[[email protected]]

Hi everyone,

I'm collecting audio recordings of dialog made on a wide variety of
cheap microphones, and writing automatic scripts to clean up and
normalize the recordings, to make them of comparable quality (some
automatic noise reduction and volume normalization).

However the microphones tend to have low response at bass and/or
treble frequencies, but by different amounts.

All the sound is male and female spoken voice.

I was wondering if there's equalization software out there that will
take a recording and *automatically* equalize it according to a target
profile, automatically upping bass and treble independently as
necessary, but not too much?

Another feature I'm trying to find would be a way to programmatically
sense if if the frequencies cut off abruptly at too low of a level
(i.e. at 5 kHz instead of 15), in order to reject those recordings...

I've been doing all my scripting in Sony Sound Forge, but the
scripting features can neither get at spectrogram analysis values or
set custom EQ values...

I'm looking for a necessarily automatic solution that works on a file-
per-file basis, because I could eventually be working with thousands
of different microphones...

Any pointers are appreciated, thanks!
Michael

[Richard Crowley]

wrote ...
I'm collecting audio recordings of dialog made on a wide variety of
cheap microphones, and writing automatic scripts to clean up and
normalize the recordings, to make them of comparable quality (some
automatic noise reduction and volume normalization).

However the microphones tend to have low response at bass and/or
treble frequencies, but by different amounts.

All the sound is male and female spoken voice.

I was wondering if there's equalization software out there that will
take a recording and *automatically* equalize it according to a target
profile, automatically upping bass and treble independently as
necessary, but not too much?

How can you possibly do that without some fixed reference?
Even manually, much less automatically.

Besides, if the recordings are made with cheap mics, boosting
the LF will likely also boost hum and ambient rumble. And
boosting HF will almost certainly boost hiss. Not clear why
either of those would be desirable?

Another feature I'm trying to find would be a way to programmatically
sense if if the frequencies cut off abruptly at too low of a level
(i.e. at 5 kHz instead of 15), in order to reject those recordings...

It seems likely that such a software application could be written.

I've been doing all my scripting in Sony Sound Forge, but the
scripting features can neither get at spectrogram analysis values or
set custom EQ values...

I'm looking for a necessarily automatic solution that works on a file-
per-file basis, because I could eventually be working with thousands
of different microphones...

You may have selected an impossible goal.

[Dec [Cluskey]]

Michael

Pointers?

If you are recording and producing thousands of microphones then that
represents a lifetime's work?

With enormous respect, I have vast experience of collecting recordings
done at various times and various situations with more than various
acoustic environments - and I have always found that producing the
eventual CD necessitated relying on my ears and only my ears.

It would be marvelous if there was an automatic, computer produced
programme to smooth the levels, the EQ and the ambient output ... but,
in my humble and experienced opinion, there is not.

There is, however, the great knowledge of EQ, the great knowledge of
compressor/limiters and the great knowledge of ambience [small room,
large room etc. on most good quality FX units] A good listen to an UK
BBC Radio 4 afternoon play will demonstrate how FX units can give an
uncanny ambience to the spoken word.

I will be fascinated to hear any other responses, but I do feel that
you have a lot of hard work to look forward to.

Dec [Cluskey]


On Jun 29, 1:17 am, wrote:
Hi everyone,

I'm collecting audio recordings of dialog made on a wide variety of
cheap microphones, and writing automatic scripts to clean up and
normalize the recordings, to make them of comparable quality (some
automatic noise reduction and volume normalization).

[Nova Music Productions]

wrote in message
oups.com...
Hi everyone,

I'm collecting audio recordings of dialog made on a wide variety of
cheap microphones, and writing automatic scripts to clean up and
normalize the recordings, to make them of comparable quality (some
automatic noise reduction and volume normalization).

However the microphones tend to have low response at bass and/or
treble frequencies, but by different amounts.

All the sound is male and female spoken voice.

I was wondering if there's equalization software out there that will
take a recording and *automatically* equalize it according to a target
profile, automatically upping bass and treble independently as
necessary, but not too much?

Another feature I'm trying to find would be a way to programmatically
sense if if the frequencies cut off abruptly at too low of a level
(i.e. at 5 kHz instead of 15), in order to reject those recordings...

I've been doing all my scripting in Sony Sound Forge, but the
scripting features can neither get at spectrogram analysis values or
set custom EQ values...

I'm looking for a necessarily automatic solution that works on a file-
per-file basis, because I could eventually be working with thousands
of different microphones...

Any pointers are appreciated, thanks!
Michael

As far as EQ alone goes, I believe Free Filter does something like this. You
may need to use a spectrum analyzer (like the one in Sound Forge) first,
tho.

Mikey Wozniak
Nova Music Productions
this sig is haiku

[Nick Brown]

Nova Music Productions wrote:
As far as EQ alone goes, I believe Free Filter does something like this. You
may need to use a spectrum analyzer (like the one in Sound Forge) first,
tho.


It does, but since these recording are of different people, their
voices will themselves have different frequency patterns (different
amounts of bass in particular), and that's surely something that needs
to be preserved, rather than EQ'd out.

-Nick

[Ty Ford]

On Thu, 28 Jun 2007 20:17:30 -0400, wrote
(in article .com):

Hi everyone,

I'm collecting audio recordings of dialog made on a wide variety of
cheap microphones, and writing automatic scripts to clean up and
normalize the recordings, to make them of comparable quality (some
automatic noise reduction and volume normalization).

However the microphones tend to have low response at bass and/or
treble frequencies, but by different amounts.

All the sound is male and female spoken voice.

I was wondering if there's equalization software out there that will
take a recording and *automatically* equalize it according to a target
profile, automatically upping bass and treble independently as
necessary, but not too much?

Another feature I'm trying to find would be a way to programmatically
sense if if the frequencies cut off abruptly at too low of a level
(i.e. at 5 kHz instead of 15), in order to reject those recordings...

I've been doing all my scripting in Sony Sound Forge, but the
scripting features can neither get at spectrogram analysis values or
set custom EQ values...

I'm looking for a necessarily automatic solution that works on a file-
per-file basis, because I could eventually be working with thousands
of different microphones...

Any pointers are appreciated, thanks!
Michael


Hi Michael,

Would that it were that easy.

My experience has been that your scripts do not take into consideration the
match between the mic and different preamps and that human voices vary quite
a bit.

Regards,

Ty Ford


--Audio Equipment Reviews Audio Production Services
Acting and Voiceover Demos http://www.tyford.com
Guitar player?:http://www.youtube.com/watch?v=4RZJ9MptZmU

[[email protected]]

OK, first of all, thanks for all the feedback. The consensus seems to
be that it is an impossible task... BUT, let me explain more precisely
what I'm doing.

I'm taking spoken audio submissions from a website (this is why there
could be 1000s of low-quality microphones involved), and then I need
to clean them up to sound fairly consistent. But NOT to sound CD-
quality or anything. I'm doing automatic noise reduction (based on
initial seconds of silence) and volume normalization. The final
results are going to be *very* low-quality MP3s with added ambient
sounds (i.e. city street sounds, etc.) masking remaining noise. The
goal is intelligibility and consistency, not audio quality (it needs
to be better than telephone quality, but not as good as radio
quality).

So far, all that works. My only problem is that some submissions sound
'tinny' and others sound muffled. I don't need to make each recording
sound the absolute best it possibly can, but rather just try to avoid
glaring tinniness (by increasing the bass frequencies) or muffledness
(by increasing treble). Again, in a very crude way, e.g. using a 3-
level graphic equalizer (not 20 levels!).

So really, the question is: does anybody think it's possible to use a
spectrogram analysis to judge the energy levels within three different
bands (e.g. up to 200, from 200 to 2000, and from 2000 up, maybe?),
detect whether the microphone is weak in bass and/or treble compared
to the energy in the mid-range, and produce an automatic adjustment
level? Since each recording is based only on voice, and I know whether
each speaker is male or female, and I've performed some noise
reduction previously, it seems like my desired overall energy levels
should be fairly consistent, no? Again, I'm not looking for 100%
perfection, more like an improvement that gets 60% or 70% there...

Leaving aside the question of whether there is software to
automatically do this, does this seem theoretically possible? I've
been looking at the spectrum analyses in Sound Forge of various
submissions, and I can't say that I find it easy to visually tell if a
recording is lacking in bass and/or treble. But maybe I've just got
the settings wrong, or don't know what to look for?

Thanks!
Michael

[Scott Dorsey]

wrote:
So far, all that works. My only problem is that some submissions sound
'tinny' and others sound muffled. I don't need to make each recording
sound the absolute best it possibly can, but rather just try to avoid
glaring tinniness (by increasing the bass frequencies) or muffledness
(by increasing treble). Again, in a very crude way, e.g. using a 3-
level graphic equalizer (not 20 levels!).

So really, the question is: does anybody think it's possible to use a
spectrogram analysis to judge the energy levels within three different
bands (e.g. up to 200, from 200 to 2000, and from 2000 up, maybe?),
detect whether the microphone is weak in bass and/or treble compared
to the energy in the mid-range, and produce an automatic adjustment
level?

Yes, but a skilled operator should be able to do a better job and it shouldn't
take any more time than the machine takes to do it automatically. Try doing
it by ear... after you get the hang of it you should be able to get a rough
EQ balance in five seconds or so.

Leaving aside the question of whether there is software to
automatically do this, does this seem theoretically possible? I've
been looking at the spectrum analyses in Sound Forge of various
submissions, and I can't say that I find it easy to visually tell if a
recording is lacking in bass and/or treble. But maybe I've just got
the settings wrong, or don't know what to look for?

There are gadgets out there that will adjust the frequency envelope so it
matches a basic reference. MagicEQ is one of them. For the most part
they do a poorer job than a person and don't save any real time.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Paul Stamler]

wrote in message
oups.com...

So really, the question is: does anybody think it's possible to use a
spectrogram analysis to judge the energy levels within three different
bands (e.g. up to 200, from 200 to 2000, and from 2000 up, maybe?),
detect whether the microphone is weak in bass and/or treble compared
to the energy in the mid-range, and produce an automatic adjustment
level? Since each recording is based only on voice, and I know whether
each speaker is male or female, and I've performed some noise
reduction previously, it seems like my desired overall energy levels
should be fairly consistent, no?

No. Voices vary all over the map, and trying to make them spectrally
consistent will make many of them sound very, very wrong.

Oh, and normalization won't make them sound like they're the same volume
either; normalization looks at the peaks, while your ears judge loudness
mainly on the average level, and unfortunately peak-to-average ratio varies
with different speakers almost as much as frequency spectrum.

In the end, your ears will be needed for both jobs.

Peace,
Paul

[Marc Wielage]

On Jun 28, 2007, commented:

I'm looking for a necessarily automatic solution that works on a file-
per-file basis, because I could eventually be working with thousands
of different microphones...
------------------------------snip------------------------------

I agree with most of the others. Another reason why an automatic mode can't
work is that the acoustics of each room in which the persons were recorded
will be vastly different from each other. Some will sound dull, some will
have a lot of echo, some will have hum and noise, some will have background
sounds, etc. No automatic device can independently EQ a voice on top of
these sounds and make them sound reasonable. You need to use your own ears
to make these choices.




The final
results are going to be *very* low-quality MP3s with added ambient
sounds (i.e. city street sounds, etc.) masking remaining noise.
------------------------------snip------------------------------

That is not going to make the final dialog more intelligible. Anything you
mix on top of the dialog is going to tend to interfere with it. To me, this
is like putting a dirty bandaid on top of the cut. Sure, you're covering up
the wound, but you're not really doing it any good.

I can see a need to mix in background sound if this were cutting back and
forth, like for a film or TV show, but it doesn't sound like this is what
you're doing.

--MFW

[Ty Ford]

On Fri, 29 Jun 2007 10:54:47 -0400, wrote
(in article .com):

OK, first of all, thanks for all the feedback. The consensus seems to be that

it is an impossible task... BUT, let me explain more precisely what I'm
doing.

I'm taking spoken audio submissions from a website (this is why there could
be 1000s of low-quality microphones involved), and then I need to clean them
up to sound fairly consistent. But NOT to sound CD- quality or anything. I'm
doing automatic noise reduction (based on initial seconds of silence) and
volume normalization. The final results are going to be *very* low-quality
MP3s with added ambient sounds (i.e. city street sounds, etc.) masking
remaining noise. The goal is intelligibility and consistency, not audio
quality (it needs to be better than telephone quality, but not as good as
radio quality).

NO! (for the same reasons as above) You have to do it by ear. Each situation
will be different and each will require a lot of tweeky noise reduction.

Regards,

Ty Ford


--Audio Equipment Reviews Audio Production Services
Acting and Voiceover Demos http://www.tyford.com
Guitar player?:http://www.youtube.com/watch?v=4RZJ9MptZmU

[[email protected]]

OK, thanks for all the feedback! Sounds like doing it by ear is the
only way. Hopefully I get halfway decent results... The patching in of
ambient sounds is actually necessary since each recording is supposed
to sound like it is being recorded in specific places, even though
it's all in front of computers..

Too bad technology isn't up to the task yet... And to think the next
thing I wanted was automatic reverb removal! Haha.. one can dream...

As you can see this is an exceedingly low-budget operation.

Thanks again!

Michael

[Dec [Cluskey]]

On Jul 1, 12:59 am, wrote:
OK, thanks for all the feedback! Sounds like doing it by ear is the
only way. Hopefully I get halfway decent results... The patching in of
ambient sounds is actually necessary since each recording is supposed
to sound like it is being recorded in specific places, even though
it's all in front of computers..

Too bad technology isn't up to the task yet... And to think the next
thing I wanted was automatic reverb removal! Haha.. one can dream...

As you can see this is an exceedingly low-budget operation.

Thanks again!

Michael

If you ever need to remove compression .... ask the group!

Now there is a trick we would all ant to learn!

Dec [Cluskey]