Side-by-side comparisons

[[email protected]]

Hi. I was wondering what workflow ya'll use in the following situation. I have a few versions of the same song that was extracted from a DAT. A recording from a couple decades ago. One transfer was done by me, and I'm sure it was digital to digital. But another version was transferred by a company that I had paid to do the transfer. I redid the transfer because there were some dropouts in my version.

Anyway, the version from the professional house is louder than mine (!). They assured me they'd do a digital to digital transfer, but I'm not so sure now.

I'd like to play back both songs and be able to flip back and forth between the two. I'd then adjust the volume to get them to match. Then after some repeated back and forths, I'd choose the version I like better.

So I'm asking, in general what software would you use to flip back and forth? In the world of digital filmmaking, a colorizers would have a line down the middle of the screen to compare colors of two versions. I suppose I could load both tracks into Logic or something, and solo one track to flip around.

Finally, I'm wondering if there's a sort of quality metric to tell me if a transfer went through the analog domain, so I can just discard that, without having to trust my ears. (The song will go through more work, so it's not done at this stage.)

Thanks,
David

[Peter Larsen[_3_]]

wrote:

> Hi. I was wondering what workflow ya'll use in the following
> situation.

You have a known good digital version with a few dropouts and a version
without dropouts. I would then match level and repair the dropouts by
pasting in short segments from the other version and think no more.

> I have a few versions of the same song that was extracted
> from a DAT. A recording from a couple decades ago. One transfer was
> done by me, and I'm sure it was digital to digital. But another
> version was transferred by a company that I had paid to do the
> transfer. I redid the transfer because there were some dropouts in
> my version.

> Anyway, the version from the professional house is louder than mine
> (!). They assured me they'd do a digital to digital transfer, but
> I'm not so sure now.

They may have normalized to zero or to slightly below. If you want to waste
time comparing then you analyze for average level and match those.

> I'd like to play back both songs and be able to flip back and forth
> between the two.

I expect any daw software out there to be able to have more than one file
open, if nothing else works then load them in multitrack view and solo as
relevant.

> I'd then adjust the volume to get them to match.
> Then after some repeated back and forths, I'd choose the version I
> like better.

It is a hopeless quest since you will never know for sure which is
"correct", you have a known good digital transfer that is 99.9 percent
complete, patch the dropouts with level matched segments from the other
version.

> So I'm asking, in general what software would you use to flip back
> and forth?

I would use some incarnation of Cool Edit, Audition included, because that
is what I have installed and in daily use.

> In the world of digital filmmaking, a colorizers would
> have a line down the middle of the screen to compare colors of two
> versions. I suppose I could load both tracks into Logic or
> something, and solo one track to flip around.

> Finally, I'm wondering if there's a sort of quality metric to tell me
> if a transfer went through the analog domain, so I can just discard
> that, without having to trust my ears. (The song will go through
> more work, so it's not done at this stage.)

You could look at noisefloor, but I do not think that you will find it easy
on your present proficiency level, so don't bother.

Once upon a time I asked a designer of digital hardware if my observation of
digital transfers from different DAT machines sounding different could be
correct, after a brief pause he said "Yes, that is thinkable ...."

Or just use the one that has no dropouts since you say a slight quality loss
is of no consequence. Being productive is some of the time also about
getting it done and move on.

> Thanks,
> David

Kind regards

Peter Larsen

[Scott Dorsey]

> wrote:
>Anyway, the version from the professional house is louder than mine (!). T=
>hey assured me they'd do a digital to digital transfer, but I'm not so sure=
> now.

If it were me, I'd measure it. Locate a peak in both, line them up,
measure the peak level. Same or different? Extract a second's worth of
data, do a diff. Same or different? Actually print the data out from a
bitscope application. Same or different?

They may well have done a digital to digital transfer and normalized. You
can do the same thing yourself.

>I'd like to play back both songs and be able to flip back and forth between=
> the two. I'd then adjust the volume to get them to match. Then after som=
>e repeated back and forths, I'd choose the version I like better.

Why not stick with your own transfer, since you know the provenance? Your
problem then only becomes figuring out what they did with theirs.

>Finally, I'm wondering if there's a sort of quality metric to tell me if a =
>transfer went through the analog domain, so I can just discard that, withou=
>t having to trust my ears. (The song will go through more work, so it's no=
>t done at this stage.)

Not really, but if the data from the bitscope is totally different and not
just scaled, it might well have gone through an analogue transfer. Still,
what you have in the best of worlds is a clean digital transfer without
dropouts, and then a digital transfer with dropouts and with some process
done to make it louder. Either way that makes me immediately want to go to
your transfer.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Peter Larsen[_3_]]

Scott Dorsey wrote:

> > wrote:

> Not really, but if the data from the bitscope is totally different
> and not just scaled, it might well have gone through an analogue
> transfer. Still, what you have in the best of worlds is a clean
> digital transfer without dropouts, and then a digital transfer with
> dropouts and with some process done to make it louder. Either way
> that makes me immediately want to go to your transfer.

Scott, it was the OP's own digital transfer that had dropouts, which is why
I suggested pasting from the outsourced transfer to repair it.

> --scott

Kind regards

Peter Larsen

[Scott Dorsey]

Peter Larsen > wrote:
>Scott Dorsey wrote:
>
>> > wrote:
>
>> Not really, but if the data from the bitscope is totally different
>> and not just scaled, it might well have gone through an analogue
>> transfer. Still, what you have in the best of worlds is a clean
>> digital transfer without dropouts, and then a digital transfer with
>> dropouts and with some process done to make it louder. Either way
>> that makes me immediately want to go to your transfer.
>
>Scott, it was the OP's own digital transfer that had dropouts, which is why
>I suggested pasting from the outsourced transfer to repair it.

Oh, well.... in that case, get a PM done on your deck and try again!
DAT machines require regular cleaning, alignment, and replacement of
rubber parts just like analogue machines do.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[[email protected]]

Hi Peter, thanks for your response.

As you suggest, I'm taking a pragmatic approach here.

After manual level-matching and listening a bit, the recordings sound pretty much the same to my (amateur) ears. I wonder if the house that transferred my DAT normalized the result, thinking they were doing me a favor. I've asked them to clarify their exact procedure, for peace of mind.

You mention looking at noisefloor. Out of curiosity, can you expand on this? Is this anything more then trying to find very quiet parts of the recording and looking / listening at the noise? (Not that this is straightforward.)

Thanks,
David

On Saturday, December 28, 2013 3:25:46 PM UTC-5, Peter Larsen wrote:
> wrote:
>
>
>
> > Hi. I was wondering what workflow ya'll use in the following
>
> > situation.
>
>
>
> You have a known good digital version with a few dropouts and a version
>
> without dropouts. I would then match level and repair the dropouts by
>
> pasting in short segments from the other version and think no more.
>
>
>
> > I have a few versions of the same song that was extracted
>
> > from a DAT. A recording from a couple decades ago. One transfer was
>
> > done by me, and I'm sure it was digital to digital. But another
>
> > version was transferred by a company that I had paid to do the
>
> > transfer. I redid the transfer because there were some dropouts in
>
> > my version.
>
>
>
> > Anyway, the version from the professional house is louder than mine
>
> > (!). They assured me they'd do a digital to digital transfer, but
>
> > I'm not so sure now.
>
>
>
> They may have normalized to zero or to slightly below. If you want to waste
>
> time comparing then you analyze for average level and match those.
>
>
>
> > I'd like to play back both songs and be able to flip back and forth
>
> > between the two.
>
>
>
> I expect any daw software out there to be able to have more than one file
>
> open, if nothing else works then load them in multitrack view and solo as
>
> relevant.
>
>
>
> > I'd then adjust the volume to get them to match.
>
> > Then after some repeated back and forths, I'd choose the version I
>
> > like better.
>
>
>
> It is a hopeless quest since you will never know for sure which is
>
> "correct", you have a known good digital transfer that is 99.9 percent
>
> complete, patch the dropouts with level matched segments from the other
>
> version.
>
>
>
> > So I'm asking, in general what software would you use to flip back
>
> > and forth?
>
>
>
> I would use some incarnation of Cool Edit, Audition included, because that
>
> is what I have installed and in daily use.
>
>
>
> > In the world of digital filmmaking, a colorizers would
>
> > have a line down the middle of the screen to compare colors of two
>
> > versions. I suppose I could load both tracks into Logic or
>
> > something, and solo one track to flip around.
>
>
>
> > Finally, I'm wondering if there's a sort of quality metric to tell me
>
> > if a transfer went through the analog domain, so I can just discard
>
> > that, without having to trust my ears. (The song will go through
>
> > more work, so it's not done at this stage.)
>
>
>
> You could look at noisefloor, but I do not think that you will find it easy
>
> on your present proficiency level, so don't bother.
>
>
>
> Once upon a time I asked a designer of digital hardware if my observation of
>
> digital transfers from different DAT machines sounding different could be
>
> correct, after a brief pause he said "Yes, that is thinkable ...."
>
>
>
> Or just use the one that has no dropouts since you say a slight quality loss
>
> is of no consequence. Being productive is some of the time also about
>
> getting it done and move on.
>
>
>
> > Thanks,
>
> > David
>
>
>
> Kind regards
>
>
>
> Peter Larsen

[[email protected]]

Hi Scott, thanks for your reply.

Yah, I have the urge to extract a second from each recording and then seeing if the values are identical but for a constant factor. I'm with Peter that this is overkill, but at the same time, I'd like to learn more and extend my skills, so I'd like to do this analysis as a rainy day project.

Are you aware of any software that would facilitate such an analysis? A quick search for "bitscope" yields hardware, where all I really need is a purely software-based analysis.

Thanks,
David

On Sunday, December 29, 2013 8:55:55 AM UTC-5, Scott Dorsey wrote:
> > wrote:
>
> >Anyway, the version from the professional house is louder than mine (!). T=
>
> >hey assured me they'd do a digital to digital transfer, but I'm not so sure=
>
> > now.
>
>
>
> If it were me, I'd measure it. Locate a peak in both, line them up,
>
> measure the peak level. Same or different? Extract a second's worth of
>
> data, do a diff. Same or different? Actually print the data out from a
>
> bitscope application. Same or different?
>
>
>
> They may well have done a digital to digital transfer and normalized. You
>
> can do the same thing yourself.
>
>
>
> >I'd like to play back both songs and be able to flip back and forth between=
>
> > the two. I'd then adjust the volume to get them to match. Then after som=
>
> >e repeated back and forths, I'd choose the version I like better.
>
>
>
> Why not stick with your own transfer, since you know the provenance? Your
>
> problem then only becomes figuring out what they did with theirs.
>
>
>
> >Finally, I'm wondering if there's a sort of quality metric to tell me if a =
>
> >transfer went through the analog domain, so I can just discard that, withou=
>
> >t having to trust my ears. (The song will go through more work, so it's no=
>
> >t done at this stage.)
>
>
>
> Not really, but if the data from the bitscope is totally different and not
>
> just scaled, it might well have gone through an analogue transfer. Still,
>
> what you have in the best of worlds is a clean digital transfer without
>
> dropouts, and then a digital transfer with dropouts and with some process
>
> done to make it louder. Either way that makes me immediately want to go to
>
> your transfer.
>
> --scott
>
>
>
> --
>
> "C'est un Nagra. C'est suisse, et tres, tres precis."

[[email protected]]

By the way, I wanted to say thanks for your responses again. I lurked on this newsgroup well over a decade ago and remember giving your posts special attention. I immediately thought of rec.audio.pro as the right place to take my question. David

On Sunday, December 29, 2013 12:08:02 PM UTC-5, Scott Dorsey wrote:
> Peter Larsen > wrote:
>
> >Scott Dorsey wrote:
>
> >
>
> >> > wrote:
>
> >
>
> >> Not really, but if the data from the bitscope is totally different
>
> >> and not just scaled, it might well have gone through an analogue
>
> >> transfer. Still, what you have in the best of worlds is a clean
>
> >> digital transfer without dropouts, and then a digital transfer with
>
> >> dropouts and with some process done to make it louder. Either way
>
> >> that makes me immediately want to go to your transfer.
>
> >
>
> >Scott, it was the OP's own digital transfer that had dropouts, which is why
>
> >I suggested pasting from the outsourced transfer to repair it.
>
>
>
> Oh, well.... in that case, get a PM done on your deck and try again!
>
> DAT machines require regular cleaning, alignment, and replacement of
>
> rubber parts just like analogue machines do.
>
> --scott
>
>
>
> --
>
> "C'est un Nagra. C'est suisse, et tres, tres precis."

[John Williamson]

On 29/12/2013 19:22, wrote:
> Hi Scott, thanks for your reply.
>
> Yah, I have the urge to extract a second from each recording and then seeing if the values are identical but for a constant factor. I'm with Peter that this is overkill, but at the same time, I'd like to learn more and extend my skills, so I'd like to do this analysis as a rainy day project.
>
> Are you aware of any software that would facilitate such an analysis? A quick search for "bitscope" yields hardware, where all I really need is a purely software-based analysis.
>
Any multi track DAW (Audacity is free, I use that or Adobe Audition)
will let you copy the same second from the two files. You adjust the
snippet timings on two tracks of the DAW so that the peaks on the
waveform display exactly coincide, if you then invert one of the
snippets, you can then mix down the two tracks to hear and see the
difference. It would be helpful if you normalise both snippets to the
same peak level before inverting one.

Which is just a description of how to do what Scott suggested.


--
Tciao for Now!

John.

[[email protected]]

I love it, that will work quite easily. The goal, of course, is no signal, visually and aurally.

You're the second person to mention Audition. I've been using Logic, as I think it has stronger MIDI support, which I need for other purposes. But perhaps I should look into Audition since it's so popular here.

d

On Sunday, December 29, 2013 3:45:24 PM UTC-5, John Williamson wrote:
> On 29/12/2013 19:22, wrote:
>
> > Hi Scott, thanks for your reply.
>
> >
>
> > Yah, I have the urge to extract a second from each recording and then seeing if the values are identical but for a constant factor. I'm with Peter that this is overkill, but at the same time, I'd like to learn more and extend my skills, so I'd like to do this analysis as a rainy day project.
>
> >
>
> > Are you aware of any software that would facilitate such an analysis? A quick search for "bitscope" yields hardware, where all I really need is a purely software-based analysis.
>
> >
>
> Any multi track DAW (Audacity is free, I use that or Adobe Audition)
>
> will let you copy the same second from the two files. You adjust the
>
> snippet timings on two tracks of the DAW so that the peaks on the
>
> waveform display exactly coincide, if you then invert one of the
>
> snippets, you can then mix down the two tracks to hear and see the
>
> difference. It would be helpful if you normalise both snippets to the
>
> same peak level before inverting one.
>
>
>
> Which is just a description of how to do what Scott suggested.
>
>
>
>
>
> --
>
> Tciao for Now!
>
>
>
> John.

[[email protected]]

By the way, in my first post when I referred to statistical analysis, what I was thinking was the following:

Imagine you concatenated one second from one recording after one second from the other recording. A compression algorithm with a window size of at least one second should be able to represent the entire two seconds with only an epsilon more information (to account for scaling).

d

On Sunday, December 29, 2013 4:01:51 PM UTC-5, wrote:
> I love it, that will work quite easily. The goal, of course, is no signal, visually and aurally.
>
>
>
> You're the second person to mention Audition. I've been using Logic, as I think it has stronger MIDI support, which I need for other purposes. But perhaps I should look into Audition since it's so popular here.
>
>
>
> d
>
>
>
> On Sunday, December 29, 2013 3:45:24 PM UTC-5, John Williamson wrote:
>
> > On 29/12/2013 19:22, wrote:
>
> >
>
> > > Hi Scott, thanks for your reply.
>
> >
>
> > >
>
> >
>
> > > Yah, I have the urge to extract a second from each recording and then seeing if the values are identical but for a constant factor. I'm with Peter that this is overkill, but at the same time, I'd like to learn more and extend my skills, so I'd like to do this analysis as a rainy day project.
>
> >
>
> > >
>
> >
>
> > > Are you aware of any software that would facilitate such an analysis? A quick search for "bitscope" yields hardware, where all I really need is a purely software-based analysis.
>
> >
>
> > >
>
> >
>
> > Any multi track DAW (Audacity is free, I use that or Adobe Audition)
>
> >
>
> > will let you copy the same second from the two files. You adjust the
>
> >
>
> > snippet timings on two tracks of the DAW so that the peaks on the
>
> >
>
> > waveform display exactly coincide, if you then invert one of the
>
> >
>
> > snippets, you can then mix down the two tracks to hear and see the
>
> >
>
> > difference. It would be helpful if you normalise both snippets to the
>
> >
>
> > same peak level before inverting one.
>
> >
>
> >
>
> >
>
> > Which is just a description of how to do what Scott suggested.
>
> >
>
> >
>
> >
>
> >
>
> >
>
> > --
>
> >
>
> > Tciao for Now!
>
> >
>
> >
>
> >
>
> > John.

[John Williamson]

On 29/12/2013 21:01, wrote:
> I love it, that will work quite easily. The goal, of course, is no signal, visually and aurally.
>
Yup.

> You're the second person to mention Audition. I've been using Logic, as I think it has stronger MIDI support, which I need for other purposes. But perhaps I should look into Audition since it's so popular here.
>
I use it for the same reason I use Photoshop. A lot of people I know use
it as well, and it's easy to swap projects between computers. One
downside of the latest vesions is that they've switched to per year
licencing fees, rather than the previous model of install and forget.

It's easier to use than Audacity, but as Audacity is free, that's what I
used before I started on Audition.

Logic can be used to do the same job as easily, it's a bread and butter
job for any DAW.

--
Tciao for Now!

John.

[John Williamson]

On 29/12/2013 21:05, wrote:
> By the way, in my first post when I referred to statistical analysis, what I was thinking was the following:
>
> Imagine you concatenated one second from one recording after one second from the other recording. A compression algorithm with a window size of at least one second should be able to represent the entire two seconds with only an epsilon more information (to account for scaling). d
>
What would you be hoping to gain from that? I'd only need statistical
analysis of any depth for comparing things like the absolute loudness of
two files averaged over a period. If you need the levels to meet a
specification, then you would be onto a winner, but if you're not, it's
extra work for nothing. IMHO, and so on.

--
Tciao for Now!

John.

[[email protected]]

Funnily enough, I know Dominic Mazzoni, author of Audacity. We jammed a few times (jazz) while we were both at Carnegie Mellon. (We're both at Google now.) d

On Sunday, December 29, 2013 4:13:50 PM UTC-5, John Williamson wrote:
> On 29/12/2013 21:01, wrote:
>
> > I love it, that will work quite easily. The goal, of course, is no signal, visually and aurally.
>
> >
>
> Yup.
>
>
>
> > You're the second person to mention Audition. I've been using Logic, as I think it has stronger MIDI support, which I need for other purposes. But perhaps I should look into Audition since it's so popular here.
>
> >
>
> I use it for the same reason I use Photoshop. A lot of people I know use
>
> it as well, and it's easy to swap projects between computers. One
>
> downside of the latest vesions is that they've switched to per year
>
> licencing fees, rather than the previous model of install and forget.
>
>
>
> It's easier to use than Audacity, but as Audacity is free, that's what I
>
> used before I started on Audition.
>
>
>
> Logic can be used to do the same job as easily, it's a bread and butter
>
> job for any DAW.
>
>
>
> --
>
> Tciao for Now!
>
>
>
> John.

[[email protected]]

I was thinking that a standard toolkit of such analyses would save time: no need to align peaks, normalize, invert, etc. I could just feed in two files and look at some summary numbers. Worth building if I would need to do this sort of comparison lots of times. Otherwise, manual inspection is fine. d

On Sunday, December 29, 2013 4:18:23 PM UTC-5, John Williamson wrote:
> On 29/12/2013 21:05, wrote:
>
> > By the way, in my first post when I referred to statistical analysis, what I was thinking was the following:
>
> >
>
> > Imagine you concatenated one second from one recording after one second from the other recording. A compression algorithm with a window size of at least one second should be able to represent the entire two seconds with only an epsilon more information (to account for scaling). d
>
> >
>
> What would you be hoping to gain from that? I'd only need statistical
>
> analysis of any depth for comparing things like the absolute loudness of
>
> two files averaged over a period. If you need the levels to meet a
>
> specification, then you would be onto a winner, but if you're not, it's
>
> extra work for nothing. IMHO, and so on.
>
>
>
> --
>
> Tciao for Now!
>
>
>
> John.

[Scott Dorsey]

> wrote:
>
>Are you aware of any software that would facilitate such an analysis? A qu=
>ick search for "bitscope" yields hardware, where all I really need is a pur=
>ely software-based analysis.

My old Sonic workstation has a function to do this.... surely there is some
tool on more modern DAW gear.

Worst case, extract it to a .wav file and use diff and od -h on the command
line.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Scott Dorsey]

> wrote:
>I was thinking that a standard toolkit of such analyses would save time: no=
> need to align peaks, normalize, invert, etc. I could just feed in two fil=
>es and look at some summary numbers. Worth building if I would need to do =
>this sort of comparison lots of times. Otherwise, manual inspection is fin=
>e. d

I bet the forensic audio guys have something like that. Wes Dooley would know.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Luxey]

Any daw will spit out max sample value and average loudness of a file. If levels are miss matched ... After leveling, try the nulling, if the residue is 0 ..., if not, maybe it's at dither level...

[[email protected]]

I like the "diff" and "od" ideas. :) Thanks.

[[email protected]]

On Tuesday, December 31, 2013 6:57:53 AM UTC-5, Luxey wrote:
> Any daw will spit out max sample value and average loudness of a file. If levels are miss matched ... After leveling, try the nulling, if the residue is 0 ..., if not, maybe it's at dither level...

Thanks!

[Arny Krueger[_5_]]

> wrote in message
...

>So I'm asking, in general what software would you use to flip back and
>forth? In the world of digital filmmaking, a colorizers would have a line
>down the middle of the screen to compare colors of two versions. I suppose
>I could

> load both tracks into Logic or something, and solo one track to flip
> around.

Try the FOOBAR2000 music player with the ABX plug in. Compare A & B.

http://www.foobar2000.org/download

http://www.foobar2000.org/components/view/foo_abx