has a 40 dB intensity with a 1 kHz tone.
It should be noted that the Fletcher/Munson curves are constructed by subjective
responses to a sinusoidal tone presented frontally. If instead a narrowband noise is
used and the sound comes from a diffuse and free field the corrections in fig. 4 need
to be done to the Fletcher/Munson-ISO curves.
4
Fig. 4. Attenuation necessary to produce the same equal loudness of a pure tone in a
diffuse and in a free sound field [Zwicker & Fastl 1999].
1.2.3 Sone scale
The sone is a unit of loudness after a proposal by Stanley Smith Stevens in 1936. The
sone scale is developed in tests where listeners are asked to define where the loudness
of a sound is doubled. One sone equals 40 phon, doubling the loudness doubles the
sones. The Son and the Phone scale coincide in Fig. 5.
Fig. 5. Conversion chart between sones and phons. According to ISO/R 131-1959.
[sengpielaudio]
5
1.2.4 Temporal aspects of loudness
The duration of the sound stimuli is also of importance. Zwicker/Fastl investigated
short sound impulses. They noticed a decreasing loudness as the sound impulse
became shorter, see fig. 6.
Fig. 6, Relative loudness of a 2 kHz tone-burst as a function of duration [Zwicker &
Fastl 1999].
Sound bursts above 100 milliseconds are steady in their loudness [Zwicker & Fastl
1999]. Sound bursts as short as 10 milliseconds have a loudness that is reduced by a
factor of 2, in other words half the loudness.
1.3 Approach to the problem
Why can we not measure loudness with a digital sample accurate peak (fast attack
times) measurement? These meters are easy to fool. Take for instance uncompressed
(actually today; non limited) pop music. The drums account for most of the peak
values. Because our perception is influenced also by duration, these instantaneous
drum sounds do not fully add up to the loudness.
If we use a meter with slower attack time these drum peaks will go by unnoticed by
the meter. So some type of slower response time is better matched by our ears
reactions.
6
2 Recent research
2.1 Different models
Recent research (Skovenborg / Nielsen) has evaluated the different loudness measure
algorithms. Below their findings are presented together with one other algorithm for
measuring loudness:
2.1.1 Leq (Linear, A-, B-, C-, D-, M-, RLB-, R2LB-weighted)
Leq is the equivalent continuous sound level. Leq is measured over time. The
measurement time may be either a whole sound file or live sound material presented
with an much shorter integration time, for example 300ms. Leq is measured as the
root-mean-square (RMS) in dB relative to a reference level that must be given
explicitly or defined by the context.
Leq is often measured together with different weighting curves to make the
measurement fit the human ears listening curves at different frequencies [Moore
1982]. The different weighting curves can be seen in fig. 7.
•
Leq(Linear) is Leq without any frequency weighting.
•
Leq(A) is Leq measured by first applying the A-weighting of the sound source.
The A-weighting is a simple approximation of the 40-phon equal loudness
curve. Usually the A-weighting is used in sound level measurements dB(A).
•
Leq(B) is Leq measured with a B-weighting curve, which is an A-weighting
modified for lower listening sound levels. The low frequencies are more
apparent here than in the A-weighting.
•
Leq(C) is Leq measured with the C-weighting frequency response filter. The
C-weighting is constructed for even lower listening levels than the Bweighting. Additional low frequencies are allowed.
•
Leq(D) is Leq measured with the D-frequency curve. Hardly used except in
“for aircraft engine noise measurement” [Todd, 2007].
7
•
Leq(M) is Leq where the M stands for Movie. It has mainly been “promoted
by Dolby, to be used for measuring the loudness off different segments of
movie soundtracks such as advertisements” [Skovenborg & Nielsen, 2004].
•
Leq(RLB) is Leq measured with the Revised Low frequency B-weighting. This
is actually a simple 50 Hz high-pass filter. RLB has since Skovenborg and
Nielsens paper been modified by the ITU-R group to also include a high
frequency boost [ITU-R; BS.1770, 2006]. The boost was added to account for
the acoustics of the head when measuring surround sound, where the head is
modelled as a rigid sphere. This new RLB has no official naming, it is seen as
an updated RLB. To avoid confusion I call this new frequency curve R2LB
and it is not included in Skovenborg and Nielsens comparison below.
Fig. 7. The different weighting filters used in Leq measure. A, B, C, M, RLB (green),
R2LB (blue). [Lund, 2007]
8
2.1.2 PPM
PPM is a Peak Program Meter. This is a Peak meter with a fast attack time, and a
slow decay time. The meter has a fast enough attack to make sure no peaks go over
the digital full scale, and it is slow enough to make sure the user has time to read the
display. Note that the attack time is not as fast as a digital peak meter. The reading is
usually in between digital peak meters and VU-meters.
For deciding the level of a whole sound file, the PPM measure is often presented in a
histogram. The level value of the sound file then becomes the 95th (or 75th or 50th)
percentile in this histogram.
Swedish Radio uses a version of the PPM called EBU/Nordic PPM.
2.1.3 Zwicker (SI++, ISO 532-B)
ISO 532-B: Zwicker has constructed a loudness model in which the frequencies are
divided into critical bands. Usually a sound’s frequency content is divided into 32 1/3octave bands. The excitation level in each critical band is calculated. The total
loudness is calculated by integrating the levels over all critical bands. The resulting
value is in sones. This model became the ISO 532-B standard. Zwicker also described
a manual method of this model [Brixen, 2001:2].
SI++: There also exists a variant of the Zwicker model which is implemented in the
acoustics software SI++. In this model, the 95th percentile of the loudness values is
used as an estimator of the loudness of a sound file. It is said that “The perceived
loudness of a long, non-stationary sound is the loudness value that is exceeded 5% of
the time in the loudness/time course.” [Akustik Technologie Göttingen, 2004]
2.1.4 CBS Loudness Indicator
This is the de facto standard in US broadcast community. This is like the Zwicker
model based on filter banks. The difference here is that only eight banks are used,
covering three critical bands each.
9
2.1.5 Moore & Glasberg
This is also a multi-band loudness model, comparable to the Zwicker model.
2.1.6 TC LARM
TC Larm is new single-band way of measuring loudness. It uses RLB-weighting, see
fig. 7, and also an asymmetrical low pass filter (with a release time slower than the
attack time) so that the peaks in the higher frequencies are accentuated.
Esben Skovenborg pointed out in private mail conversation that; ”Essentially, LARM
was conceived to demonstrate that another full-band model could be (at least) as
accurate as the Leq(RLB)“. [Skovenborg, 2007]
2.1.7 TC HEIMDAL
TC Heimdal is a multi-band model. The sound source is filtered into 9 bands. These
are further processed. This method also uses an asymmetrical low pass filter. As this
model was patent pending at the time of writing, the exact details were not available .
Skovenborg further explained that ”HEIMDAL then confirms that a multi-band model
is required, in order to achieve an even better accuracy of loudness measurements especially for input signals with atypical spectra”.
2.1.8 Replay gain
Replay gain is published openly as a community developed measuring system
[Replay Gain, 2001]. Work with Replay gain began in 2001 by David Robinson. It is
widely used, for example at Swedish Radio in the music ingestion system A-Wave.
The publisher stresses the different loudness measures, short or long term. In this
10
sense the difference short / long term is between CD track mode (“Radio”) or cd
album mode (“Audiophile”).
The loudness is calculated by first applying two filters: first, an IIR filter which
simulates the high frequency response of an average of Fletcher/Munson curves
above. To this Replay gain adds a high-pass filter to match the Fletcher/Munson
curves. Secondly, the RMS of the sound file is calculated in 50ms windows. These
values are plotted in a histogram, and the 95th percentile (sorted from the highest
values) is chosen as the loudness of the sound file.
Audio engineers at Swedish Radio believe that this algorithm works very well for pop
music. It has been very useful when ripping pop music CD’s into files. These files
have often been broadcasted by self-operators using only headphones. No audio
engineers with good acoustical environments using loudspeakers have been involved
in the level adjustment process.
11
2.2 Comparison of models
In “Evaluation of Different Loudness Models” by Skovenborg/Nielsen published at
AES in 2004, the models were classified into different classes based on their overall
performance compared to subjective listener tests at two separate locations (TC
electronics, McGill University). Their results are presented in table 1.
Performance
class
Models,
best in class first
Median
Absolute
error in
dB
Class 1 (best)
TC HEIMDAL
TC LARM
Leq(RLB)
Leq(C)
Leq(Linear)
Leq(B)
PPM(50th percentile)
Zwicker-ISO
Zwicker&Fastl(95th
percentile)
Leq(D)
Leq(A)
Leq(M)
0.52
0.61
0.67
0.72
0.77
0.84
1.12
1.22
1.22
95th
percentile
maximum
absolute
error in
dB
1.50
1.64
1.58
1.95
2.16
1.59
2.70
1.95
2.92
1.42
1.78
1.68
3.00
4.13
3.83
Class 2
Class 3
Class 4 (worst)
Table 1. Classes of loudness models, based on the overall evaluation. Median
(Absolute error chosen from the TC data set because TC HEIMDAL and TC LARM
were optimised using the McGill data set.)
The evaluation states that TC HEIMDAL and TC LARM have a mean absolute error
of only 0.5-0.6 dB compared to subjective tests at TC. Leq(RLB) is not much worse
in practical use. The 95th percentile maximum absolute error is even smaller for
Leq(RLB) than for TC LARM.
The standard error of different subjects for the TC data set is 0.43 dB. This data set
consisted of 8 subjects. The mean absolute error of Leq(RLB) is not far off with its
standard error being 0.67 dB.
12
2.2.1 Conclusions – comparing methods
The Skovenborg/Nielsen research suggests that it may be possible to devise a
universal machine for calculating loudness. This machine could calculate the gain
setting in an automatic broadcast system.
Skovenborg/Nielsen pointed out that the most exact measure is Leq(RLB). In the
following chapter everything is measured using the ITU recommendation Leq(R2LB).
13
14
3 Loudness at the Swedish Radio
3.1 Measured loudness. Comparing before and after the
Swedish Radio final dynamic processors.
The final dynamic processors at the Swedish Radio for P1 (the speech channel) and
P3 (mixed pop music and speech channel for young listeners) are constructed with an
automatic gain controller and a multi band dynamic processor in series.
The measurements were focused around the question; Do broadcasters need
Leq(R2LB) meters in live broadcast and automatic pre-broadcast measurement of
sound files using the Leq(R2LB), or can we handle the varying loudness satisfactorily
with the same final dynamic processors that we use today? Also the previous
unmeasured Replay gain algorithm has been tested against the main final processes at
the Swedish Radio.
3.1.1 Method
For this study, the P1 and P3 channels at the Swedish Radios were recorded both
directly before (pre process) and directly after (post process) the dynamic processing,
see section 10 of this thesis for a detailed description of the points of recording.
P1 is mainly the speech channel with news, discussions, documentaries, theatre, some
music and listener contact through telephone. P1 audio engineers for many years have
been using one compressor on the master from the mixing desk. Because of the new
digital mixing desks with separate compressors on every channel, every microphone
had its own dynamic settings at the time of my recordings.
P3 is a channel for the younger listeners. It is a high tempo pop music channel with a
high degree of speech content. The speech is often loud and heavily compressed at the
mixing desk. The music is level controlled when it is imported into the play out
system using the Replay gain standard.
15
These two radio channels were recorded digitally on an 8 channels Pro Tools-888
system. The recordings have a sampling rate of 48.000 Hz as this is the sampling rate
standard in the Swedish Radio on-air system Digas. The signal both from pre and post
process was digital AES/EBU, received through a break-out box in the central
apparatus room. The router matrix for the outgoing channels at Swedish Radio is in
parts still analogue, therefore the pre-process signal was recorded after the main A/D
converters. These A/D converters have different reference levels because the different
channels use their PPM EBU/Nordic meters in different ways. The P1 reference level
is set to -18 dB equivalent to test tone in studio at TEST level (PPM EBU/Nordic 0
dB). P3’s reference is set to -22 dB equivalent to test tone in studio at TEST level
(PPM EBU/Nordic 0 dB).
Recordings were done during October 2006. This report is based on recordings with
varying types of programme material, for this reason recordings from October the 26th
were chosen. The P1 recordings from this day included
•
Phone-in-programmes containing a mixture of narrowband telephones and
broadband studio microphones.
•
News, both a few short 3 minute versions and one half hour long version from
other studio.
•
Weather, live from SMHI, the weather bureau, without audio engineers.
•
Presenter, live in channel master studio.
•
Trailer, pre recorded louder than other programme material.
•
Trailer, live from other studio.
•
Theatre, pre recorded with greater dynamic usage than other programmes.
•
Interviews, pre recorded.
•
Dialogue in studio, live.
•
Sound effects, church bell every day at noon.
16
•
Music, pre-recorded.
Most of the material broadcasted on P1 is mono so only the left channel was used to
measure the Leq(R2LB).
3.1.2 Different software measuring systems
The Leq(R2LB) measurements in this paper were done using a meter by the
Communications Research Centre (Ottawa, Canada). It is called the CRC Loudness
Meter, see fig. 8. This is one of the two measuring devices available at the time of
writing this paper that properly uses the R2LB frequency weighting. The other
measuring software with the correct frequency weighting is the Loudness Meter
Comparison Utility (LMCU) by the Australian Broadcasting Corporation, se fig. 9.
In this report the CRC was chosen because it can be used as a live meter, which also
made it possible to measure live sounds at an early stage of the research. All
measurements referred to in this paper are pre-recorded and edited files. The LMCU,
on the other hand, has many advantages. It has the facility for easy comparison
between PPM, VU and Leq(R2LB) loudness. The LMCU also lets the user set a
sound level threshold which in effect would make the meter only react to sound, and
not letting the Leq(R2LB) be biased by silence between syllables and sentences in
speech. It also gives the user an option of selecting an output filter. This controls how
the Leq(R2LB) is biased by the length of the sounds. The fast setting is based on
Zwicker/Fastl’s tone-burst calculations in fig. 6.
Both the CRC and LMCU meters calculate a cumulative loudness value.
Another software measuring system that was tested is BBC’s Baptools, see fig. 10.
This software can only measure live input, and it only measures Leq(R2LB). I and
another mixing engineer working for Swedish Radio have been using this meter with
very pleasing results.
17
Fig. 8. The CRC meter by Gilbert Soulodre.
Fig. 9. The LMCU meter by Australian Broadcasting corporation.
Fig. 10. The BBC baptools meter by Andrew Mason, BBC.
18
3.1.3 Measurements
The above mentioned study [Skovenborg & Nielsen, 2004] have classified Leq(RLB)
as the best measure method apart from TC’s own methods. In this report this method
and its extension Leq(R2LB) is used. This method is also proposed by the ITUBS.1770 standard.
All measurements are done both pre- and post- the final processing at Swedish Radio.
The measurements have been done both on short (less than 20 seconds) excerpts of
sound and on whole radio shows. The measured Leq(R2LB) before processing gives a
hint of the differing loudnesses in the incoming signal. It gives an answer to the
question: How big difference do we receive in the incoming material to the central
apparatus room? The post processing measurement shows us how much the final
processors helped in smoothing the level jumps to improve the listening experience.
By measuring both short excerpts and long shows we can test the different methods of
measure . One hypothesis is that an audio engineer or a producer works to find an
even loudness of her own show, but with no control over the adjacent shows or
trailers. Will the 20 second measurement give a good enough hint of the loudness for
the full show? Will a 20 second quick listen be enough for an audio engineer to
smooth the play out levels between shows? Is it better to use processing equipment for
control of loudness inside shows, but use some form of overall loudness calculation as
a help to the processors?
As an additional measure, Replay gain is compared to the processing equipment at
Swedish Radio. This measurement is done to see how well the Replay gain algorithm
correlates with Leq(R2LB) and the final processing units at Swedish Radio.
19
3.2 Results of measurements
3.2.1 Results – short recordings
The first part of the results concerns the short recordings. This contains excerpts from
Swedish Radio the 26th October 2006. The excerpts are actively chosen to exemplify
special cases in broadcast, for example some examples are chosen because there is a
narrowband telephone and others are chosen because there are loud applauses among
the studio guests after a short live song.
3.2.1.1 P1 speech channel
Table 2 shows the Leq(R2LB) results from short file excerpts both before and after
1
2
3
4
5
6
7
8
9a
10
11
12
13
14
15
16
17
18
19
male_female_iv
male_iv
female_studio
male2_iv
male1LoudStreet_iv
female1_tel
male_studio_Presenter
maleAndmale_trailer
female_news
maleFemale_trailBlockEnd
maleFemale_trailBlockStartNext
female_ivQuiet
maleAndmale_studio
male_studioLoud
male_tele
female_tele
female_studioLoud
male_teleSoft
male_teleLoud
20
20
20
20
15
20
17
20
20
14
12
15
19
10
19
19
13
20
09
-29,1
-28,3
-23,6
-24,5
-25,7
-24,7
-28,8
-26,0
-24,4
-30,5
-23,8
-35,1
-27,3
-23,9
-24,0
-25,9
-24,5
-28,4
-24,1
CRC Leq
Post Processor -
CRC Leq
Pre Processor -
seconds
Duration in
Info
Sample number
the final dynamic processors.
-21,2
-20,1
-18,4
-18,5
-18,2
-18,1
-20,3
-18,8
-18,1
-21,3
-19,4
-23,6
-20,3
-18,4
-17,9
-18,6
-17,4
-19,3
-17,6
Table 2. Leq(R2LB) measurements from the speech channel, P1, at Swedish Radio.
In fig. 11 the measurements are plotted graphically.
20
P1 Short files Pre/Post Final Processors
P1 PRE processors
P1 POST processors
0,0
0
2
4
6
8
10
12
14
16
20
18
-5,0
-10,0
-15,0
-20,0
-25,0
-30,0
-35,0
-40,0
Sample
Fig. 11. The Leq(R2LB) short files recordings from the speech channel P1 at Swedish
Radio.
Already from this graph it is clear that the final dynamic processing does a good job
with material that is mainly speech. Fig. 12 shows the basic statistics of these data.
P1 Short files Pre/Post Final Processors
20,0
10,0
Leq(R2LB
)
0,0
-10,0
-20,0
-30,0
-40,0
Mean Short P1
StandardDev
Short P1
StandardErr
Short P1
Max distance to
Range Short P1 Max Short P1
meanShort P1
P1 PRE processors
-26,5
3,0
0,7
8,6
11,5
-23,6
-35,1
P1 POST processors
-19,2
1,6
0,4
4,4
6,3
-17,4
-23,6
Min Short P1
Fig. 12 shows the mean, standard deviation, standard error and max distance from
mean for short recordings from the speech channel P1 at Swedish Radio.
21
3.2.1.2 P3 pop music / speech channel
Table 3 shows the Leq(R2LB) results from short file excerpts both before and after
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
9b
female_studLoud
male_studEND
music_showID
music_aCappellaPunkRocker
music_EminemSmackThat
MaleAndMale_TrailerStuIv
FeMale_Studio
Music_RedHotChillipeppersSnow
Music_ShowID
Female_Studio
MusicIntro_MartinStenmarkSjumilakliv
MusicSong_MartinStenmarkSjumilakliv
FemaleAndFemale_studio
Female_studioNews
MaleAndMale_ivNews
MaleAndFemale_teleLittleStudio
MaleAndFemale_ShoutAndLaughter
Music_GainsbourgTheSOngsThatWeSing
FemaleAndFemale_Studio
Music_LiveAccordionSong
Studio_FewAppleause
female_news
20
18
20
20
20
20
18
17
15
17
12
20
21
12
07
17
09
10
18
20
05
20
-28,5
-28,1
-28,2
-29,9
-30,6
-24,8
-28,1
-25,9
-29,6
-33,2
-35,0
-28,1
-30,3
-34,9
-31,0
-30,8
-26,4
-27,6
-28,7
-24,7
-33,6
-28,4
Table 3. Leq(R2LB) measurements from the pop music and speech channel, P3, at
Swedish Radio.
In fig. 13 the measurements are plotted graphically.
22
CRC Leq
Post Processor -
CRC Leq
Pre Processor -
seconds
Duration in
Info
Sample number
the final dynamic processors.
-17,1
-18,0
-15,6
-16,7
-16,9
-15,8
-17,2
-13,9
-16,2
-19,9
-18,1
-14,3
-18,2
-19,0
-18,7
-18,9
-17,2
-14,4
-17,0
-14,7
-22,6
-17,9
P3 Short Files Pre/Post Final Processors
P3 PRE processors
P3 POST processors
0,0
0
5
10
15
20
25
-5,0
-10,0
-15,0
-20,0
-25,0
-30,0
-35,0
-40,0
Sample
Fig. 13. The Leq(R2LB) short files recordings from the pop music and speech channel
P3 at Swedish Radio.
For this channel, it is not obvious that the final dynamic processing does a good job.
The Leq(R2LB) does not indicate any major difference pre and post the final dynamic
processing. It should be noted both that the music on P3 is pre-processed using the
Replay gain algorithm before it enters the play out system, and that parts of the final
P3 dynamic processing system works at a lower ratio. Fig. 14 shows the basic
statistics of these data.
23
P3 Pre/Post Final Processors
15,0
10,0
5,0
0,0
Leq(R2LB
)
-5,0
-10,0
-15,0
-20,0
-25,0
-30,0
-35,0
-40,0
Mean Short P3
StandardDev
Short P3
StandardErr
Short P3
Max distance to
Range Short P3 Max Short P3
mean Short P3
P3 PRE processors
-29,4
2,9
0,6
5,7
10,4
-24,7
-35,0
P3 POST processors
-17,2
2,0
0,4
5,4
8,7
-13,9
-22,6
Min Short P3
Fig. 14. shows the mean, standard deviation, standard error and max distance from
mean for short recordings from the pop music and speech channel P3 at Swedish
Radio.
3.2.2 Results – long recordings
The previous results concerned the use of a short-term meter measuring selected
sound bytes. The results give us some answers to the question; are our final dynamic
processors good enough?
In real production, perhaps with an automatic pre-processor that decides what
corrections should be made to the sound file prior to broadcast, some research has to
be made on the files of complete shows.
Below is the speech channel, P1, for the same day, but this time edited so that one file
example is exactly one program. This is the way a pre-broadcast measurement device
would work.
24
3.2.2.1 P1 speech channel
Table 4 shows the Leq(R2LB) results from long file excerpts both before and after the
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
Ring P1 - Phone in show
Trailer 1
Trailer 2
Ekot10 - Short News
Trailer 3
Trailer 4
Meny - Studio show, Food
Trailer 5
Trailer 6
Ekot11 - Short News
Presenter 1 - Christoffer Murray
Trailer 7
Presenter 2
Tendens - Interview show
Presenter 3
Pre Theatre Talk studio
Theatre book reading
Presenter 4
Noon Church Bells
38.56
00.18
00.31
02.58
00.30
00.33
54.42
00.32
00.30
02.59
01.22
00.35
00.30
29.29
00.26
01.26
22.24
00.43
00.54
-26,4
-28,9
-27,5
-25,1
-28,3
-26,8
-29,6
-29,7
-24,9
-25,3
-28,6
-25,9
-29,2
-25,8
-32,7
-29,4
-30,2
-32,5
-25,8
CRC Leq
Post Processor -
CRC Leq
Pre Processor -
Duration mm.ss
Info
Sample number
final dynamic processors.
-18,7
-20,1
-18,7
-18,4
-19,5
-19,1
-20,8
-20,5
-19,7
-18,7
-20,5
-18,9
-21,1
-19,0
-22,6
-20,1
-20,6
-22,5
-17,4
Table 4. Leq(R2LB) measurements from the speech channel, P1, at Swedish Radio.
In fig. 15 the measurements are plotted graphically.
25
P1 Long files Pre/Post Final Processors
P1 long PRE processors
P1 long POST processors
0,0
100
105
110
115
120
125
130
135
-5,0
-10,0
-15,0
-20,0
-25,0
-30,0
-35,0
-40,0
Sample
Fig. 15. The Leq(R2LB) long files recordings from the speech channel P1 at Swedish
Radio.
Fig. 16 shows the basic statistics of these data.
P1 Long files Pre/Post Final Processors
15,0
10,0
5,0
0,0
Leq(R2LB
)
-5,0
-10,0
-15,0
-20,0
-25,0
-30,0
-35,0
-40,0
Mean Long P1
StandardDev
Long P1
StandardErr
Long P1
Max distance
to mean Long
P1
Range Long
P1
Max Long P1
Min Long P1
P1 Long PRE processors
-28,4
2,5
0,4
5,3
10,4
-23,1
-33,5
P1 Long POST processors
-20,2
1,5
0,3
4,3
7,1
-17,4
-24,5
Fig. 16. shows the mean, standard deviation, standard error and max distance from
mean for long recordings from the speech channel P1 at Swedish Radio.
26
3.2.3 Comparing Replay gain with Leq(R2LB)
For pop music, Swedish Radio uses the Replay gain algorithm for measuring and
correcting the loudness of imported CD-tracks, before storing the file in the central
database. Swedish Radio uses a Replay gain software implementation called A-Wave
by Fmj-Software (www.fmjsoft.com).
Can this software be used on full-length pre produced programs? Below is a
comparison between the post processing equipment at Swedish Radio and A-Waves
Replay gain algorithm.
Table 5 shows the Leq(R2LB) results from long file excerpts both after the final
dynamic processors and the same unprocessed files gain corrected by A-Wave.
27
Ring P1 - Phone in show
Trailer 1
Trailer 2
Ekot10 - Short News
Trailer 3
Trailer 4
Meny - Studio show, Food
Trailer 5
Trailer 6
Ekot11 - Short News
Presenter 1 - Christoffer Murray
Trailer 7
Presenter 2
Tendens - Interview show
Presenter 3
Pre Theatre Talk studio
Theatre book reading
Presenter 4
Noon Church Bells
38.56
00.18
00.31
02.58
00.30
00.33
54.42
00.32
00.30
02.59
01.22
00.35
00.30
29.29
00.26
01.26
22.24
00.43
00.54
-18,7
-20,1
-18,7
-18,4
-19,5
-19,1
-20,8
-20,5
-19,7
-18,7
-20,5
-18,9
-21,1
-19,0
-22,6
-20,1
-20,6
-22,5
-17,4
- CRC Leq
Replay gained files
CRC Leq
Post Processor -
Duration mm.ss
Info
Sample number
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
-31,4
-30,9
-30,2
-30,0
-30,3
-29,9
-30,9
-30,1
-29,6
-30,2
-28,3
-29,7
-28,4
-30,3
-28,5
-29,4
-28,8
-27,8
-29,8
Table 5. This shows the measurements of post processor versus the Replay gainalgorithm. Long recordings (full programs) from the speech channel P1 at Swedish
Radio.
In fig. 17 the measurements are plotted graphically.
28
P1 Long files Post Final Processors / Replay Gain calculation
P1 long Replay Gain
P1 long POST processors
0
100
105
110
115
120
125
130
135
-5
-10
-15
-20
-25
-30
-35
Sample
Fig. 17. This shows the measurements of post processor versus the Replay gainalgorithm. Long recordings (full programs) from the speech channel P1 at Swedish
Radio.
P1 Long files Post Final Processors / Replay Gain calculation
10,0
5,0
0,0
Leq(R2LB
)
-5,0
-10,0
-15,0
-20,0
-25,0
-30,0
-35,0
-40,0
Mean Long P1
StandardDev
Long P1
StandardErr
Long P1
Max distance
to mean Long
P1
Range Long
P1
Max Long P1
Min Long P1
P1 Long Replay Gain
-29,8
1,3
0,2
3,3
5,8
-27,3
-33,1
P1 Long POST processors
-20,2
1,5
0,3
4,3
7,1
-17,4
-24,5
Fig. 18 compares post processor versus the Replay gain-algorithm. Long recordings
(full programs) from the speech channel P1 at Swedish Radio. All measurements done
according to Leq(R2LB)
29
In fig. 18 it can be seen that the Replay gain algorithm as measured by Leq(R2LB)
outperforms the post processors in use today at Swedish Radio. The standard
deviation is only 1.3 dB. The important factor “Max distance to mean” shows how
poorly the worst case would end up in an automatic system for this type of material.
3.3 Conclusions of measurements
All the above measures are calculated with the use of Leq(R2LB), which is the
proposed ITU standard BS.1770. According to Skovenborg/Nielsen’s research, this
type of measure is subjectively the best in terms of listener satisfaction.
The effect of introducing measurements to files prior to broadcast can be seen in fig.
16. This figure displays the values measured from complete sound files from the talk
channel. The maximum distance to the channel’s mean loudness is the maximum
error from the listeners’ point of view. If an automatic levelling system processed the
full radio programme file, this figure shows how big the maximum error in dB would
be for the listener. Using this measure, the maximum error would be 4.3 dB compared
to a Leq(R2LB) measurement. In the measurement with shorter excerpts from the
speech channel (see fig. 12) the maximum error is 4.4 dB. The Replay gain algorithm
was slightly closer to the mean value with a maximum distance of 3.3 dB, see fig. 18.
The Final processors at the Swedish Radio do even out the loudness. Previous
research also tells us that pre measurement using Leq(R2LB) evens out the loudness.
An implementation of equal loudness at Swedish Radio could be firstly measurement
with Leq(R2LB) meters at production, secondly pre measurements of files using
either Leq(R2LB) or Replay gain and thirdly the same final processing that is already
in use today.
The pop music channel, P3, did not show much difference in measured Leq(R2LB)
pre or post the final processing, see fig. 14. The maximum error to mean is 5.7 for the
pre process and 5.3 dB for the post process measurement. All music channel
measurements were done with short excerpts. The music channel would probably gain
in equal loudness if the final processors were used more aggressively.
30
4 Workflows and levels at the Swedish Radio
4.1 The pre-digital era
Before digital media became available, most recordings at the Swedish Radio were
made (in the very early days) on shellac vinyl or steel tape and later from about 1950
on ¼ inch open reel tape. I have been assessing a great amount of recordings made at
different times through my collaboration with Swedish Radios archive channel, SR
Minnen. The oldest recordings date back to the 1940’s and the newest are from 2007.
The early recordings have a much narrower bandwidth. The only treble that exists is
from hiss and cracks from the shellac records. During the Second World War the
recordings were mainly done on 800 recycled ¼ inch open reel tapes. The recordings
that were to be kept in the archives were transferred to shellac records. In the early
nineteen-fifties, Swedish Radio began to save open reel tapes in the archive. The
bandwidth and the dynamics of the recordings increased, and hence the sound quality
improved. The dynamics were quite large at that time, presumably due to less overall
control and no availability of dynamic processing during the recordings.
There was still a place in the living room where the family spent time together and
where listening was done in full concentration, close to the radio. Radio theatre shows
from this time are dynamic. They are totally unusable today without remastering, gain
riding or heavy compression, as preparation for usage in an iPod or retransmission on
the digital archive channel. But the shows probably worked quite well so long as the
radio was the main focus.
Soon there was mobile listening in cars and small plastic radios. Gone was the living
room with full-concentration listening. Now the radio had to compete with a growing
number of loud sources. All radio broadcasts have a maximum modulation level. If
your transmitters level exceeds a maximum level of modulation it will interfere with
adjacent radio frequencies. The solution to control transmission levels is the
compressor. It was also heavily used when recording shows.
Around 1960, other channels started their transmissions (Radio Nord) and something
had to be done to maintain the competitiveness of public service radio. Soon new
31
public service radio channels were on air, and these had much less dynamics,
probably because everyone looked to the U.S., where the radio commercials drove the
need for high listener ratings. The thesis was formulated that a louder station caught
more listeners, and so the race for more compression was on. Since the compressor
began its days in radio, not much has changed in the production process, from the
loudness perspective.
4.1.1 Methods and workflow – pre digital era
•
Recording in the field to 1/4 inch open reel tape, later in time recording to
DAT.
•
Editing by cutting and splicing the original tape, or an open reel copy if the
original recording was on DAT.
•
Correcting levels manually during the copying process. An audio engineer
copies the edited tape, constantly moving the fader to counteract differences in
loudness. At this stage compression and equalisation may also be applied. At
this stage the copy could have been made to either 1/4 inch open reel or later
to DAT tape. The sound levels are controlled using a PPM EBU/Nordic bar
graph. Usually a RTW meter. For P1 the maximum levels touched +6 dB (0 =
-18 dBFS), for P3 the maximum levels were allowed to peak at +9 dB.
•
Analogue broadcast were constantly controlled in the FM-continuity
(Swedish: Programkontroll). Manual level control was applied so that
adjoining radio shows could be heard without the need for the listener to reach
for the volume control on their radio. This last part was preferably done with
the same audio engineer controlling the flow for as many continuous hours as
possible.
•
The A/D conversion is done with 18 dB headroom to Full Scale (the music
channel P3 is converted utilising 22 dB headroom).
•
Digital Final processing at the Master Control of Swedish Radio. This
processing consisted of a limiter as the last precaution before leaving the
signal for final transmission. Swedish Radio had full control of the settings in
32
these processors even if they physically were placed close to the FM
transmitters, operated by Televerket Radio (new name: Teracom). The control
of these processors has always belonged to Swedish Radio.
•
Today the processors consist of two units in series. First there is an automatic
gain controller, jointly developed by Swedish Radios Torbjörn Wallentinus
and the company Factum Electronics. The Cadenza was developed during a
ten-year period of time. It began with an analogue prototype and later it
became a digital DSP based unit. The second processor in the row is the multi
band processor Orban 8200. Today, this unit is used mainly as a top limiter.
These processors are used on all Swedish Radio’s FM channels except the
classical music channel P2 which instead uses a processor from Omnia. The
web channels solely use the Cadenza for levelling.
The sound level measuring device at the Swedish Radio has been a PPM meter using
the EBU/Nordic Scale.
4.2 Digital era
Here, we define the digital era as the era beginning with non-linear editing with
computer software. The workflows below are presented graphically in section 10.
4.2.1 Methods and workflow – pre-produced in the digital era
•
The journalist uses a compact flash or hard disk recorder. The recordings are
done in either uncompressed PCM 16-bit 48kHz stereo files or directly
encoded into MPEG I Layer II at 384kbit/s.
•
At the radio station, the recorded sound files are copied into the sound
managing system DiGAS (by DaVID Gmbh in Munich). If originally recorded
in PCM, they are at this point converted to Layer II 384 kbit/s.
•
Editing is done mainly in DaVID’s Multitrack editor where plug-ins for
dynamics and equalization are applied. This editing is often done on the
ordinary office computer without functional loudness meters.
33
•
The final mix is often made together with an audio engineer using a PPM
EBU/Nordic Meter. Sometimes the final mix is done by the journalist using a
PPM EBU/Nordic Meter in a small pre-production studio.
•
The final mix is then placed in the play list by an engineer who also makes
sure that the levels are correct by using his or her ears.
•
The broadcast server automatically transmits the packaged radio.
•
The final processors are the automatic gain controller Cadenza by Factum
Electronics followed by a multi band 8200-processor by Orban.
•
The signal is fed to the transmitter operator, Teracom, via digital J.57 linear
PCM circuits. In principle, no gain shift should occur through this chain.
•
The signal is also fed through a break-out box to web-feeds, reference
recordings etc.
4.2.2 Methods and workflow – live radio in the digital era
The sound engineer mixes the live show with the help of a Nordic/EBU PPM meter.
•
Sometimes the pre-recorded sound segments are evaluated prior to broadcast
utilising a Nordic/EBU PPM meter.
•
The final mix is transmitted to the central apparatus room.
•
The final processors are the automatic gain controller Cadenza by Factum
Electronics followed by a multi band 8200-processor by Orban. The latter is
used mostly as a top limiter.
•
The signal is fed to the transmitter operator, Teracom, via digital J57 linear
PCM circuits.
•
The signal is also fed through a break-out box to web-feeds, reference
recordings etc.
34
4.3 The future – method and workflows
4.3.1 Production monitor sound levels?
As can be seen in fig. 3, the ear does not have a linear frequency response at differing
sound levels. How loud should the production environment be?
The sound production facility should be listening at the same intensity as the typical
listener. The typical listening levels for TV in actual homes is 60 dBA [Benjamin,
2004].
With the new patterns of broadcast consumption this level is not enough. The listeners
use both headphones and speakers. And each listening device has its own preferred
sound level, see fig. 19.
Fig. 19. Preferred listening levels for different groups of employees at the Danish
Radio & TV. 1 Administration (non-engineer) 2 Journalists (non-engineer) 3
Classical music engineers 4 Pop/Rock music engineers 5 Noise engineer [Brixen
(2001)].
35
Many believe that there never is a correct volume to listen at during production. One
must listen at all levels, both headphones and loudspeakers. If it is a passage where
music or sound effects is layered with the speech, this part should be test listened in at
least two different configurations.
4.3.2 How to implement fully controlled loudness levels in an
automatic broadcast
Swedish Radio uses the sound production system DiGAS by DaVID Gmbh
[http://www.david-gmbh.de]. This system contains all parts of modern radio
production:
•
DBM is the database manager. Contains all sound files and all their metadata.
These metadata contains among other things the play out volume.
•
Multitrack is the sound editor.
•
BCS is the play out backbone database system. It contains all data regarding
the delivery of the files.
•
Digairange is the front end of the BCS. It is used in packaging the pre
produced sound files. This tool is also used for recording of speech between
adjacent sound files (voice tracks).
•
Digaroc is the play out part of the BCS. It handles play out of both pre
produced shows and items in a live show.
Automatic loudness control can be used in Swedish Radios workflow in two ways;
1. As an automatic loudness leveller early in the production flow.
2. As an automatic loudness leveller processing of complete sound files or their
metadata prior to play out.
36
4.3.3 Manually calculating parts of a show
The journalist records in the field without absolute control of the sound levels. The
journalist’s focus should be on the interviewed person. The two important aspects for
the journalist level wise is digital overload and a recording made at too low level.
A topic of discussion relating to loudness is that of True Peak problems. This area of
audio technology studies how an almost full-scale digital signal can distort. [Lund &
Nielsen, 2004] As a precaution the sound level for MPEG-1 Layer II 384kbit/s is
recommended to never exceed -3 dBFS on the digital peak meter on the recording
device.
When the journalist returns from the field, all the recorded sounds are transferred into
the DiGAS database. The show item is edited in the Multitracker or appropriate sound
editor. During the editing process the show item is levelled with a short term meter.
Later a button can be clicked to calculate the Leq(R2LB) of the complete show item.
The usage would be: save the file (bounce) from the Multitracker, click on the
loudness button and the appropriate sound level correction would be noted in the files
metadata.
When this file later is planned for play-out inside a show, the sound level correction is
automatically transferred to the planner Digairange, to the play out database BCS and
to the delivery system Digaroc.
This system could e-mail its correction levels back to the journalist. This would in
turn allow the journalist to improve her level usage in future productions. This would
also warn the journalist and let the journalist manually override the automatic level
settings.
4.3.4 Full sound file levelling
The levels can also be automatically set volumes as they are imported in to the
planner Digairange. If a show item is planned a computer software could easily be
37
programmed to measure the sound files Leq(R2LB) and then set the play out level
according to the measurements correction gain value.
38
5 Real life meter usage
I have been using the BBC Leq(R2LB) meter
together with a mixing engineer in
production for over a year. How well does it
perform in real life production? Below are
some comments of situations where the
Leq(R2LB) measure does not function as expected.
The meter is set up in loudness mode. It is calibrated so that normal speech keeps the
needle close to 4.
Peaks are controlled with a security limiter post metering.
We biased our readings to make sure the same loudness was achieved, see table 6.
Situation
Actual reading to obtain equivalent
loudness
Normal speed speech
4
Fast speech
3
Close voices (proximity effect)
3
Telephone hybrid
5
Old recordings (from the 1940’s)
4
Modern music without speech
3
Table 6. Biasing meter readings to accomplish equal loudness over different
programme material.
39
40
6 Discussion concerning a new meter
In the above description it can be seen how the Leq(R2LB) meter functions in real
life. The meter can be easily fooled. A user must judge the meter readings. Could we
possibly build a meter that did this for us? It is out of the scope of this report, but I
believe a meter could be built to work better for non-technical staff or an automatic
levelling system.
First of all, the meter should classify the material. Is it music, speech or telephone? Is
someone too close to the microphone? Is it broadband sound from a recent recording
or is it an old narrowband recording. Is it a quick talker? Is it a dynamic talker? Is the
measured signal pre-limited?
All the above-mentioned different types of sound can after classification be biased
before measurement so that the meter becomes a help in loudness measurement. This
meter is hereafter called category meter.
The idea of a category meter has some resemblance of measuring with the help of
critical bands, but in a less universal implementation. In turn, with less universality
the meter might improve its exactness. The new category meter could probably be a
meter with a needle that constantly should stay in a well-defined space. The
implementation could also be used to bias the reading if for example the broadcaster
chooses to suppress the music volume.
41
42
7 Other comments closely related to loudness in
broadcasting
7.1 Balance speech/music
During the spring 2002 I worked as a audio engineer at Swedish Radios young music
and speech channel, P3. Usually for morning and day shows we mixed speech 3-6 dB
above the highly limited CD music. But in the evening there is a show containing
Live Music, P3 LIVE. We all understood that these live listeners sitting at home with
the volumes turned up, when the short speech segments were played out they would
blow their ears away. We usually fixed this by mixing the speech comparatively much
softer in these late live shows.
7.2 Classical music
Classical music cannot be measured and automatically processed in the same manner
as speech or pop music. Classical music is dynamic, and the listeners often want this
wide usage of dynamics to be intact the whole way to the receiver. It can be argued if
this will be true in the future with a new generation of listeners and their new listening
devices. Can an iPod user on the bus in the morning rush hours really hear the quiet
passages in classical music without turning up the volume, and at the same time
turning up the peaks in the music. Maybe the iPod’s of tomorrow will include a
compressor for these types of dynamics problems.
7.3 Different meter usage between channels
P1’s audio engineers let the speech peaks after compression touch the +6 dB mark on
the PPM EBU/Nordic meter. P3’s engineers let these peaks touch +9 dB instead. At
the Swedish Radio’s central apparatus room the P3 is turned down 4 dB relative P1.
This discrepancy was introduced at P3 when they during the seventies installed
43
mixing desks with 22 dB head room instead of the old desks with a head room of 18
dB. This difference between the channels works fine as long as it is a simple FM
transmission. The problems arise later in the production.
The unprocessed P1 and P3 signals are recorded in the central apparatus room for
later reprise use. If this signal is fed back a second time in the broadcast chain, the P3
signal now is 4 dB even lower, a total of 8 dB lower.
A different problematic situation is when different channels use each other’s material
live. For example, the archive channel SR Minnen is set up as P1 with reference level
of -18 dB equivalent to test tone in studio at TEST level (PPM EBU/Nordic 0 dB).
Alternative transmission if the main play out computer breaks is the classical music
channel P2. P2 too uses a different reference level, being -15 dB equivalent to test
tone in studio at TEST level (PPM EBU/Nordic 0 dB). When the broadcast is
redirected to use P2 instead of SR Minnen some listener almost always phones in
reacting to this massive volume increase (apart from complaining about the missing
main programme).
44
7.4 Different dynamics for different usages
Today transmission of material from Swedish Radio is done to many listeners in
many different ways. Table 7 gives examples of all the different forms of distribution,
and their most usual listening preference.
Distribution Form
Assumed Listening Device
FM transmission
Car stereo, Small speaker radio, Living
room stereo system with large speakers
and subwoofer
AM transmission
Advanced Small speaker radio
DAB transmission
Car stereo, Small speaker radio, Mobile
telephone’s hands free and loudspeaker,
Living room stereo system with large
speakers and subwoofer
Web streaming transmission
(cheap) Computer speakers, Soon mobile
telephone’s hands free and loudspeaker
Web On Demand transmission
(cheap) Computer speakers
Podcast
Earphones
Table 7 describing different distribution forms and different listening devices.
These different listening situations demand differing dynamic ranges. Dynamic range
is the distance between the softest and loudest sound in a recording, usually measured
between quietest and loudest speech level. Thomas Lund at TC Electronics has
studied varied broadcast consumption. In fig. 20 he shows how the differing
consumption patterns need different dynamic ranges.
45
Fig. 20. Dynamic Range Tolerance for consumers under different listening conditions
[Lund, 2006]
Consumption of media is today done at almost any location. Different locations have
varying background disturbances, see table 8.
SPL A weighted
SPL C weighted
Living Room, Suburban
45 dB
Living Room, Urban
55 dB
70 dB
Inside Car
65 dB
85 dB
Inside Jet
75 dB
90 dB
Walk in Traffic
80 dB
92 dB
Subway
90 dB
100 dB
Table 8. Typical surrounding noise levels measured by Lund. All environments are
realistic for broadcast consumption today.
46
8 Acknowledgment
I would like to thank the following people in helping me with this paper
•
My boss at Swedish Radio, Mikael Cohen, for discussions and help with
sound levels. He taught me the importance of even sound levels, mainly in
documentary production.
•
My professor, Sten Ternström, at Kungliga Tekniska Högskolan in
Stockholm. He has opened the world of research for me, and helped with my
English. He is also really good at deciding deadlines – and forgiving when I
missed them. Without these this paper would never have been completed.
•
My supervisor at Swedish Radio, Lars Jonsson, for all help with my texts. I
would also like to thank for all valuable contacts both inside Swedish Radio,
and outside in the world of research.
•
Pelle Holmquist in the central apparatus room for explaining the final dynamic
processes and also letting me record pre-and post-process signals.
•
Fredrik Nilsson. A sound engineer who has helped me evaluate the BBC
loudness meter.
•
Björn Melander, my tutor in the studio. Produces relaxation music with
relaxing speech. Taught me the neurotic behaviour of million level breakpoints
in every production to ensure even loudness, very important for relaxation (to
the listener, not the engineer).
•
Gösta Konnebäck. The grand old man in radio. He opened my eyes to the
world of radio in my early teens.
•
My wife Naomi and my two kids Hannah and Maya for helping me test
loudness. Every morning and evening I turned the volume of the TV set
slowly down, noting where they started complaining. At the same time I
measured the dynamic content of the cartoons. Hannah and Maya also helped
me with the yellow marking pen at the time of reading all the previous
research.
47
48
9 Glossary
AES/EBU – A standard for digital sound, both dataformat in the cable and the cables
connectors.
DAT – Digital audio tape. A digital tape system. Not used in production anymore.
DiGAS – A broadcasting system consisting of a sound database, recorder, editor,
multitrack mixer, play-out system.
Master Control at Swedish Radio – The room at Swedish Radio where all the
nationwide signals are collected before transmission to Teracom, the FM-distributor.
In this room the final processing units are positioned.
Metadata – Extra data associated with a sound file such as production number, title,
length and most importantly for loudness purposes; sound play-out volume correction.
By writing this figure in metadata instead of recalculating the file saves processing
power and leaves the sound file untouched.
Nordic PPM – The meter used at the time of this writing at Swedish Radio.
Teracom – The FM-distributor used by Swedish Radio. Swedish Radio feeds the
master signal to them for further distribution.
49
50
10 Production Flowchart for Swedish Radio
51
52
11 References
Akustik Technologie Göttingen, 2004. Webpage
http://www.akutech.de/mainpage/psychoa.htm mentioned in Skovenborg and
Nielsen, “Evaluation of different Loudness Models” AES convention paper
2004.
Backus, John (1977) “The Acoustical Foundations of Music”, 2nd Ed, W W Norton,
New York.
Benjamin E (2004) “Preferred Listening Levels and Acceptance Windows for Dialog
Reproduction in the Domestic Environment”. 117th AES convention, San
Fransisco. Preprint 6223
Bonello (2007) “Multiband Audio Processing and Its Influence on the Coverage Area
of FM Stereo Transmission” JAES March 2007
Brixen (2001) “Report on Listening Level in Headphones”. Document KKDK-06801-ebb-1 for the Danish Radio, Copenhagen.
Brixen (2001:2) “Audio Metering”, Broadcast Publishing & DK Audio A/S, Denmark
Hyperphysics (loud) Internet; http://hyperphysics.phyastr.gsu.edu/hbase/sound/loud.html
Hyperphysics (place) Internet; http://hyperphysics.phyastr.gsu.edu/hbase/sound/place.html
Hyperphysics (eqloud) Internet; http://hyperphysics.phyastr.gsu.edu/hbase/sound/eqloud.html
ITU-R (2006) proposition for BS.1770, from Srg-List at yahoogroups.com
2006-04-26
53
Lund, Thomas & Nielsen, Søren H. (2003) “Overload in Signal Conversion”.
Presented at 23rd AES International conference, Copenhagen, Denmark,
2003-05-23.
Lund, Thomas (2006) NAB BEC Proceedings.
Lund, Thomas (2007) Presentation at AES 2007-05-05. Vienna. Privately mailed
keynote.
Moore 1982, “An introduction to the Psychology of Hearing”
ReaplayGain (2001) http://replaygain.hydrogenaudio.org/
Sengpielaudio Internet; www.sengpielaudio.com/calculatorSonephon.htm
Skovenborg & Nielsen (2004) “Evaluation of different Loudness Models”, AES
convention paper.
Skovenborg (2007) Private mail to the author of this text 2007-04-24
Ternström, Sten, (2002) “Ljud”, (2. ed.), KTH
Todd, Craig (2007) In e-mail to the maililnglist Srg3-List at yahoogroups.com
2007-03-23
Zwicker E. & Fastl H. (1999) "Psychacoustics: Facts and Models" (2. ed.), Springer
Series in Information Sciences, 22, Berlin: Springer-Verlag.
Zwicker E. & Flottorp G. & Stevens S. S. (1957) “Critical bandwidth in loudness
summation.” J. Acoustical Society of America, 29(548).
54

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project