Audio editing and converting

I thought I'd start a new thread here on this subject, since the YouTube one
had drifted off subject.

LostGalliFreyan wrote:

Bill, if you want to hear a 32 kbps MP3 test of a clip
about 52 seconds long, I posted it in alt.binaries.test.test
"For Bill, a small MP3 at 32 kbps test"
The MP3 is there, plus the WAV clip too. I don't know how it
will compare with WMA, I'm just exploring MP3 to see if it
will even go there, and it's not too shabby.

I spent a couple of hours trying to get those files LGF, but had only
partial success (and NOT with OE). The collection of WAV fragments, when
combined and decoded, came out to 29 seconds, instead of the expected 52
seconds (as in the mp3), for some weird reason (and yes, I tried doing the
downloading and combining again, but to no avail (using Xananews, which is
about as far as I could go).

But at least it was listenable, and so I could at least compare the
abbreviated WAV file with the mp3 (which did come out in full). So, so
much for downloading binaries; it seems to be a bit problematic at times
(i.e., in successfully combining and decoding the multipart fragments).

However, since it was in mono, I couldn't do a good comparison with WMA,
using a joint stereo, low bitrate, A/B comparison (32 kbps 22 KS joint
stereo).

However, I did notice that the mp3 was a bit duller (lacked the highs) as
compared to the source WAV file, unlike what happened when using 32 kbps 22
KS sampling WMA, as one might expect. But I didn't find an option to
convert to a 32 kpbs MONO 22KS file format, only stereo, which renders the
comparison useless. So unless you can convert some stereo WAV file (with
some music) to both MP3 and WMA formats at 32 kbps joint stereo, 22 KS, to
really push the bar, we may be stuck here. Again, I needed the 32 kbps AND
stereo for the reprocessed Command Performance broadcasts to get some
presence in the file (presence meaning like you feel you are right there in
the audience listening to the performances, instead of using just mono).
And as I pointed out in the other thread, IF you use a good pseudostereo
plug-in like PSP, the "presence" effect is quite good (on headphones), AND
it's pretty much mono-compatible, to boot.

I will say this, however. The loss of highs was noticeable in your mp3.
Not too bad, but definitely noticeable (I mean noticeably duller than the
source WAV file).

-----

BTW, if anybody else wants to jump into an audio editing and conversion
topic, feel free to do so at any time. :-) I'll post what I know.

"Bill in Co" wrote in
news
I will say this, however. The loss of highs was noticeable in your mp3.
Not too bad, but definitely noticeable (I mean noticeably duller than the
source WAV file).


I can't follow this up, I need to do other things, but I have to say that I
mentioned that. The conditions for a test have to be reasonable, I described
in extreme and repeated detail that I was showing that LAME does not default
to the best settings for low bitrate encoding, and described a change in
filtering that would REDUCE the high frequencies a bit more to gain bits for
a clean encode free of artifacting, such that a simple tone control can fix
the HF enough to restore legibility. So of course the MP3 sounds a bit
duller. Not only would I expect you to notice, I pointed it out, so there's
no new finding there.

As far as WMA goes, I'll NEVER use it, period. I will never waste my time
trying to prove someone else's case for a format I have stated good reasons
for me to avoid. Others can do what they want. That means I can do what I
want too. All I'm concerned with is that people become aware that MP3 can go
further then they likely expect. I did say I thought that it might contend
with WMA, but it really doesn't matter. WMA = smaller file = bigger playback
overheads. Doesn't matter HOW you cut it, this is true, and people pick the
compromises that suit them best. The reason that my finding matters is
simple: it offers improvements in the SAME context that many people already
choose.

"Bill in Co" wrote in
news
I will say this, however. The loss of highs was noticeable in your
mp3. Not too bad, but definitely noticeable (I mean noticeably duller
than the source WAV file).


Last poing from me, just closing the circle to where I came in on this.

An MP3 encoded slightly 'duller', with less HF, is far less objectionable
than artifacting in the mioddle of the audio band. It amounts to listening to
a loudspeaker off axis, at worst, something most of us do every day. We don't
even care, we don't notice until we decide to consciously test to see what
the on-axis sound is like. Now compare that MP3 method to the dreck often
posted in files TWICE the size. Come to that, compare ANY encoding wioth the
original WAV, at 32 kbps you're going to hear a difference. If WMA could
REALLY encode a stereo 32 kbps WAV so well that there is no difference, no-
one would be using LAME, never mind lossless compression. Obviously it's not
that good.

So, never mind the WMA thing which I thing is a red herring given that I
started out offering a way to make a LAME encode better than the default
offered, for anyone who does want to try the methods of reducing HF as a
price to pay for lower artifacting at low bitrates, here are two settings to
try:

LAME --preset 80 --resample 24 --lowpass 11

(My original, for getting FM broadcast quality from a CD source using only 80
kbps for modern high quality drama that includes complex batural soundtrack
and high quality music. Compare this with the defaukle LAME --preset 80, to
verify the method is viable! Note the default filter they use, and hear the
difference, you won't need to ABX test this, it's very evident.).

LAME.EXE --preset 32 --resample 22.05 --lowpass 12 --lowpass-width 6 -mm

(Recent modification to test at extreme compression, to 32 kbps. It's
designed for sources where HF at 9 KHz is below -48 dB, specifically designed
for old mono radio broadcasts, and uses a wide shallow filter that begind
slightly ABOVE the Nyquist frequency but still manages to avoid aliasing
because the source is already 'dull' as all such old recordings are. A lot of
people like those old shows, and most Usenet posts for them are either in
files so big that it's silly, or small ones that sound desperately poor,
often unlistenable. Worse, some people transcode the bad ones into BIG ones,
thinking they'll get better! THAT, ultimately, is the reason for my efforts.
There IS a better way, even without abandoning MP3.

Last point: The more you can do to restore a sound, or change it usefully,
AFTER the encode/decode stages, the smaller the file will be. Analogous to
Dolby, there is no reason a method like mine could not have been consistently
modelled and applied to ALL lossy formats as a way to save bits for storage
and transfer. I have taken that notion further with a method based on noise
reduction and harmonic restoration, but even in the simpler EQ form, it has
not been used as a standard as far as I know. Like noise shaping, such
methods seem to have taken a back seat in favour of psychoacoustics, which in
my view was the wrong way to go as it depends more on subtleties we can't
know, than on those we can, and was therefore dependent on large volume
statistical analyses. I chose methods that do not. WHile it won't make TINY
files, I strongly recommend WAVpack's lossy format for those who normally
prefer LAME at bitrates of 192 or better. It amazed me when I heard how well
it works with no psychoacoustics, and no artifacting of the kind we're
normally lumbered with when trying to compress audio. I haven't played around
with that like I did with MP3, but there may be scope for it. But I haven't
got the time. I'm way to good at wasting that already, which is why I'm
closing my side of this talk with a final complete statement on my methods
and intent.

Lostgallifreyan wrote:
"Bill in Co" wrote in
news
I will say this, however. The loss of highs was noticeable in your mp3.
Not too bad, but definitely noticeable (I mean noticeably duller than the
source WAV file).


I can't follow this up, I need to do other things, but I have to say that
I
mentioned that. The conditions for a test have to be reasonable, I
described
in extreme and repeated detail that I was showing that LAME does not
default
to the best settings for low bitrate encoding, and described a change in
filtering that would REDUCE the high frequencies a bit more to gain bits
for
a clean encode free of artifacting, such that a simple tone control can
fix
the HF enough to restore legibility. So of course the MP3 sounds a bit
duller. Not only would I expect you to notice, I pointed it out, so
there's
no new finding there.

Sorry, I didn't intend to imply that there was. I was just pointing out my
observations. I think you have already made your point that with tweaking
you can get some improvements over the default mp3 settings, and that was
never in dispute.

As far as WMA goes, I'll NEVER use it, period. I will never waste my time
trying to prove someone else's case for a format I have stated good
reasons
for me to avoid. Others can do what they want. That means I can do what I
want too. All I'm concerned with is that people become aware that MP3 can
go
further then they likely expect. I did say I thought that it might contend
with WMA, but it really doesn't matter. WMA = smaller file = bigger
playback
overheads. Doesn't matter HOW you cut it, this is true, and people pick
the
compromises that suit them best. The reason that my finding matters is
simple: it offers improvements in the SAME context that many people
already
choose.

OK. I just thought there was a chance you might want to at least try it,
and I was clearly wrong. I did indeed try yours out however. What I
found weird was that the 15 part WAV file didn't come out fully intact for
some reason, although it appeared to have equal length segments before being
combined.

Lostgallifreyan wrote:
"Bill in Co" wrote in
news
I will say this, however. The loss of highs was noticeable in your
mp3. Not too bad, but definitely noticeable (I mean noticeably duller
than the source WAV file).


Last poing from me, just closing the circle to where I came in on this.

An MP3 encoded slightly 'duller', with less HF, is far less objectionable
than artifacting in the mioddle of the audio band.

I agree completely.

It amounts to listening to
a loudspeaker off axis, at worst, something most of us do every day. We
don't
even care, we don't notice until we decide to consciously test to see what
the on-axis sound is like. Now compare that MP3 method to the dreck often
posted in files TWICE the size. Come to that, compare ANY encoding wioth
the
original WAV, at 32 kbps you're going to hear a difference. If WMA could
REALLY encode a stereo 32 kbps WAV so well that there is no difference,
no-
one would be using LAME, never mind lossless compression. Obviously it's
not
that good.

No, nothing is THAT good. I didn't intend to imply it was.

So, never mind the WMA thing which I thing is a red herring given that I
started out offering a way to make a LAME encode better than the default
offered, for anyone who does want to try the methods of reducing HF as a
price to pay for lower artifacting at low bitrates, here are two settings
to
try:

LAME --preset 80 --resample 24 --lowpass 11

(My original, for getting FM broadcast quality from a CD source using only
80
kbps for modern high quality drama that includes complex batural
soundtrack
and high quality music. Compare this with the defaukle LAME --preset 80,
to
verify the method is viable! Note the default filter they use, and hear
the
difference, you won't need to ABX test this, it's very evident.).

LAME.EXE --preset 32 --resample 22.05 --lowpass 12 --lowpass-width 6 -mm

(Recent modification to test at extreme compression, to 32 kbps. It's
designed for sources where HF at 9 KHz is below -48 dB, specifically
designed
for old mono radio broadcasts, and uses a wide shallow filter that begind
slightly ABOVE the Nyquist frequency but still manages to avoid aliasing
because the source is already 'dull' as all such old recordings are. A lot
of
people like those old shows, and most Usenet posts for them are either in
files so big that it's silly, or small ones that sound desperately poor,
often unlistenable. Worse, some people transcode the bad ones into BIG
ones,
thinking they'll get better!

That is pretty pathetic. But I've seen evidence of that, too, in some of
the questions I've seen posted (i.e. some people think they can improve the
quality by reencoding the original at a higher bitrate - go figure!).

THAT, ultimately, is the reason for my efforts.
There IS a better way, even without abandoning MP3.

And that was never in dispute. :-)

Last point: The more you can do to restore a sound, or change it usefully,
AFTER the encode/decode stages, the smaller the file will be. Analogous to
Dolby, there is no reason a method like mine could not have been
consistently
modelled and applied to ALL lossy formats as a way to save bits for
storage
and transfer. I have taken that notion further with a method based on
noise
reduction and harmonic restoration, but even in the simpler EQ form, it
has
not been used as a standard as far as I know. Like noise shaping, such
methods seem to have taken a back seat in favour of psychoacoustics, which
in
my view was the wrong way to go as it depends more on subtleties we can't
know, than on those we can, and was therefore dependent on large volume
statistical analyses. I chose methods that do not.

I don't know why either.
I am impressed, however, that you have taken the time to go into all of this
stuff on your own.
At any rate, I'm sorry if I offended you (by trying to compare mp3 with wma
in the quest for best compressibility) which wasn't my intent, but we can
leave it at that. I just found those tests to be interesting, and thought
I'd share my observations.

"Bill in Co" wrote in
news
OK. I just thought there was a chance you might want to at least try
it, and I was clearly wrong. I did indeed try yours out however.
What I found weird was that the 15 part WAV file didn't come out fully
intact for some reason, although it appeared to have equal length
segments before being combined.



No WMA for me. Nope. Never.

About the WAV bits, I don't know. I sent it via a btinternet server using
PowerPost, which reported it sent ok, and I just looked on the test group
with Xnews, and it's all there too. I can't check any further, but if you
can't see it all it means some propogation failure beyond my control
happened. Anyway, you don'ty need all of it, you heard some, which is enough.
I only set it to 52 seconds for a neat break in the stuff being said in the
show..

(I might have suggested a good binaries grabber but I don't know of one. I
wanted one to rival PowerPost, and there is actually a PowerGrab but it's
horrible, only the names seem anything like related to each other. I use
Xnews for binaries, except when a freind lets me use an Easynews account for
big stuff).

"Bill in Co" wrote in
:

Lostgallifreyan wrote:
"Bill in Co" wrote in
news
I will say this, however. The loss of highs was noticeable in your
mp3. Not too bad, but definitely noticeable (I mean noticeably duller
than the source WAV file).


Last poing from me, just closing the circle to where I came in on this.

An MP3 encoded slightly 'duller', with less HF, is far less
objectionable than artifacting in the mioddle of the audio band.

I agree completely.

It amounts to listening to
a loudspeaker off axis, at worst, something most of us do every day. We
don't
even care, we don't notice until we decide to consciously test to see
what the on-axis sound is like. Now compare that MP3 method to the
dreck often posted in files TWICE the size. Come to that, compare ANY
encoding wioth the
original WAV, at 32 kbps you're going to hear a difference. If WMA
could REALLY encode a stereo 32 kbps WAV so well that there is no
difference, no-
one would be using LAME, never mind lossless compression. Obviously
it's not
that good.

No, nothing is THAT good. I didn't intend to imply it was.

So, never mind the WMA thing which I thing is a red herring given that
I started out offering a way to make a LAME encode better than the
default offered, for anyone who does want to try the methods of
reducing HF as a price to pay for lower artifacting at low bitrates,
here are two settings to
try:

LAME --preset 80 --resample 24 --lowpass 11

(My original, for getting FM broadcast quality from a CD source using
only 80
kbps for modern high quality drama that includes complex batural
soundtrack
and high quality music. Compare this with the defaukle LAME --preset
80, to
verify the method is viable! Note the default filter they use, and hear
the
difference, you won't need to ABX test this, it's very evident.).

LAME.EXE --preset 32 --resample 22.05 --lowpass 12 --lowpass-width 6
-mm

(Recent modification to test at extreme compression, to 32 kbps. It's
designed for sources where HF at 9 KHz is below -48 dB, specifically
designed
for old mono radio broadcasts, and uses a wide shallow filter that
begind slightly ABOVE the Nyquist frequency but still manages to avoid
aliasing because the source is already 'dull' as all such old
recordings are. A lot of
people like those old shows, and most Usenet posts for them are either
in files so big that it's silly, or small ones that sound desperately
poor, often unlistenable. Worse, some people transcode the bad ones
into BIG ones,
thinking they'll get better!

That is pretty pathetic. But I've seen evidence of that, too, in some
of the questions I've seen posted (i.e. some people think they can
improve the quality by reencoding the original at a higher bitrate - go
figure!).

THAT, ultimately, is the reason for my efforts.
There IS a better way, even without abandoning MP3.

And that was never in dispute. :-)

Last point: The more you can do to restore a sound, or change it
usefully, AFTER the encode/decode stages, the smaller the file will be.
Analogous to Dolby, there is no reason a method like mine could not
have been consistently
modelled and applied to ALL lossy formats as a way to save bits for
storage
and transfer. I have taken that notion further with a method based on
noise
reduction and harmonic restoration, but even in the simpler EQ form, it
has
not been used as a standard as far as I know. Like noise shaping, such
methods seem to have taken a back seat in favour of psychoacoustics,
which in
my view was the wrong way to go as it depends more on subtleties we
can't know, than on those we can, and was therefore dependent on large
volume statistical analyses. I chose methods that do not.

I don't know why either.
I am impressed, however, that you have taken the time to go into all of
this stuff on your own.
At any rate, I'm sorry if I offended you (by trying to compare mp3 with
wma in the quest for best compressibility) which wasn't my intent, but
we can leave it at that. I just found those tests to be interesting,
and thought I'd share my observations.




No worries. I wasn;t lookign either for agreement or competition, is all,
just to put that MP3 thing out because I think I likely won't be the only one
who gets something out of it. The whole thing about comparisons and testing
gets to the point where I learned to stay with one direct assertion and stay
there. The whole debate about LAME settings used to get beyond imagining. I
learned to stay well out of it. Part of the reason for my posting of LAME
settings is to show that despite YEARS of frenetic wrangling between self-
professed gurus and audiophiles, there are STILL easy ways for novices to get
better results with it.

Lostgallifreyan wrote:
"Bill in Co" wrote in
news
OK. I just thought there was a chance you might want to at least try
it, and I was clearly wrong. I did indeed try yours out however.
What I found weird was that the 15 part WAV file didn't come out fully
intact for some reason, although it appeared to have equal length
segments before being combined.



No WMA for me. Nope. Never.

About the WAV bits, I don't know. I sent it via a btinternet server using
PowerPost, which reported it sent ok, and I just looked on the test group
with Xnews, and it's all there too.

Yup. All 15 parts show up in the listing.

I can't check any further, but if you

That's ok.

can't see it all it means some propogation failure beyond my control
happened. Anyway, you don'ty need all of it, you heard some, which is
enough.
I only set it to 52 seconds for a neat break in the stuff being said in
the
show..

No, I did see all 15 parts, and they appeared to be ok from the listings
(and with equal reported lengths), but when decoded and combined into one
wav file, it came out to be only 29 seconds long when played, and it wasn't
contiguous. I mean there were jumps in it (or parts skipped in it) when it
was played (really weird). But I got the gist of it, and it was a bit
humorous, too.

(I might have suggested a good binaries grabber but I don't know of one. I
wanted one to rival PowerPost, and there is actually a PowerGrab but it's
horrible, only the names seem anything like related to each other. I use
Xnews for binaries, except when a freind lets me use an Easynews account
for
big stuff).

Lostgallifreyan wrote:
"Bill in Co" wrote in
:

Lostgallifreyan wrote:
"Bill in Co" wrote in
news
I will say this, however. The loss of highs was noticeable in your
mp3. Not too bad, but definitely noticeable (I mean noticeably duller
than the source WAV file).


Last poing from me, just closing the circle to where I came in on this.

An MP3 encoded slightly 'duller', with less HF, is far less
objectionable than artifacting in the mioddle of the audio band.

I agree completely.

It amounts to listening to
a loudspeaker off axis, at worst, something most of us do every day. We
don't
even care, we don't notice until we decide to consciously test to see
what the on-axis sound is like. Now compare that MP3 method to the
dreck often posted in files TWICE the size. Come to that, compare ANY
encoding wioth the
original WAV, at 32 kbps you're going to hear a difference. If WMA
could REALLY encode a stereo 32 kbps WAV so well that there is no
difference, no-
one would be using LAME, never mind lossless compression. Obviously
it's not
that good.

No, nothing is THAT good. I didn't intend to imply it was.

So, never mind the WMA thing which I thing is a red herring given that
I started out offering a way to make a LAME encode better than the
default offered, for anyone who does want to try the methods of
reducing HF as a price to pay for lower artifacting at low bitrates,
here are two settings to
try:

LAME --preset 80 --resample 24 --lowpass 11

(My original, for getting FM broadcast quality from a CD source using
only 80
kbps for modern high quality drama that includes complex batural
soundtrack
and high quality music. Compare this with the defaukle LAME --preset
80, to
verify the method is viable! Note the default filter they use, and hear
the
difference, you won't need to ABX test this, it's very evident.).

LAME.EXE --preset 32 --resample 22.05 --lowpass 12 --lowpass-width 6
-mm

(Recent modification to test at extreme compression, to 32 kbps. It's
designed for sources where HF at 9 KHz is below -48 dB, specifically
designed
for old mono radio broadcasts, and uses a wide shallow filter that
begind slightly ABOVE the Nyquist frequency but still manages to avoid
aliasing because the source is already 'dull' as all such old
recordings are. A lot of
people like those old shows, and most Usenet posts for them are either
in files so big that it's silly, or small ones that sound desperately
poor, often unlistenable. Worse, some people transcode the bad ones
into BIG ones,
thinking they'll get better!

That is pretty pathetic. But I've seen evidence of that, too, in some
of the questions I've seen posted (i.e. some people think they can
improve the quality by reencoding the original at a higher bitrate - go
figure!).

THAT, ultimately, is the reason for my efforts.
There IS a better way, even without abandoning MP3.

And that was never in dispute. :-)

Last point: The more you can do to restore a sound, or change it
usefully, AFTER the encode/decode stages, the smaller the file will be.
Analogous to Dolby, there is no reason a method like mine could not
have been consistently
modelled and applied to ALL lossy formats as a way to save bits for
storage
and transfer. I have taken that notion further with a method based on
noise
reduction and harmonic restoration, but even in the simpler EQ form, it
has
not been used as a standard as far as I know. Like noise shaping, such
methods seem to have taken a back seat in favour of psychoacoustics,
which in
my view was the wrong way to go as it depends more on subtleties we
can't know, than on those we can, and was therefore dependent on large
volume statistical analyses. I chose methods that do not.

I don't know why either.
I am impressed, however, that you have taken the time to go into all of
this stuff on your own.
At any rate, I'm sorry if I offended you (by trying to compare mp3 with
wma in the quest for best compressibility) which wasn't my intent, but
we can leave it at that. I just found those tests to be interesting,
and thought I'd share my observations.




No worries. I wasn;t lookign either for agreement or competition, is all,
just to put that MP3 thing out because I think I likely won't be the only
one
who gets something out of it. The whole thing about comparisons and
testing
gets to the point where I learned to stay with one direct assertion and
stay
there. The whole debate about LAME settings used to get beyond imagining.
I
learned to stay well out of it. Part of the reason for my posting of LAME
settings is to show that despite YEARS of frenetic wrangling between self-
professed gurus and audiophiles, there are STILL easy ways for novices to
get
better results with it.

And they keep coming out with lame revision updates, but I think I'm still
using an older version of 3.93, but I can't recall now. Plus it depends on
which program I'm using, as some install a DLL into their own program
directories, which overrides the one I have in the system folder.

Maybe you should submit some of your presets to the Lame group, and see if
they might want to add them (like that one for broadcast AM).

"Bill in Co" wrote in
:

And they keep coming out with lame revision updates, but I think I'm
still using an older version of 3.93, but I can't recall now. Plus it
depends on which program I'm using, as some install a DLL into their own
program directories, which overrides the one I have in the system
folder.

Maybe you should submit some of your presets to the Lame group, and see
if they might want to add them (like that one for broadcast AM).


Oh no.. I wouldn't do that. The moment I do that it quickly becomes insane. I
mean, there were just two of us, and there was potential for
misunderstanding. Now imagine a bunch of people all with cat avatars on
Hydrogenaudio all vying for their place in the pyramid. If a person wants to
get shot down in flames, by all means go there, otherwise just try things,
based on methods we can all use. LAME's development all but got derailed due
to the politicking and such at one point, and the core developers had to put
distance between themselves and some of those fighting to establish their
ideas of the best preset. The fact that I can show a variation worth trying
is as much to do with that as anything else, it's clear;ly not all about real
science, and being psychoacoustically based it's hard to be such anyway, as
statistical bases are more useful than any 'objective'.

In short, I have a better chanmce of beign heard by ONE thoughtful individual
here, than there. I also posted the method (and some files made by it) in the
radio groups I hung out in for a while. There were no complaints, but I doubt
I changed their world either.

LAME versions changem, so presets also change. I'm using 3.98.2, and I think
those presets may have been the same in v3.93 but I don't know. ABR presets
got improved, I know that much, but not what releases were involved in that
change at its greatest shift. I explored it mainly because it had improved
while I wasn't looking, before.


About the WAV file, I grabbed it to see if it worked ok. It did. If yours is
9,289,832 bytes, try for an MD5 checksum of C753CA7E193CFF576D3D1616F91F65EF.
I don't need to know the outcome, but you might. It's possible that your
client is breaking it, but I don't know. Checking for integrity is all I can
think of to try.

Got to go... but not as long as Captain Oates. This year I have no appetite
for cold weather anyway.