Re: Radio 3 Freeview & Satellite recordings
davidr...@postmaster.co.uk
> > In message <1155731750....@74g2000cwt.googlegroups.com>
> > "davidr...@postmaster.co.uk" <davidr...@postmaster.co.uk> wrote:
> >
> >> As a follow up to...
> >>
> >> http://www.digitalradiotech.co.uk/samples/proms_2006/
> >>
> >> I've now listened to the pieces in full. The trumpet in the
> >> Prom2006_39_Schnittke_DVB-s.mp2 file has obvious audible artefacts. If
> >> you want to hear them, 4:13 into the file is the note where the
> >> artefacts are most obvious. The muted trumpet in the 6th minute also
> >> suffers.
> >>
> >> Of course the DVB-T file is even worse, but I don't think the DVB-S
> >> broadcast is acceptable.
[snip]
> If you load them into Audition you may see something rather interesting.
> I decode with foobar 2000 to 32 bit without dither. The sat version
> has a load of mush between 16Khz and 20Hhz that has nothing to do with
> the music (no harmonics from the music pass into this area) and a deep
> dip at around 15Khz. So on paper bet that sat version sounds brighter as
> the freeview version has no such problems.
Thanks for mentioning that Paul. I've done the same as you, and see the
same.
I think the situation is even more bizarre than you suggest.
Firstly, there's a very stable asynchronous sample rate conversion
between the two sources. Taking the Mozart, there's a 12674 sample
delay between the two files at the start, but a 12679 sample delay
between the two files at the end.
Secondly, that "mush between 16Khz and 20Hhz" isn't signal, but it
isn't "mush" either - it's 16-bit dither. At that level, it's inaudible
and largely irrelevant to the sound (though it's inefficient to waste
bits encoding it!), but it makes it possible to guess how the signal
was encoded.
That 16-bit dither is encoded into the mp2 file (which itself has no
native bit depth). This means the DVB-S signal has been encoded
_without_ a low pass filter (and without the absolute threshold of
hearing model enabled!) from a previous "generation" source which was
16-bits, already low pass filtered, and dithered. It's possible the low
pass filter was fixed at 16kHz, but also possible it was dynamic. (The
final mp2 encoding makes it hard to tell what the previous generation
was.) The most likely explanation is that the DVB-S signal is encoded
from a previous (mp2?) encoding.
Thirdly, the lack of "mush" in the DVB-T version is nothing positive.
There is simply a 16kHz fixed low pass filter which removes most of it.
However, if you look closely around 14-16kHz, you'll see that there's
both the "mush" you refer to above at ~16-bit dither level, and
additional lower level mush in the "holes" at ~18-bit dither level from
yet another generation of encoding/dithering.
Making a wild guess, the signal chain goes something like this...
1. 48kHz linear PCM from the Albert Hall.
2. Encoded (may be mp2) with dynamic LPF around 16kHz for internal
distribution
3. Decoded to 16-bits.
4. Re-encoded to mp2 with no LPF* for DSat
5. Decoded to 18-bits.
6. Re-encoded to mp2 for DTT
* the 20.3kHz low pass filter is a hard limit of the mp2 format itself
The re-encoding stages include, at the very least, asynchronous sample
rate conversion. It could even be digital>analogue>digital conversion.
More importantly, each re-encoding stage dramatically lowers the
quality.
The DVB-T version is obviously at least one further generation away
from the source than the DVB-S version. Whether the DVB-T version uses
the DVB-S version as a source, or if there's a separate string of mp2
encoders feeding the DVB-T version is hard to say. It's irrelevant
really - either way, the quality suffers.
What's interesting (and sad) is that someone set the DVB-S encoder up
for what they (foolishly) believed would give high quality. People who
don't understand psychoacoustic encoding often disable the low pass
filter. So that's one mistake. The bigger mistake is encoding, not from
a linear PCM source, but from an already encoded source - which by
definition lowers the quality.
It makes you wonder what on earth they're playing at, and if there's an
audio engineer anywhere in the building!
Cheers,
David.
Richard Evans
<SNIP>
>
> It makes you wonder what on earth they're playing at, and if there's an
> audio engineer anywhere in the building!
Perhaps they thought it would be cheaper to employ a 13 y/o schoolboy to
set up the encoding :(
Richard E.
Paul Webster
wrote:
>Paul Edwards wrote:
>> > In message <1155731750....@74g2000cwt.googlegroups.com>
>> > "davidr...@postmaster.co.uk" <davidr...@postmaster.co.uk> wrote:
>> >
<big snip>
Could any of that odd encoding be "watermarking".
i.e. putting information into the broadcast that can be:
a) checked for copyright infringement later
b) be detected over the air by external (home-based) equipment
gathering listening stats
Some old references:
http://www.aes.org/sections/uk/meetings/0103.html
http://www.bbc.co.uk/rd/pubs/whp/whp078.shtml
http://www.aes.org/events/116/papers/C.cfm
http://www.bbc.co.uk/rd/pubs/whp/whp101.shtml
--
Rgds
Paul Webster
Richard Evans
audio coding, and it sounds like he has looked rather carefully at these
samples. So I would tend to think his original conclusions would be
about rite.
Richard E.
hwh
So now there is "proof" that the quality of digital broadcasting is not
only influenced by the chosen bitrates, but serious engineering
mistakes as well.
I hope someone with influence at the BBC does something about this, it
looks like improvement is relatively simple and not expensive.
gr, hwh
davidr...@postmaster.co.uk
They're good references, though patent searches for the relevant
companies (e.g, Verance) will tell you more about commercial
implementations, as does the hack SDMI challenge.
There is a plan/proposal/idea to watermark _all_ broadcast content to
help with RAJAR/BARB type activities. I don't know how advanced this
is, or if it will definitely happen.
There could well be audio watermarks in the extracts, and they could
contribute slightly to the sound quality issues. However, what's
clearly visible in the spectral analysis, and what's clearly audible
(especially in the Freeview version) isn't watermarking as far as I can
tell - unless it's gone _very_ wrong!
A watermark (in the true sense of the word) wouldn't normally be placed
90dB down at 16-20kHz. If the goal was some kind of copy protection, it
would be relatively easy to remove intentionally and very easily lost
by accident! If the goal was some kind of audience monitoring, it
wouldn't be reproduced reliably via cheaper speaker systems. "Good"
audio watermarks are hidden in the important part of the audible
frequency range (typically 500Hz-3kHz, which even gets through a
telephone). If you know the algorithm, they're detectable (obviously!),
but you can't "see" _most_ of them on a normal spectrum analysis (there
are exceptions, e.g. Copycode, but they're not good "watermarks").
Cheers,
David.
David A Stocks
in news:1156239634....@74g2000cwt.googlegroups.com:
> Paul Webster wrote:
>> "davidr...@postmaster.co.uk" <davidr...@postmaster.co.uk>
>> wrote:
>>
>> >Paul Edwards wrote:
>> >> > In message <1155731750....@74g2000cwt.googlegroups.com>
>> >> > "davidr...@postmaster.co.uk"
>> >> > <davidr...@postmaster.co.uk> wrote:
>> >> >
>> <big snip>
>> Could any of that odd encoding be "watermarking".
>> i.e. putting information into the broadcast that can be:
>> a) checked for copyright infringement later
>> b) be detected over the air by external (home-based) equipment
>> gathering listening stats
>>
>> Some old references:
>> http://www.aes.org/sections/uk/meetings/0103.html
>> http://www.bbc.co.uk/rd/pubs/whp/whp078.shtml
>> http://www.aes.org/events/116/papers/C.cfm
>> http://www.bbc.co.uk/rd/pubs/whp/whp101.shtml
>
> There is a plan/proposal/idea to watermark _all_ broadcast content to
> help with RAJAR/BARB type activities. I don't know how advanced this
> is, or if it will definitely happen.
>
The BARB service uses sound signature technology to identify broadcast
content for audience mesaurement - AIUI this is based on matching
signatures captured by the meter against those from a reference site and
there is no dependence on watermarking.
HOWEVER, BARB and RAJAR are doing a joint trial of personal metering
using the Arbitron Portable People Meter (PPM) system:
" ... in a new joint venture with BARB, RAJAR has commissioned TNS to
recruit, install and operate an electronic measurement panel using the
Arbitron Portable People Meter (PPM). This panel will initially act as a
learning outlet for the entire radio industry, whilst preparing the way
for a move to a full electronic system for radio as and when it becomes
feasible. Data from the panel is to be analysed separately and will not
be integrated in to the main currency. The electronic measurement panel
is to run for two years and will deliver weekly data from 50 national
and London radio stations within the M25 across four platforms:
analogue, DAB, DTV and Internet. It is anticipated that RAJAR will
provide regular reports regarding key listening insights from this
panel."
quote from
<<http://www.rajar.co.uk/bulletinShow.cfm?uid=060705041656&bulletinid=96
&part=two>>
From the following blurb on the Arbitron website it would appear that
this *does* use some form of watermarking.
<<http://www.arbitron.com/portable_people_meters/thesystem_ppm.htm>>
D A Stocks
Mark Carver
David A Stocks wrote:
> " ... in a new joint venture with BARB, RAJAR has commissioned TNS to
> recruit, install and operate an electronic measurement panel using the
> Arbitron Portable People Meter (PPM).
'PPM' eh ? And I see it's been registered as a service mark. Mind you
PPMs are just there these days to stick 'Post-Its' to, aren't they ?
davidr...@postmaster.co.uk
Thanks for that David.
The Arbitron US patents explain some methods they could be using. Given
that no one will try to strip these watermarks, or speed up / slow down
the content at all, they don't have to be that clever compared to those
proposed for digital rights management.
I haven't found any listening test results to demonstrate that their
technology really is transparent when cascaded with mp2, NICAM, etc,
but maybe I need to dig around their website a little more.
Cheers,
David.
ted msn
<davidr...@postmaster.co.uk> wrote in message
news:1156322396....@b28g2000cwb.googlegroups.com...
What the device seems to do is ident what IT hears not what the person
hears. I have the radio on while reading the newspaper at breakfast brain
does 95% reading and about 3% radio 2% "noise" but the device would say 100%
radio. If I used headphones??
Their key phrase is "versatile audience measurement system that can track
consumer exposure" it then goes on to say "determine what consumers listen
to" which is not the same thing at all. It is also miles away from what the
consumer thinks or understands from what they "heard"
ted
David A Stocks
news:db2dnYJJOuF...@bt.com:
>
> <davidr...@postmaster.co.uk> wrote in message
> news:1156322396....@b28g2000cwb.googlegroups.com...
>>
>> Thanks for that David.
>>
>> The Arbitron US patents explain some methods they could be using.
>> Given that no one will try to strip these watermarks, or speed up /
>> slow down the content at all, they don't have to be that clever
>> compared to those proposed for digital rights management.
>>
>> I haven't found any listening test results to demonstrate that their
>> technology really is transparent when cascaded with mp2, NICAM, etc,
>> but maybe I need to dig around their website a little more.
>>
>> Cheers,
>> David.
>>
> Hum !
> What the device seems to do is ident what IT hears not what the person
> hears. I have the radio on while reading the newspaper at breakfast
> brain does 95% reading and about 3% radio 2% "noise" but the device
> would say 100% radio. If I used headphones??
>
This is a criticism of most audience measurement systems. The BARB
definition of "viewing" is that a panellist has declared their presence
in a room with the TV set switched on and tuned to a channel. A few
years ago someone from one of the (then) BARB contractors told me that
their QA procedures would occasionally pick up the fact that a perfectly
valid (according to the above definition) "viewer" on the BARB panel had
died.
For TV measurement the potential advantages of metering the people
rather than TV sets are that you can pick up out of home viewing (e.g.
sports in pubs/clubs), and avoid errors introduced by the need for
panellists to declare their presence in the room.
David A Stocks
opinions are mine, not my employers.