R&D White Paper

WHP 051

October 2002 (revised July 2004)

Audio Description:
what it is and how it works

N.E. Tanton, T. Ware and M. Armstrong

Research & Development

BRITISH BROADCASTING CORPORATION
 BBC Research & Development
White Paper WHP051

Audio Description: what it is and how it works

Nick Tanton, Trevor Ware & Mike Armstrong

Abstract

For understandable reasons many television programmes rely on visual content and
composition to help to tell their story. The provision of additional description of the scene or
action can therefore provide a considerable benefit to understanding, especially to the visually
impaired.
Audio description (AD) is an ancillary component associated with a TV service which delivers a
verbal description of the visual scene as an aid to understanding and enjoyment particularly (but
not exclusively) for viewers who have visual impairments.
This White Paper describes the user requirements for audio description, how audio description
can be implemented for digital television and how it works for digital terrestrial television (DTT).
It also briefly describes two particular practical implementations.

Additional key words: Access services, accessibility, DTV,

Document revision history

v1.0 2002 original

v2.0 2004 added explicit specification for practical pan characteristic,
updated implementation section and minor additions

© BBC 2004. All rights reserved.

 White Papers are distributed freely on request.
Authorisation of the Chief Scientist is required for
publication.

© BBC 2004. All rights reserved. Except as provided below, no part of this document may
be reproduced in any material form (including photocopying or storing it in any medium by
electronic means) without the prior written permission of BBC Research & Development
except in accordance with the provisions of the (UK) Copyright, Designs and Patents Act
1988.
The BBC grants permission to individuals and organisations to make copies of the entire
document (including this copyright notice) for their own internal use. No copies of this
document may be published, distributed or made available to third parties whether by
paper, electronic or other means without the BBC's prior written permission. Where
necessary, third parties should be directed to the relevant page on BBC's website at
http://www.bbc.co.uk/rd/pubs/whp for a copy of this document.
What is Audio Description ?
For understandable reasons many television programmes rely on visual content and pictorial
composition to help to tell their story. The provision of additional description of the scene or action
can therefore provide a considerable benefit to understanding, especially to the visually impaired.
Audio description (AD) is an ancillary component associated with a TV service which delivers a
verbal description of the visual scene as an aid to understanding and enjoyment particularly (but
not exclusively) for viewers who have visual impairments 1 & 2.
The description is typically confined to gaps in the normal programme narrative. The opportunities
to describe a scene are therefore dependent on the programme genre and on the editing of the
main programme sound. Some programmes are naturally more suited to description than others.
For example news programmes provide little opportunity for description and anyway tend to be
self-documenting. Similarly the presenters of some (but not all) cookery programmes provide a
seamless uninterrupted description of what they are doing on screen which precludes significant
opportunities for description.
Science and informative programmes, on the other hand, tend to include relatively long gaps in the
narrative which are associated with significant visual content; in such cases the gaps provide
ample opportunity to fully describe the concurrent visual images.
At the other extreme, action drama and “soaps” are edited more tightly and typically provide
predominantly brief windows in the dialogue which allow only concise description events. Because
drama and soaps make significant use of purely visual imagery for dramatic purposes AD is
particularly beneficial despite the briefness of the description windows.
In any programme there may be long periods during which there is no suitable opportunity for
description. Any means of delivering AD must therefore accommodate extended periods of
description silence (representing near-continuous narrative in the main programme sound) and so
should provide the user with means of confirming that, under these conditions, description silence
does not necessarily imply failure in delivery of the service or in the receiving equipment.

1
Readers of this note who have yet to be persuaded of the benefits of AD are invited to try fully comprehending an
episode of a soap or drama or of a science and features programme (such as “Horizon”) with their eyes closed.
2
Audio description became available with some US analogue terrestrial channels more than a decade ago. A brief
and small-scale UK experiment (called Audetel) using CELP-coded audio conveyed as teletext data was led by the
ITC in the early 1990s. This further demonstrated the value of such a service and, en passant, exposed several
practical issues and user requirements.
Audio Description – determining the requirements
An ancillary service such as AD will not suit all viewers so the first requirement is that it is provided
as a “closed” or “elective” system where the user elects whether or not to hear the description.
This is directly analogous to “closed captioning“ or subtitles.
Those gaps in programme narrative which present opportunities for description will nevertheless
often include sound effects or music which can make an added description hard to discern.
Furthermore the loudness of programme sound during these description opportunities will vary
from one gap to the next. An important requirement of AD is therefore to be able to adjust the
relative level of programme sound and description in the mix which the AD user hears and,
significantly, to be able to adjust this relative level on a passage-by-passage basis. Determining
the appropriate depth of this fade is best be done by the programme maker under controlled
conditions typically using the final transmission copy of the programme at the stage when the AD
component is being authored.
Individual AD users will have different aural acuity, different describers will have their own style of
vocal delivery (voice pitch and timbre), several voices may be used to describe a single
programme and there are, in practice, differences in audio signal level for different home receivers.
This makes it very desirable for the AD user to be able to make minor adjustments to the volume of
the description signal to suit his or her condition3.
Description content is voice only. It thus makes sense to assume that the description signal need
not have the same audio bandwidth as programme sound and that, where appropriate, a separate
description signal can be conveyed as mono rather than stereo – this saves bandwidth or bit-rate.
The combined programme sound and description signal as heard by the AD user will however
need the normal bandwidth appropriate accurately to convey the programme sound.
As noted above there will sometimes be considerable intervals between successive description
passages. The AD user who has selected a described channel during one of these gaps will
therefore find it hard to determine from the audio itself whether the temporary absence of
description is intentional or is the result of a fault in scheduling, transmission or decoding. There is
a strong requirement for the user to be able to confirm that AD is being transmitted as scheduled,
preferably using an audible indication since many AD users will have visual impairments.

In summary the user requirements for AD are

• a closed system,
• ability to adjust relative volume of description and
• ability to promptly determine that a programme is currently being described.

From the service provider/multiplex operator’s viewpoint the requirements should include
• bandwidth or bit-rate frugal delivery of the service,
• a delivery mechanism that uses existing & open standards (e.g. ISO/IEC 13818-x, DVB etc.),

Additional desirable features for AD decoder implementations include having separate hi-fi and
VCR outputs 4 and providing an output for headphones should the AD user wish to listen in the
company of others who do not wish to hear the description.

3
A similar clear requirement for user control of description level was been identified in a recent (2003) usability study
of “spoken subtitling” conducted by SVT in Sweden.
4
Many set-top boxes for example already have separate SCART connections for the TV and VCR.
Audio Description on Digital Terrestrial Television
Digital television offers considerable flexibility for the delivery of new types of service or of
additional service components; it therefore provides an excellent opportunity for adding AD to
appropriate television services.
The deliverable bit-rate for each platform (satellite, terrestrial or cable) is not, however, limitless
and bit-rate efficient methods of delivery are important. This is as true for the vision component of
a digital television service as for subtitles or AD; thus service providers & multiplex operators all
take steps (such as statistical multiplexing and opportunistic data insertion) to ensure that bit-rate
is used effectively. Of the three platforms DTT has the least deliverable bit-rate per multiplex and
the following text in this note describes the mechanism by which AD is coded, signalled and
decoded for DTT 5. The technique 6 could equally be applied to DSat and to DCable.
One of the practical conclusions of the ”Audetel” project was that means should be provided of
controlling the level of the main programme sound during a description passage and restoring it
when the description was over. The level of the AD and overall level of main mixed with
description would preferably be under the control of the listener. The principles are shown
diagrammatically below in figure 1.

user control of
description volume

decoded audio
description
mono

R
decoded main
programme
stereo
L

programme provider control of user control of

programme volume overall volume
during description passages

fig 1 : functionality of AD processing

To support this functionality there is thus a need for the programme provider to signal to the
receiver/decoder so as to fade the main programme sound to a suitable level when the
accompanying AD is active. For practical reasons this signalling is best done by embedding the
information in the AD stream, thereby ensuring appropriate timing and conveniently binding the
control signal to the component itself. As the level of fade depends on the level of the programme
sound during each particular description passage the fade value needs to be adjustable rather than
simply “on” or “off”.

Fading the programme sound

In UK DTT the transmitted fade instruction is an unsigned byte value, 0x00 representing 0 dB,
each increment representing a nominal 0.3 dB, 0xFE therefore representing approximately –77 dB
whilst the fade value 0xFF represents completely mute programme sound. The programme
provider is normally able to ensure standard signal levels in the description authoring process and
throughout studio infrastructure and, where gaps in the narrative permit, subtle stepped fades are
possible. This obviates having to have very conspicuous “crashes” in and out of a description

5
The provision of audio description with digital terrestrial television services was mandated in the UK Broadcasting Act
1996. The UK Communications Act 2003 sets out a framework for providing AD on digital satellite and cable
platforms as well as DTT.
6
This is called “receiver-mix AD” and is documented in ETSI document TR101 154 v1.5.0 (2004-01). AD broadcast as
a pre-mixed combination of programme sound and description is called “broadcast-mix AD”.
passage when the corresponding programme sound is very loud. If the opportunity to describe is
only short the fade and consequent recovery may however need to be very prompt.

description

programme
sound after fade

programme
sound before
fade

0x22 = 10 dB 0x28 = 12dB

fade value
0x00 = 0dB

figure 2 – principles of fading programme sound during description passages

Figure 2 above illustrates these principles using real waveforms from a described BBC
programme.
During the first description passage (leftmost) the programme sound requires only 10 dB of fade to
avoid obscuring the description but the amount of description required to explain the scene means
that the fade must be abrupt.
During the second description passage (right) more attenuation of the programme sound is
required but there is less to describe; a more gradual fade in and out of the description passage is
therefore possible.

Panning the description

As the means of embedding this control data in the compressed audio stream leaves some small
bit-rate resource unused, a second control value (pan) is also transmitted. This allows the
decoded AD signal to be panned within the sound stage of the main programme sound. Pan
control enables the programme maker to place the “describer” at any preferred horizontal position
within the sound field (in the same way that speech from out-of-vision commentators is sometimes
placed to one side of the stereo image).
As with the fade, transmitted pan is a byte value, 0x00 representing centre front, each increment
representing about 1.4° clockwise looking down on the listener (see figure 3 below).
For stereo the pan value will be restricted to ±30° of the centre front (i.e. to the range 0xEB..0xFF
& 0x00..0x15) but the syntax of the signalling allows for any future use in which an AD component
might be provided with a surround-sound main programme audio.

limits for
stereo

pan = 0x00
CENTRE

0xEB 0x15

(front) LEFT (front) RIGHT

30°

110°
0xC0 0x40

(rear) LEFT (rear) RIGHT

0xB2 0x4E

0x80

figure 3 : interpretation of audio description pan value

(seen from above the listener; includes mapping onto multi-channel sound presentation)

AD pan and line up levels

UK television broadcasters use line up levels of -18 dBfs for all audio signals, whether mono or
stereo, with peak level at -10 dBfs (otherwise known as M6 line up where mono signal is derived
by the formula M = (A + B) - 6 dB). To avoid incompatibility and clipping, AD signals need to be
recorded with the same line up as the main audio channels.
In the main audio central speech usually peaks up to -10 dBfs in both channels, but if the voice is
panned to one channel it can still only peak to -10 dBfs in that channel. Thus to retain compatibility,
the pan law for the mono AD signal needs to operate by attenuating one or other channel whilst
keeping the gain of the other channel constant.
The line up of an AD decoder can be summarised in the following requirements.
• Reference level on the main stereo channels and reference level on the AD channel should
both appear at the same level on the output when the AD is panned centrally and the user
mix adjustment is at the default (0 dB gain) setting.
• Reference level on the AD channel should appear at the same level as a centrally panned
signal in one channel and be attenuated by at least 20 dB in the other channel when the
description is fully panned to one side and the user mix adjustment is at the default setting.
• Throughout the pan from left to right channel, reference level in the AD channel should
appear at the same level in the louder channel and at a lower value in the other.

AD pan law 7
To implement the AD pan in a way which tracks the 1.4° steps across the stereo image, the
required relative levels for this movement can be calculated from the “stereophonic law of sines8”.
For loudspeakers placed at ±30° this gives the formula
sin α = (A – B) / 2.(A + B) where α is the angle from centre.
Therefore for the right half of the stereo image (where 0° ≤ α ≤ 30°) the required levels of the two
channels become
A = M.(1 + 2.sin α) / (1 – 2.sin α) and B = M
and for the left half of the image
L = M and R = M.(1 – 2.sin α) / (1 + 2.sin α) .
Annex 1 shows the appropriate levels for each value of pan byte from 0xEB - 0xFF & 0x00 - 0x15.
For a practicable implementation of AD pan, a reasonable match to these values can be obtained
by attenuating the signal in one channel at a rate of 1dB per pan value. This gives an accuracy of
better than ±1dB over the range of ±20° and a maximum attenuation of 21 dB which is sufficient to
move the position of the sound into the louder of the two speakers [1].

left law of sines

right law of sines
AD pan laws compared
left 1dB steps
right 1dB steps
12.000

6.000

0.000
level dBs

-6.000

-12.000

-18.000

-24.000

-30.000
0xEB 0xF2 0xF9 0x00 0x07 0x0E 0x15
Pan Byte Value

7
The pan law described here supercedes other strategies previously suggested (such as constant power).
8
Quoted from “Phasor Analysis of Some Stereophonic Phenomena” B.B. Bauer, JASA, vol 33, no. 11, 1961.
Signalling the presence of audio description
The MPEG-2 System syntax defined in ISO/IEC 13818-1 [2] provides an optional simple fixed
length field for PES_private_data with a total capacity of 16 bytes (128 bits) per packetised
elementary stream (PES) packet.
Fade and pan control information is therefore coded in PES_private_data within the PES
encapsulation of the coded AD component.

The structure and syntax of this field are as follows.

AD_descriptor(){
reserved 1111 4 bslbf
AD_descriptor_length 1000 4 bslbf
AD_text_tag 0x4454474144 40 bslbf (5 bytes)
revision_text_tag 0x31 8 bslbf
AD_fade_byte 0xYY 8 bslbf (FADE byte)
AD_pan_byte 0xYY 8 bslbf (PAN byte)
reserved 0xFFFFFFFFFFFFFF 56 bslbf (7 bytes)
}

The semantics are as follow.

AD_descriptor_length : the number of significant bytes following the length field (i.e. 8).
AD_text_tag : a string of 5 ASCII characters forming a simple and unambiguous
means of distinguishing this from any other PES_private_data.
A receiver which fails to recognise this tag should not interpret this
audio stream as audio description.

revision_text_tag : the AD_text_tag is extended by a single ASCII character version

designator (here “1” indicates revision 1).
Descriptors with the same AD_text_tag but a higher revision
number shall be backwards compatible with this specification – the
syntax and semantics of the fade and pan fields will be identical but
some of the reserved bytes may be used for additional signalling.

AD_fade_byte : takes values between 0x00 (representing no fade of the main

programme sound) and 0xFF (representing a full fade)
Over the range 0x00 to 0xFE one lsb represents a step in attentuation
of the programme sound of approximately 0.3dB giving a range of
about 77 dB. The fade value of 0xFF represents no programme
sound at all.
The rate of signalling and the expected behaviour of a decoder to
changes in fade byte are described below.

AD_pan_byte : takes values between 0x00 representing a central forward

presentation of the audio descriptor and 0xFF, each increment
representing a 360/256 degree step clockwise looking down on the
listener (ie. just over 1.4 degrees, see figure 3).
The rate of signalling and the expected behaviour of a decoder are
described below.

reserved : the remaining 7 bytes are set to 0xFF and reserved for future
developments if and when required.
A PES-packet from an audio stream [3] carrying audio description will therefore typically
commence thus :
packet_start_code_prefix 0x000001 24 bslbf
stream_id 0xC0 (0xC0-0xDF ≡ audio streams)
PES_packet_length 0xYYYY (as appropriate)
'10' 10 2 bslbf
PES_scrambling_control YY 2 bslbf (as appropriate)
PES_priority Y 1 bslbf (as appropriate)
data_alignment_indicator Y 1 bslbf (as appropriate)
copyright Y 1 bslbf (as appropriate)
original_or_copy Y 1 bslbf (as appropriate)
PTS_DTS_flags 10 2 bslbf (PTS present)
ESCR_flag 0 1 bslbf
ES_rate_flag Y 1 bslbf (as appropriate)
DSM_trick_mode_flag 0 1 bslbf
additional_copy_info_flag 0 1 bslbf
PES_CRC_flag Y 1 bslbf (as appropriate)
PES_extension_flag
* 1 1 bslbf
PES_header_data_length 0x10 8 uimsbf (16d)
'0010' 0010 4 bslbf
PTS[32..30] YYY 3 bslbf (as appropriate)
'1' 1 1 bslbf
PTS[29..15] YYYYYYYYYYYYYYY 15 bslbf (as appropriate)
'1' 1 1 bslbf
PTS[14..0] YYYYYYYYYYYYYYY 15 bslbf (as appropriate)
'1' 1 1 bslbf
if (ES_rate_flag ==1'1') {etc.}
if (PES_CRC_flag ==1'1') {etc.}
PES_private_data_flag 1 1 bslbf
pack_header_field_flag 0 1 bslbf
program_packet_sequence_counter_flag 0 1 bslbf
P-STD_buffer_flag 0 1 bslbf
reserved 111 3 bslbf
PES_extension_flag_2 0 1 bslbf
AD_descriptor(){
reserved 1111 4 bslbf
AD_descriptor_length 1000 4 bslbf (16d)
AD_text_tag 0x4454474144 40 bslbf
revision_text_tag 0x31 8 bslbf
AD_fade_byte 0xYY 8 bslbf (FADE byte)
AD_pan_byte 0xYY 8 bslbf (PAN byte)
reserved 0xFFFFFFFFFFFFFF 56 bslbf
}
for (i=0; i<N1; i++) {stuffing_byte} (if required)

Thereafter commences the coded audio elementary stream data.

The presentation time-stamp (PTS) is the ISO/IEC 13818-1 mechanism for synchronising the
presentation of a decoded stream by the decoder. The PTS for each time-critical component of a
service is referenced to the service programme clock reference (PCR). One result of this temporal
binding is that good audio-vision synchronisation can be maintained. The PTS field encapsulated
in a PES packet refers to the first “access unit” 9 (AU) which commences in that PES packet.
The maximum rate of signalling of fade and pan values is determined by the number of audio PES
packets per second for that AD stream. For bit-rate efficiency it is usual to encapsulate several
access units of audio within one PES packet. Note that the fade and pan values in each
AD_descriptor are deemed to apply to each of the AUs encapsulated within, and which commence
in, that PES packet.
It is usual for audio AUs to be aligned with the PES packet and for the PES packet to encapsulate
an integer number of AUs. In this case the PTS refers to the first AU within that PES packet and
the fade and pan value are deemed to apply to all of the AUs therein. In principle this integer
number could also vary from one PES packet to the next .
It is also possible (but relatively unusual) for the PES packetisation to be asynchronous with the
AUs and for the PES packet to commence with and/or to end with an incomplete AU. In this case
the PTS refers to the first AU which commences within that PES packet and the fade and pan
value are deemed to apply to all the AUs which commence therein. Decoders should be capable
of accommodating each of these forms of PES packetisation.
In practice the encapsulation of several AUs within one PES packet means that fade and pan
values are transmitted typically every 120ms to 200ms. This allows the programme provider to
have some control over the attack and decay of a fade and for fades to be reasonably gentle (i.e.
taking several intermediate values between no-fade and the final target) where the gap in narrative
permits.
In the UK the rate at which fade and pan values are transmitted shall never exceed 10 per second
but successive PES packets may well convey different fade and pan values (e.g. during a fade)
which must be accurately represented in the AD decoder output.
An AD decoder must maintain the relative timing between the decoded description signal and the
decoded programme sound signal and between the appropriate fade and pan values and the
decoded description signal.
As noted above, the description signal is mono speech and will typically be coded at a relatively
low bit-rate (e.g. 64 kbits/s). Whilst broadcasters tend to use 48 kHz sampling for digital audio in
their studio infrastructures, other sample rates are possible for digital television services [4]. A
simple practicable constraint is that the sampling rate of the AD audio shall be identical to that
of the programme sound for that TV service 10. AD decoders should therefore be capable of
decoding MPEG1 layer II or MPEG 2 mono signals at bit-rates between 64 kbits/s and 256 kbits/s
and of supporting the audio sampling rates relevant to applicable digital television services.
AD stream components with DTT services in the UK are distributed unscrambled regardless of any
scrambling which might be applied to other service components.
For the duration of a described programme the AD for DTT is transmitted as described above.
During programmes for which there is no description however there is little point in transmitting an
AD stream of continual silence; during these periods the bit-rate accorded to AD could be
reassigned for other purposes 11. In an 18 Mbit/s DTT multiplex with 4 linear TV services this could
yield 300 kbits/s or more of reusable bit-rate during periods when none of these services is being
described. Decoders should therefore be able to respond promptly to the addition of an AD
component at the start of a described programme.

9
An “access unit” is a unit of coded data – e.g. a frame of video or a 24 ms “frame” of MPEG1 layer II audio.
10
This is a new but entirely practicable constraint designed to simplify receiver implementations.
11
e.g. by using null-packet harvesting and using opportunistic data insertion for time non-critical application data such
over-the-air receiver software upgrades etc..
Signalling AD in the PSI & SI
ISO/IEC 13818-1 (MPEG-2) and DVB-SI rules provide straightforward methods of referencing and
labelling the individual stream components of a digital television service. For such a service on
DTT the audio description component is signalled in the Program Map Table (PMT) of the
Programme Specific Information (PSI) in a similar manner to the signalling for the programme
sound component.
The streams for programme sound and for audio description are distinguished in the PSI by the
use of the ISO_639_language descriptor. The audio_type field within the descriptor associated
with programme sound is assigned the value 0x00 (“undefined”) whilst the equivalent descriptor
associated with audio description has its audio_type field assigned the value 0x03 (“visual impaired
commentary”).
This is illustrated below from a typical DTT PMT
…..
// main programme audio details
{
stream_type 0x03 ; Audio MPEG1
reserved 111b
elementary_PID 0x0259 ; PID for programme sound for this service
reserved 1111b
ES_info_length 0x009
{
descriptor_tag 0x52 ; stream identifier descriptor
descriptor_length 0x01
component_tag 0x02
}
{
descriptor_tag 0x0A ; ISO 639 language descriptor
descriptor_length 0x04
ISO_639_language_code “eng” ; English
audio_type 0x00 ; undefined
}
}
…..
// audio description details
{
stream_type 0x03 ; Audio MPEG1
reserved 111b
elementary_PID 0x025A ; PID for AD component for this service
reserved 1111b
ES_info_length 0x009
{
descriptor_tag 0x52 ; stream identifier descriptor
descriptor_length 0x01
component_tag 0x03
}
{
descriptor_tag 0x0A ; ISO 639 language descriptor
descriptor_length 0x04
ISO_639_language_code “eng” ; English
audio_type 0x03 ; visual impaired commentary
}
}

In the event that a service has AD in several languages (e.g. English and Welsh) the PMT
reference to each stream would have the appropriate ISO_639_language_code and the AD
decoder would discriminate between them on the basis of the preferred language chosen in the
user settings.
If a service includes several AD streams (e.g. for different languages) and if there is no mechanism
for the user to select between these streams then the AD decoder should decode the first AD
stream referenced in the PMT for that service.
Decoder behaviour in the presence of AD
In the presence of a valid AD descriptor in the encoded description signal for the selected service
the AD decoder should present the appropriate mix of programme sound and description signal to
the user.
AD decoders should be capable of decoding MPEG1 layer II or MPEG 2 mono signals at bit-rates
between 64 kbits/s and 256 kbits/s and of supporting the audio sampling rates relevant to
applicable digital television services.
The relative timing of programme audio, description audio and the signalled fade/pan values
should be maintained correctly at all times.
When the fade value is 0x00 or in the absence of an AD stream the programme sound level should
be unattenuated.
In the presence of a valid AD descriptor, the AD decoder should attenuate the programme sound
by 0.3 dB per fade value increment. If the AD decoder cannot support such small steps then the
implemented attenuation should match the intended attenuation as closely as possible. For
example if only 1 dB steps are possible then fade values of 0x00 and 0x01 should map to 0db,
0x02, 0x03 and 0x04 should map to1dB, 0x05, 0x06, 0x07 & 0x08 to –2db etc..
In a stereo environment the AD decoder should interpret any pan values outside the ranges
0xEB..0xFF and 0x00..0x15 in the following manner. Pan values from 0x16 to 0x7F inclusive
should be mapped to the value 0x15 (i.e. stereo hard right). Pan values from 0x80 to 0xEA should
be mapped to the value 0xEB (i.e. stereo hard left).
If, whilst listening to a described programme, the user selects a new programme, the AD decoder
should mute the decoded description signal and restore the programme sound to the unfaded
state. This restoration should not be abrupt - it is recommended that under such conditions the
value of fade and of pan are ramped to the default values (0x00) over a period of at least 1 second.
AD decoders should present to their VCR output a mix of programme sound and description
modulated as appropriate by fade and pan but before any attenuation applied by the user control of
overall volume control shown diagrammatically in figure 1.

Decoder behaviour in the presence of errors

If the AD decoder detects an error in, or absence of, the AD descriptor in the encoded AD signal, it
should have a strategy which leads to muting the decoded description signal, restoring the
programme sound to its default unfaded amplitude and setting the effective fade and pan values to
0x0013.
When the AD signal is suddenly lost or regained the AD decoder behaviour as experienced by the
user should never be abrupt. It is recommended that in the event of an error or the absence of AD
signal the value of fade and of pan implemented by the AD decoder be ramped from the signalled
values to the default values (0x00) over a period of at least 1 second 14. Equally, on recovery from
an error or on the reappearance of the AD signal the value of fade and of pan should be ramped to
the signalled values from the default values (0x00) over a similar period.

13
e.g. the AD decoder might flywheel through isolated errors caused by uncorrected transmission errors but respond
appropriately to successive instances of loss caused by problems at the “headend”.
14
Under normal operation the AD decoder should follow at signalled speed any changes in the transmitted fade and
pan values
Decoder user indications
As many potential users of AD will be visually impaired, the user interface should not rely solely on
visual clues (lights or on-screen display logos) to indicate status information such as the presence
or absence of description. Audible indications are desirable and designers should consider how to
distinguish different states using, for example, contrasting tones.
The user should be able to interrogate the decoder (e.g. using a remote control) and be given an
indication that all is well (e.g. a recognisable “beep” and flashing LED). This will indicate presence
in the stream of decodable AD, even when the description may at that particular moment be silent
and the fade value 0x00. Equally a distinguishable indication of the detected absence of the AD
signal is highly desirable (e.g. a using a second style of “beep”).
Other desirable controls with distinctive audible indications include the ability to mute the combined
sound and to adjust the description and overall volumes.
Any user tones applied to the headphone or hi-fi outputs of the decoder should not be added to the
decoder VCR output.

Test streams
The UK DTG Test Centre maintains a suite of test streams to help developers ensure correct
behaviour of DTT AD decoders.
These streams provide tests for include service selection, audio level, relative timing, fade and pan
response and behaviour in the presence of AD signalling errors.

Standardisation
The syntax and signalling for receiver-mix AD described in this white paper is as defined in
annex G of the ETSI DVB guidelines document TR 101 154 v1.5.0 (2004-01).
Currently however TR 101 154 does not explicitly define a preferred pan characteristic –
indeed it suggests (as a footnote example) a constant power model.
The preferred pan law for AD described above has been developed since the text for
annex G of TR 101 154 was written. It is straightforward to implement, produces
manageable peak levels in the receiver and can also be used to produce a fully broadcast-
compatible pre-mix for broadcaster-mix AD under M6 line-up.
This pan law has also been validated in a commercial product (see below) and should be
the pan characteristic used for future decoder implementations of receiver-mix AD 15.

15
The pan characteristic described in this document is also the appropriate characteristic for equipment used to
generate “broadcaster-mix” AD on DTV platforms where that particular signal-form is used.
Implementations
PCMCIA card
The UK DTT multiplex operators represented by TDN commissioned the development of a
PCMCIA card module which fulfils the requirements of an audio description decoder and plugs into
any DTT receiver which has a working “Common Interface” (CI) socket [5]. It was designed and
built by SCM under contract to TDN with technical oversight and functional testing provided by
BBC R&D. The ADM has been successfully demonstrated with an integrated digital TV receiver
and with a suitably equipped DTT set-top-box.
The design was initiated at a time when UK DTT first carried a number of encrypted services.
Contemporary commercial sensitivity about including decryption in the AD module (ADM)
functionality resulted in a design in which the programme sound input to the module is analogue
and from the receiver phono sound output whilst the audio description signal (broadcast by
agreement “in the clear”) is demultiplexed, decoded and processed within the ADM itself.
The ADM also manages the fade and pan processing together with the user interface. This
interface is via a separate and simple infra-red remote controller and external IR receiver pod
which affixes to the top of the receiver.
Separate outputs are available for headphones, hi-fi and VCR. The remote control provides
means of adjusting the description level and of the level of the overall programme-sound/AD mix,
of querying the status of AD on the selected programme and of muting the headphone and hi-fi
outputs. Distinctive tones are also added to these outputs to provide audible confirmation of the
remote control keystrokes and of the AD status.

figure 4: The TDN Audio Description Module

Netgem i-Player
Shortly before Christmas 2003, Netgem launched a version of their i-Player DTT set-to-box
product in which the AD decoding and mixing is performed in software using existing resources
within the original product.
Netgem have clearly considered the usability of this product. Control of the description volume is
integrated on the existing remote control and audio tones are used to distinguish between
presence or absence of AD as the user selects a new service. The decoder also enunciates the
name of the service by playing and decoding short MP3 speech files stored in the product.

Nebula
Nebula Electronics have implemented AD decoding and processing in software on both the PCI
card and USB versions of their DigiTV DTT decoder product.

Philips
Philips Semiconductors have demonstrated AD decoding and processing in software on their
existing PNX1300 (“TriMedia”) chip could implement receiver-mix AD in future DTV decoding
products.

References
[1] BBC RD Report 1979/7
[2] ISO/IEC 13818-1 MPEG 2 system syntax
[3] ISO/IEC 11172-3 MPEG 1 audio syntax
[4] TR 101 154 v1.5.0 Digital Video Broadcasting (DVB); Implementations guidelines
(2004-01) for the use of Video and Audio Coding in Broadcasting Applications
based on the MPEG-2 Transport Stream.
[5] EN 50221 Common Interface Specification for Conditional Access and other
Digital Video Broadcasting Decoder Applications

Glossary of acronyms
Like many contemporary technologies, digital television generates acronyms at an alarming rate.
This glossary aims to unzip just the acronyms used in this white paper.
AD Audio Description
DTT Digital Terrestrial Television
MPEG Moving Pictures Experts Group (ISO common interest group)

DVB Digital Video Broadcasting (EU industry body)

PSI Programme Specific Information

SI Service Information
DTG Digital Television Group (UK industry body)

DSat Digital Satellite

DCable Digital Cable
PES Packetised Elementary Stream
VCR Video Cassette Recorder
PCMCIA Personal Computer Memory Card International Association (industry body)

CI Common Interface (DVB standardised interface)

TDN The Digital Network (UK industry body)

Annex 1 – Pan values which follow the law of sines

pan pan pan angle left right left right

byte value degrees sine law sine law level dB level dB
0xEB -21 -30.0 1.000 0.000 0.000 -∞
0xEC -20 -28.5 1.000 0.022 0.000 -33.061
0xED -19 -27.1 1.000 0.046 0.000 -26.784
0xEE -18 -25.7 1.000 0.071 0.000 -23.000
0XEF -17 -24.2 1.000 0.097 0.000 -20.233
0xF0 -16 -22.8 1.000 0.126 0.000 -18.022
0xF1 -15 -21.4 1.000 0.156 0.000 -16.159
0xF2 -14 -20.0 1.000 0.188 0.000 -14.534
0xF3 -13 -18.5 1.000 0.222 0.000 -13.082
0xF4 -12 -17.1 1.000 0.258 0.000 -11.759
0xF5 -11 -15.7 1.000 0.297 0.000 -10.537
0xF6 -10 -14.2 1.000 0.339 0.000 -9.393
0xF7 -9 -12.8 1.000 0.384 0.000 -8.312
0xF8 -8 -11.4 1.000 0.432 0.000 -7.283
0xF9 -7 -10.0 1.000 0.484 0.000 -6.295
0xFA -6 -8.5 1.000 0.541 0.000 -5.340
0xFB -5 -7.1 1.000 0.602 0.000 -4.413
0xFC -4 -5.7 1.000 0.668 0.000 -3.506
0xFD -3 -4.2 1.000 0.740 0.000 -2.616
0xFE -2 -2.8 1.000 0.819 0.000 -1.738
0xFF -1 -1.4 1.000 0.905 0.000 -0.867
0x00 0 0.0 1.000 1.000 0.000 0.000
0X01 1 1.4 0.905 1.000 -0.867 0.000
0x02 2 2.8 0.819 1.000 -1.738 0.000
0x03 3 4.2 0.740 1.000 -2.616 0.000
0x04 4 5.7 0.668 1.000 -3.506 0.000
0x05 5 7.1 0.602 1.000 -4.413 0.000
0x06 6 8.5 0.541 1.000 -5.340 0.000
0x07 7 10.0 0.484 1.000 -6.295 0.000
0x08 8 11.4 0.432 1.000 -7.283 0.000
0x09 9 12.8 0.384 1.000 -8.312 0.000
0x0A 10 14.2 0.339 1.000 -9.393 0.000
0x0B 11 15.7 0.297 1.000 -10.537 0.000
0x0C 12 17.1 0.258 1.000 -11.759 0.000
0x0D 13 18.5 0.222 1.000 -13.082 0.000
0x0E 14 20.0 0.188 1.000 -14.534 0.000
0x0F 15 21.4 0.156 1.000 -16.159 0.000
0x10 16 22.8 0.126 1.000 -18.022 0.000
0x11 17 24.2 0.097 1.000 -20.233 0.000
0x12 18 25.7 0.071 1.000 -23.000 0.000
0x13 19 27.1 0.046 1.000 -26.784 0.000
0x14 20 28.5 0.022 1.000 -33.061 0.000
0x15 21 30.0 0.000 1.000 -∞ 0.000