Hi,

I don't know if this is the right group - sorry if it isn't.

But i was wondering if someone has some advice: I have a recording from
a convetion in mp3 - its not too good, you hear the audience more than
the people speaking through the loud speaker.

So i was wondering if it wasn't possible to employ some filters in a
sound editor to try and improve the quality (ie suppress some things and
enhance others)?

I just don't quite know how to do that, so i was wondering if someone
had some tips?

On Mon, 26 Jul 2004 22:14:52 +0200, P. Burrows wrote:


Hi,

I don't know if this is the right group - sorry if it isn't.

But i was wondering if someone has some advice: I have a recording from
a convetion in mp3 - its not too good, you hear the audience more than
the people speaking through the loud speaker.

So i was wondering if it wasn't possible to employ some filters in a
sound editor to try and improve the quality (ie suppress some things and
enhance others)?

I just don't quite know how to do that, so i was wondering if someone
had some tips?


You can do clever things with noise-reduction systems to remove
unwanted sound of a particular type - a mains hum or air-conditioning
rumble perhaps. But I'm afraid there isn't a filter that can be
told: "I want to hear voice A but not voice B".

In article ,
says...
On Mon, 26 Jul 2004 22:14:52 +0200, P. Burrows wrote:

I just don't quite know how to do that, so i was wondering if someone
had some tips?


You can do clever things with noise-reduction systems to remove
unwanted sound of a particular type - a mains hum or air-conditioning
rumble perhaps. But I'm afraid there isn't a filter that can be
told: "I want to hear voice A but not voice B".

Well, i guess exaggerated a bit, its mostly laughts and claps from the
audience. But the voices are far away, and i guess a bit tinny.

P. Burrows writes:


Hi,

I don't know if this is the right group - sorry if it isn't.

But i was wondering if someone has some advice: I have a recording from
a convetion in mp3 - its not too good, you hear the audience more than
the people speaking through the loud speaker.

So i was wondering if it wasn't possible to employ some filters in a
sound editor to try and improve the quality (ie suppress some things and
enhance others)?

I just don't quite know how to do that, so i was wondering if someone
had some tips?

As Laurence already said, the term for this type of operation is "noise
reduction." You can google to see if there's anything free out there. Adobe
Audition has a Noise Reduction function available that will do what you need
if you have a nice "clean" section of the noise only.
--
% Randy Yates % "And all that I can do
%% Fuquay-Varina, NC % is say I'm sorry,
%%% 919-577-9882 % that's the way it goes..."
%%%% % Getting To The Point', *Balance of Power*, ELO
http://home.earthlink.net/~yatescr

Laurence Payne writes:
[...]
But I'm afraid there isn't a filter that can be
told: "I want to hear voice A but not voice B".

Have you ever heard of a (classic analog) Wiener filter? It maximizes
the SNR based on the power spectral density of the noise and the
signal. But typically it is as you say - noise reduction systems use
non-linear techniques to do their job.
--
% Randy Yates % "My Shangri-la has gone away, fading like
%% Fuquay-Varina, NC % the Beatles on 'Hey Jude'"
%%% 919-577-9882 %
%%%% % 'Shangri-La', *A New World Record*, ELO
http://home.earthlink.net/~yatescr

In article , says...

I just don't quite know how to do that, so i was wondering if someone
had some tips?

As Laurence already said, the term for this type of operation is "noise
reduction." You can google to see if there's anything free out there. Adobe
Audition has a Noise Reduction function available that will do what you need
if you have a nice "clean" section of the noise only.

Yeah i tried the demo, doesn't seem to work for this. The noise
reduction doesn't seem to improve the tinny loud speaker speech. Plenty
of other filters, but no idea how the work.
Oh well.

P. Burrows writes:

In article , says...

I just don't quite know how to do that, so i was wondering if someone
had some tips?

As Laurence already said, the term for this type of operation is "noise
reduction." You can google to see if there's anything free out there. Adobe
Audition has a Noise Reduction function available that will do what you need
if you have a nice "clean" section of the noise only.

Yeah i tried the demo, doesn't seem to work for this. The noise
reduction doesn't seem to improve the tinny loud speaker speech.

That's a different problem. You apparently, then, have two problems
you're trying to correct for, 1) a noisy background, and 2) a tinny
sounding speaker. I would try applying noise reduction first, then
following that with some type of equalizing filter. If you look at
the spectrum of the noise-reduced output (especially over a bunch
of samples where speech is present, like a few seconds worth), you
might get a picture of the response of the speaker. Try designing
a filter that is the *inverse* of that and applying it as the
equalizing filter.

Plenty
of other filters, but no idea how the work.
Oh well.

See above.
--
Randy Yates
Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
, 919-472-1124

"P. Burrows" wrote:

Hi,

I don't know if this is the right group - sorry if it isn't.

Close enough.

But i was wondering if someone has some advice: I have a
recording from a convetion in mp3 - its not too good, you hear
the audience more than the people speaking through the loud
speaker.

Is it understandable? - if so: make a transcript instead and the
information will be made available.

So i was wondering if it wasn't possible to employ some filters
in a sound editor to try and improve the quality (ie suppress
some things and enhance others)?

Yes.

I just don't quite know how to do that, so i was wondering
if someone had some tips?

You need to understand that noise is an expression of mathematical
uncertainty, you can not put precision into a data set that wasn't there
in the first place. Noise reduction generally works poorly, if at all,
if the starting signal to noise dynamic range is poorer than some 35 dB.
The brain will generally do better speech extraction than any machine
yet conceived by man.

Things can be done with noise, also with extranous noise, such as
traffic noise, but what is unfixable is a poor relationship between
direct and reflected sound, it will sound "roomy" no matter what.

Things can be done, but coming up with a cookbook is not possible,
mostly however you would equalize first and do digital nr later to avoid
increasing the audibility of the nr artifacts.

Gentle nudges with a suitable (!) expander function may improve the
reducability of the noise or reduce the audibility of the artifacts. You
can use white or pink noise at a suitable (!) level as noise example if
it is not possible to find a usable example in the recording or in case
the background noise is non-static, such as traffic noise or room noise.

The noise reduction demo example on my site (one channel processed, one
channel unprocessed, dual mono) is processed with Cool Edit 2000's nr
plug in. Other software exists, and it is a less obvious recommendation
for the first time purchaser to say go get Adobe Audition (that is how
to get than nr engine now!).

I never did get to test it, because the webshop would not allow me to
choose the german language version (newer!) of the Magix stuff and
insisted that I wanted the english language version based on my IP
address, but it looks as highly cost-efficient entry level software.


Kind regards

Peter Larsen

--
*******************************************
* My site is at: http://www.muyiovatki.dk *
*******************************************

Randy Yates wrote:

Yeah i tried the demo, doesn't seem to work for this. The noise
reduction doesn't seem to improve the tinny loud speaker speech.

That's a different problem. You apparently, then, have two problems
you're trying to correct for, 1) a noisy background, and 2) a tinny
sounding speaker. I would try applying noise reduction first, then
following that with some type of equalizing filter.

It will not get a better recording if the background sounds all sound
wrong.

If you look at the spectrum of the noise-reduced output (especially
over a bunch of samples where speech is present, like a few seconds
worth), you might get a picture of the response of the speaker.

I don't think that he can get usable results that simply, I have done a
great deal of work on such strategies - I think it was Nannestad that
demonstrated at an AES meeting here in Copenhagen just how well a 78 rpm
recording of a string quartet could be made to sound when equalized to
suit the spectrum of a modern, similar performance.

Try designing a filter that is the *inverse* of that and applying
it as the equalizing filter.

In this context it is probably faster for him to experiment with moving
a slider at at the time of a virtual or physical octave band equalizer.

Plenty
of other filters, but no idea how the work.
Oh well.

See above.

It takes time to learn and to gather an experience and knowledge base.

Randy Yates


Kind regards

Peter Larsen

--
*******************************************
* My site is at: http://www.muyiovatki.dk *
*******************************************

Peter Larsen writes:

Randy Yates wrote:

Yeah i tried the demo, doesn't seem to work for this. The noise
reduction doesn't seem to improve the tinny loud speaker speech.

That's a different problem. You apparently, then, have two problems
you're trying to correct for, 1) a noisy background, and 2) a tinny
sounding speaker. I would try applying noise reduction first, then
following that with some type of equalizing filter.

It will not get a better recording if the background sounds all sound
wrong.

Don't follow you here. A spectral subtraction scheme should work as long
as you can detect the difference between signal+noise and noise-only.

If you look at the spectrum of the noise-reduced output (especially
over a bunch of samples where speech is present, like a few seconds
worth), you might get a picture of the response of the speaker.

I don't think that he can get usable results that simply, I have done a
great deal of work on such strategies - I think it was Nannestad that
demonstrated at an AES meeting here in Copenhagen just how well a 78 rpm
recording of a string quartet could be made to sound when equalized to
suit the spectrum of a modern, similar performance.

Are you saying that it's not that easy to get an estimate of the
system response (i.e., the one that is to be equalized)? That may or
may not be true. It depends on the type of non-noise signal that is
present since the spectrum of that signal weights the response of the
system that will be observed. But if the tinniness is severe it may
still be close enough to do some good. If ill-conditioning is involved,
then you can do some heuristic adjustments such as introducing a
frequency-dependent maximum gain which the equalizer will not exceed.
--
% Randy Yates % "I met someone who looks alot like you,
%% Fuquay-Varina, NC % she does the things you do,
%%% 919-577-9882 % but she is an IBM."
%%%% % 'Yours Truly, 2095', *Time*, ELO
http://home.earthlink.net/~yatescr

In article ,
says...

But i was wondering if someone has some advice: I have a
recording from a convetion in mp3 - its not too good, you hear
the audience more than the people speaking through the loud
speaker.

Is it understandable? - if so: make a transcript instead and the
information will be made available.

Yeah, people have been trying to make a transcript, and some areas are
clear and others are very much not. (It's public, you could have a
listen if you liked: http://www.fireflyfans.net/feature.asp?f=60 )


You need to understand that noise is an expression of mathematical
[snip]

So basically, there are no magic potions, sometimes one might be able to
do something, but one would generally need to have some experience.

I guessed it would be something like that. Thanks for your reply (and
the others

"P. Burrows" wrote:

In article ,
says...

But i was wondering if someone has some advice: I have a
recording from a convetion in mp3 - its not too good, you hear
the audience more than the people speaking through the loud
speaker.

Is it understandable? - if so: make a transcript instead and the
information will be made available.

Yeah, people have been trying to make a transcript, and some areas are
clear and others are very much not. (It's public, you could have a
listen if you liked: http://www.fireflyfans.net/feature.asp?f=60 )

You need to understand that noise is an expression of mathematical
[snip]

So basically, there are no magic potions, sometimes one might be able to
do something, but one would generally need to have some experience.

I guessed it would be something like that. Thanks for your reply (and
the others

Ever seen the film " The Conversation " with Gene Hackman ?

Might give you some ideas :-)


Graham

Randy Yates wrote:

That's a different problem. You apparently, then, have two
problems you're trying to correct for, 1) a noisy background,
and 2) a tinny sounding speaker. I would try applying noise
reduction first, then following that with some type of
equalizing filter.

[my comment]

It will not get a better recording if the background sounds all sound
wrong.

Don't follow you here. A spectral subtraction scheme should work
as long as you can detect the difference between signal+noise
and noise-only.

I am speaking about the effect of EQ'ing for the loudspeakers oddities
on the sound of the recorded room tone.

[Randy]

If you look at the spectrum of the noise-reduced output
(especially over a bunch of samples where speech is present,
like a few seconds worth), you might get a picture of the
response of the speaker.

[my comment]

I don't think that he can get usable results that simply,

Are you saying that it's not that easy to get an estimate of the
system response (i.e., the one that is to be equalized)?

Yes, I am saying that a few seconds is not enough, you need to average
over a considerable segment of audio. You also have to know what to
considerer "linear", i.e. how the type of signal is likely to come
across in a room of whatever size that room is. Not much information on
such issues is - to my knowledge - publicly available. Nannestad's
strategy for 78 rpm recording restoration was described as comparing
with a modern recording of the same oeuvre. Mind you, I didn't say that
it can not be gotten less bad simply by getting it "probable".

% Randy Yates


Kind regards

Peter Larsen

--
*******************************************
* My site is at: http://www.muyiovatki.dk *
*******************************************

"P. Burrows" wrote:

Yeah, people have been trying to make a transcript, and some areas
are clear and others are very much not. (It's public, you could
have a listen if you liked: http://www.fireflyfans.net/feature.asp?f=60 )

IMO not recovable due to poor mic positioning. BTW, why are you wasting
server space and bandwidth with a two channel file of what appears to be
mono? - it would take someone like Gene Hackmann to fix it, but he
would have had multiple mics recorded individually. Next time if you
record stereo, then record the podium mike on one track and the audience
mike on the other. Using a wide dynamics recording format and static
gain may be wiser than using something with automatic gain control.

Your mileage may vary.


Kind regards

Peter Larsen


--
*******************************************
* My site is at: http://www.muyiovatki.dk *
*******************************************