A better way to needledrop? (Using Voxengo Curve EQ)

Discussion in 'Audio Hardware' started by back2vinyl, Mar 11, 2014.

Thread Status:
Not open for further replies.
  1. kevintomb

    kevintomb Forum Resident


    No I think he is simply trying to get one thing to sound "tonally" like another.

    I do not think he is assuming anything, nor implied anything earth shaking.

    Overall frequency response and tonality are by far the main thing we hear when comparing 2 sound choices.
     
  2. vinyldoneright

    vinyldoneright pbthal

    Location:
    Ca
    Perhaps in your world that is by far the main thing, but not in mine. As someone who has probably done more or as many needledrops as anyone and has had them compared to other sources by LOTS of people I can say that is not the case in my world
     
  3. darkmass

    darkmass Forum Resident

    This will be fairly off-topic for this thread, but given that the general tone concerns modifying EQ as an accent to needledropping it won't be entirely unrelated. When you work with Blue, this could be an aid to you (though I wouldn't rule out that you may develop your own/better technique). And, oh, is this post going to be long! :)

    In our previous conversation I mentioned that towards the end of the first cut on Blue, "All I Want", there was an accentuation of the bass which seemed unique to that portion of the cut, otherwise I felt matching EQs over the entire selection would have addressed things. That is, I found that both Steve's DCC CD and the original Reprise vinyl had more bass in that specific region of the cut than the HDtracks contained. And, I mentioned that in working with the HDtracks version I had to bring up the bass in that specific region...I did not state at the time how I did that.

    Now my two most significant tools for working with sound (and video) are Sony Sound Forge Pro 11 and Sony Vegas Pro 12. Neither of them have any kind of time-based EQ modification, and that is also the case for my various other stand-alones and plug-ins. An EQ may only be applied overall.

    Here's my approach...

    If this represents the time-line of the "All I Want", from left to right, it can be thought of as:

    A-----------------------B-----C----D

    A is the selection start,

    B is immediately following the final vocal in the cut,

    B to C is a brief instrumental interval which is principally dulcimer,

    C to D is the emphasized bass guitar region, and

    D is the cut end.

    Now I didn't discover that the full cut equalization of the HDtracks version didn't work till I had precisely matched the overall HDtracks EQ to the DCC EQ for the cut. On playback and comparison of the HDtracks cut, it was clear that the C to D interval was not correct, even though the A to C region didn't seem bad at all. Looking at real-time EQ while I played the DCC cut, I could see peaking in the 90 Hz area within the C to D region. The peaking looked roughly one octave wide.

    I two made extra copies of the EQed HDtracks cut, and split the first copy at point C, throwing away the C to D interval. The second copy I split at point B, throwing away the A to B interval.

    Proportional fonts and compression of blank spaces will probably kill this, but what I had at that point for the HDtracks "All I Want" was:

    1) A-----------------------B-----C

    2) B-----C----D

    In the trimmed copy consisting of the B to D interval, I used Sound Forge and put in a parametric EQ gain centered at 90 Hz and made it one octave wide. I don't recall what parametric EQ gain dB value I ended up with, but I settled on a gain following some comparative listening to the DCC version. The dulcimer in the B to D region was unaffected by the parametric bump since the dulcimer's tones were all well above 90 Hz.

    Once I was happy with the parametric gain, I brought the A to C section and the EQ corrected B to D section into Vegas Pro as parallel tracks, carefully sliding the B to D track till the B to C areas of the respective tracks were exactly overlapping and aligned with each other. (With Vegas Pro, keeping the tracks in parallel allows great clarity of alignment when zooming way-in to a musical region...if they were to reside in the same track that clarity would be lost.) Once the proper alignment was clear, I dragged the B to D track on top of the A to C track keeping the precise overlap of the respective B to C regions. Once I had that, I applied a liner cross-fade within the overlapped B to C region. That is, when the A to C section was playing, the linear fade-out in the B to C portion of it was exactly compensated for by the linear fade-in within the B to C portion of the B to D segment. (It also meant that any bass occurring within the B to C portion would be faded-in to the level occurring within the C to D portion.)

    After I recorded the modified entire A to D section HDtracks of "All I Want" out, I once again did an overall EQ match of it to the original DCC version since I felt the bass bump had an effect on average EQs.

    This stuff probably reads more complicated than the concept actually is (and I sure hope I kept all my alphabet letters correct), so it'll likely take a reading or several. If there were a lot of such sections within a given cut I'm not sure I would bother with it. But perhaps if there were a lot of such sections within a given cut, the initial cut's over-all EQ matching would give reasonably acceptable results. At any rate, I considered this one more learning experience.

    I'm looking forward to hearing how things work out for you on Blue.
     
    Last edited: Mar 20, 2014
  4. darkmass

    darkmass Forum Resident

    Sometimes when I'm editing real time, I just don't do my best.

    Of course that should have been, ''I made two extra copies...".
     
  5. Ulli

    Ulli Forum Resident

    Location:
    Germany
    This is a very good idea that I'm applying all the time, also using Voxengo CurveEQ. Not always to transfer EQ characteristics from LP to CD, but that's one application.

    I go one step further by applying the EQ modifications on the fly (all my music is server-based) using convolution that is triggered by tags selecting the correct EQ impulse response file completely automatically. This method offers the exciting possibility of exchanging EQ profiles of recordings on this forum, by simply posting the small impulse response files (wav files containing only a few hundred samples). In fact, I'm in the middle of writing a long article on this topic for the forum, including several examples that demonstrate the power of this method. Stay tuned...
     
    Robert C and Dan Steely like this.
  6. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    It's very kind of you to share all that.

    As you say, I would need to read it two or three times.

    But the first thing that occurs to me is this.

    Why not split the track into separate parts; do an EQ match on each part: and then stick them back together again? Wouldn't that be simpler? As long as it's only the bass that's changing, and you split at a point where the bass is not playing, the "join" wouldn't be any more obvious than in the original recording.
     
    darkmass likes this.
  7. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    I'm looking forward to reading your article. Convolution is an approach I haven't considered but it sounds very interesting.

    With Voxengo Curve EQ, you can already share EQ profiles very easily. You can save any EQ profile as a CSV file, which is very small, and send it to anyone else. The only problem is, I think it would probably only work for someone who also had Voxengo Curve EQ.
     
  8. darkmass

    darkmass Forum Resident

    Of course, I didn't make the discovery that the accentuated bass needed to be managed till after I did an overall EQ match. So that became my starting point.

    Still, I have since thought about what you are suggesting, thinking that could lead to some easier automation. I see couple of problems with separate matching followed by a join.

    First, there is probably some overlap in the spectrum of Joni's voice and the dulcimer (and, the bass guitar is still present, just not boosted in gain). The later section does not contain Joni, but still contains dulcimer (and the boosted bass guitar). If the dulcimer ended up receiving different EQs in the two sections due to other factors making contributions (or not), would there be a noticeable shift in the dulcimer's tonality across the sections?

    Secondly, there is the matter of overall normalization of volume in each of the two sections if they are separately treated. If the EQ matching process works to keep a constant overall volume level during the EQ modifications, individual elements within the EQ spectrum could still be changed in volume level (and the bass boost in the end section of the cut might work to push the volume of the dulcimer downward). Through a little trial and error, the volume levels could be brought into agreement, but that might be a bit tedious and with some degree of uncertainty of the final result--though that depends, of course, on the ear and the tools brought to the task.

    Even if there were no shifts in tonality or volume I'd recommend some degree of overlap and crossfade for cutting and joining, otherwise there could be a noticible "tic" at the join point.

    My originally outlined approach keeps all overall EQ and gain modifications across the entire cut, so to my thinking, at least, there is no probability of the sections ending up with any degree of mismatch.

    But no matter what, somehow there must be a different EQ treatment of the tail end of "All I Want" if a person is trying to make HDtracks sound more like Steve's work. My above points are based strictly on "thought experiment" and may have little practical significance. I encourage you to try whatever makes sense to you.
     
  9. Ulli

    Ulli Forum Resident

    Location:
    Germany
    Yes, that's the problem, one is tied to one particular plugin. With convolution, one becomes completely independent of any program. I like to think of that method as "sampling hard- or software", and my article (which is basically finished, I'm just working on the sound file examples) explains a bit what it is and where it's being used apart from what we're after.
     
    back2vinyl likes this.
  10. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    I agree that the shift in tonality and the shift in amplitude are two potential problems. But... let's imagine for a moment that Steve had his fingers on the faders for this particular track and made a shift in EQ at the point you refer to. If he did, then surely he too would have run into both the problems you describe? So if you or I do the same thing at home, then theoretically, it shouldn't be any worse. With the emphasis on "theoretically"!
     
  11. darkmass

    darkmass Forum Resident

    Your point is a good one. And an awful lot comes down to theoretical differences versus perceivable differences.

    In the last several hours I've decided it would be a personally fun and informative experiment, since I have exactly the tools I need, to split the 192k/24 HDtracks and Steve's DCC of "All I Want" at the identical point precisely before the accentuated base begins (and that will be a singular point with no overlap and cross fading used for reassembly), and use CurveEQ to separately EQ match the respective sections before and after the split point. Then I'll stick the HDtracks sections back together and give things a listen through my Senn 650s.

    This'll be in the next couple of days when I've set aside the time for it. I'll report back, but however things fall we will be learning something I think. Don't know if I'll run a second experiment running Har-Bal instead of CurveEQ, but I might, I just might. I can be strange that way.
     
  12. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    Excellent! I must admit, if Steve EQd it "live" with faders, he would probably have got a smoother transition from one part to another than we would get by cutting it into sections and using Curve EQ. If you can find a way of crossfading, it will surely be for the better.
     
  13. darkmass

    darkmass Forum Resident

    I flat love crossfading...and I can always find a way. I may do that before this is done. However, I also wanted to investigate what would happen with a very carefully chosen single split point with no crossfading. I figured that would verify or not verify my statement that a "tic" might be heard at that point, and I thought if there was a shift in volume that non-faded split could make that more clear. That could be a starting point leading to improvements.

    I've already spent some time on this small project and have some quite interesting, even unexpected, observations. I'll hold off on posting my observations for the moment, there are still some thoughts I want to pursue. But why don't I let this on: I chose my split point very carefully, you could say microscopically, and in listening to a processed and stuck back together file I hear nothing showing there is a transition taking place. :)
     
    back2vinyl likes this.
  14. darkmass

    darkmass Forum Resident

    Time to look at some "practicality"...

    As I've looked further into Steve's DCC of "All I Want" from Joni Mitchell's Blue album, with respect to the respective 192k/24 HDTracks version, and have looked into EQ modifications of the HDTracks version using Voxengo CurveEQ and Har-Bal 3.0, almost every phase has resulted in a complete surprise to me. Every phase has at first left me confused...followed by a struggle to understand what it was I was hearing--and what I was seeing in graphical representations. There is a lot to pass on, so I'll break things into three successive posts. The first post will concern what I've learned about the CurveEQ and Har-Bal equalization tools and add some additional foundation information. The second post will be about the bass in "All I Want", which is what I initially set out to explore. And the third post will be about what I learned about the treble region of "All I Want". But let me give a quick summary of my results up front: Steve's 44.1k/16 DCC of "All I Want" (and certainly of the entire Blue album) is the gold standard, the "silk purse", of digital versions--but contrary to my expectations, the best the HDTracks version can be transformed into, by using the respective DCC version as an equalization target, is not quite up to "pig's ear" standards. There are regions where the HDTracks of "All I Want" just cannot go.

    I'll explain why that is so, but to start I need to lay some foundation. This will not be the order that discoveries came to me, but I hope I can add some clarity.

    First, though both CurveEQ and Har-Bal allow an equalization curve to be copied from a reference to a target, and although both allow that for a "mid" combination of left and right stereo channels (probably the easiest and best approach for transferring EQ), it turns out the two products do not form their "mid" in the same way. By all indications (and I've confirmed this through the use of the "Channel Phasing" track from Stereophile Test CD 2), Har-Bal does a simple summation of the left and right stereo tracks and next performs a frequency analysis of the summation. The key thing to understand here is that if left and right channels are completely in phase with each other, the "mid" will be formed by a simple sum of the left and right channels; however, if the two channels are exactly 180 degrees out of phase throughout the analysis interval, the sum that goes into the "mid" will be zero throughout the interval. As a corollary to this, if some frequencies in the overall analysis interval aren't completely in phase in some region, there will be a partial energy cancellation leading to a reduction of the left and right sum for those frequencies. There is nothing wrong with this approach, it could be thought of as representing what happens in a theoretical free space with those particular frequencies.

    On the other hand, when CurveEQ prepares its "mid" for analysis, it could be said to sum the "absolute value" of the energy of any frequency, out of phase or not. If left and right channels are completely 180 degrees out of phase, CurveEQ will form the same analysis result as if they were completely in phase. As a corollary, if some frequencies are out of phase by any amount, CurveEQ does not see that--they are still accounted for as if they were completely in phase.
    Now none of this "mid" forming methodology is incorrect (indeed, if there are no left/right phase differences, the summed "mid" should be the same with either EQ tool), and none of it has any direct bearing on the bass situation I started out investigating. But the respective summing methods are still important to "All I Want" overall...and it's indirectly important because it probably helped lead to my raising the bass in the last few seconds of the song when I was working with Har-Bal.

    Just for the record, let me state here that when I did all my listening during my initial EQ transference work on Blue I was using asynchronous USB transfer to my Wadia 121 Decoding Computer (call it a DAC), which was feeding a pair of Focal Professional - CMS 65 internally-powered near field monitors--this is what I use as part of my audio/visual workstation. And let me also state that during my analysis over the last few days, I made extensive use of the spectrum analysis feature of Sound Forge Pro 11--using the spectrum analysis feature I can look at both left and right channel spectrums simultaneously, and I can overlay the spectrum from one file on another. I also summed both channels to mono (the summation in Sound Forge Pro works the same as the "mid" summation in Har-Bal) and overlaid spectrums from various files. In all my curve plots I first normalized all files that went into the plots to the same average RMS level. Whenever I used CurveEQ, I used it in the "60 point" option. When using the DCC version of "All I Want" as an equalization target or in comparative spectrum analysis plots, I used a 192k/24 uprezzed version of the DCC file that I prepared from the 44.1k/16 by first using the "iZotope MBIT + Dither" feature of Sound Forge Pro to convert to a 24-bit depth, followed by using the "iZotope 64-Bit SRC" (with "use simplified Quality setting" checked, and with the "Quality" slider set full to the "100.00/Highest quality" position) feature of Sound Forge Pro to uprez to 192k.

    I had first tried CurveEQ with the DCC EQ target at 44.1k/16 and the HDTracks cut at 192k/24, and results were produced that were not very listenable, with somewhat saturated bass. Things worked much better when the DCC target and the HDTracks were both at 192k/24.
     
  15. darkmass

    darkmass Forum Resident

    It's time to talk about the bass. I'll break my examination down into steps:

    1) As you know, I split "All I Want" into two sections--both for Steve's DCC (uprezzed to 192k/24) and for the corresponding HDTracks cut--the second of the sections containing the emphasized bass guitar in the last few seconds of the song.

    2) I then used CurveEQ independently on each of the two sections to map the DCC EQ onto the corresponding HDTracks sections.

    3) I joined the two HDTracks sections back together (using no cross-fading), then set that result aside.

    4) I next ran a CurveEQ transference from an unsplit copy of the 192k/24 uprez of the DCC "All I Want" onto a full copy of the HDTracks version of the cut which had never been spit and processed.

    5) I brought both the "never split" HDTracks result and the split/joined HDTracks result into Sound Forge Pro, normalized the two to identical RMS values, then overlaid the separate left and right channel spectrum analysis from the "never split" version on top of the corresponding split/rejoined spectrum analysis.

    Are you ready for this? I wasn't. There was no difference between the "never split" and the "split/rejoined" versions at any point of the spectrum. And that was true for both the left and right channels. Yet, I had been convinced that I had originally heard something I had to compensate for! However, using CurveEQ there seemed to be absolutely nothing exceptional going on. I was now feeling really bad. The need for separately emphasizing the bass at the end of "all I Want" was something I had talked about at some length, and I'd actually gone through quite a lot of work to "fix"--but there was no actual problem there? Maybe it was time for me to fall on my sword.

    I'll talk about some indirect phase effects on the bass in my next post, but first let me show you a picture that, once I noticed the distinction, left me dumbfounded. I've come to believe the bass in the end region of the "All I Want" left my ears confused. The picture will probably help explain the confusion.

    partB.jpg

    That's the end section of "All I Want". The full portion that got its separate EQ. The top pair is the DCC version (with the topmost of the pair the left channel), the lower pair is the unEQed version from HDTracks. To aid the comparison I set both to the identical normalized RMS value. The big "blobs" are bass guitar...the smaller "spikes" are dulcimer. In the DCC, the bass is not at all present in the left channel! (Yet, while listening both through speakers and through headphones to the HDTracks, there is enough of a difference between channels that the bass seems to be primarily in the right channel, not in the center.) The HDTracks doesn't sound as monophonic as it looks, but for whatever ununderstandable reason, it seems that the source for HDTracks has the left channel mixed into the right. If it's even technically possible, it would not be easy to make the end region of this particular song into Steve's version.

    Looking at overall spectrum plots, both the plot of the overall EQ spectrum of my Har-Bal EQ remapping of "All I Want" and the corresponding plot of the CurveEQ remapping are quite similar up to about 3k Hz. For both plots, "mid" spectrums look identical between about 175 Hz and roughly 3k Hz. Below 175 Hz the two traces closely follow each other, but the Har-Bal is almost one dB higher at about 45 Hz, with the CurveEQ result about one dB higher at 120 Hz. However, in a plot of "peaks" in the spectrum, in that same bass interval the Har-Bal effort has peaks a fairly consistent three dB higher than the bass peaks of the CurveEQ. Remember that after I parametrically bumped up the response at 90 Hz when I was working with Har-Bal, I followed by once again doing an overall match to the DCC EQ. That forced an overall spectrum match of the average response, but still left bass peaks in a region roughly centered at 90 Hz. Those emphasized peaks struck me as a satisfactory bass correction.
     
  16. darkmass

    darkmass Forum Resident

    All right, for the moment that's enough about the bass region. I certainly understand the bass is all I initially set out analyze. However, during the course of things I saw that the CurveEQ and the Har-Bal spectrums were very similar up to about 3k Hz--above that they were quite dissimilar. In my original Blue set of posts I mentioned that the dulcimer strings had an extra "metallic edge" following my Har-Bal EQ mapping. During this investigation of CurveEQ and Har-Bal equalization I learned what gave rise to that "metallic edge" I had noticed. I mentioned the phase distinctions between the way Har-Bal and CurveEQ form their "mid" spectrums for analysis, it's time look at how that relates to the DCC and the HDTracks versions of "All I Want". Actually, what I learned about the phase differences will bring me back to the bass effects I believed I heard, so I'll get back to the bass towards the end of this.

    Here's the situation... In the HDTracks rendition of "All I Want" (and I believe this is probably the case for every HDTracks cut on the Blue album), the left and right channels are not quite in phase with each other in the region between about 3k Hz and 12k Hz--with the peak phase difference occurring between 5k Hz and 10k Hz (spectrum plots show up to 5dB of phase-related cancellation, in the 5k Hz to 10k Hz region, with respect to signals with no phase differences). That's a broad enough region of the spectrum. These phase differences are not present in the DCC rendition of "All I Want". Because of the distinctions between the way CurveEQ and Har-Bal form their respective "mid" spectrums for EQ matching, this has very real effects on any resulting EQ match--particularly when what is being matched has no such phase differences. I thought about posting a set of screen captures of spectrum analysis traces, and if you want, I'll do that, but let me attempt to describe the differences.

    Har-Bal is forcing a source spectrum partially canceled in level (by the way Har-Bal sums phase differences), to match a target spectrum with no such cancellation. This means extra energy will be put into the adjusted spectrum results to compensate for the cancellation effects. Given that this is a somewhat high frequency region, the extra energy leads to an exaggerated "brightness"...my "metallic edge" (observable when phase cancellation cancellation does not occur in "free air" around stereo speakers, and also observable with headphone listening).

    CurveEQ, on the other hand, is not affected by phase differences when it prepares "mid" spectrums for EQ matching. There will be no extra energy put into each channel to compensate for phase cancellation effects. However, that also means that the results of the matching (which will still contain all phase differences) can sound diminished if any phase cancellation does occur in "free air" around stereo speakers. Of course this phase cancellation can also occur when playing unmodified HDTracks files, but that hardly takes away from the error of having the phase differences in the first place.

    Tying this back to my bass "compensation", I had extra "brightness" in the dulcimer playing in my Har-Bal results. The bass in the HDTracks had significant, ear-confusing, problems even without the accentuated brightness--but the exaggerated brightness (specifically in a area where bass was intended to be very notable), encouraged me to lift the bass to make the bass level sound relatively more correct with respect to the dulcimer. Well, that's what I'm going to claim anyway.

    In the course of this investigation I've gained a good deal of respect for the abilities of both Har-Bal and CurveEQ. They both need to be understood, but given that source and target files are well-behaved, I have seen that both can do an excellent job of EQ matching. While I think that Har-Bal can present a lot of useful information in the process (a lot more than I've used so far), CurveEQ can give no less good EQ results a lot more easily and quickly, using its excellent automation.
     
  17. Grant

    Grant Life is a rock, but the radio rolled me!

    I can do the same thing with iZotope Ozone.

    I have only used it once or twice, and thought I would find lots of uses for it, but I haven't. Over the years, I have developed a pretty good ear for EQ'ing blind.

    I've only skimmed a few posts so far, but, from what I can tell, a few of those people don't seem to understand exactly what this EQ matching does. All it does is allow one to copy the EQ of a source file and apply it to another so that they both sound similar. No one is talking about trying to make a CD sound like a record, or vice-versa. Say I have completed a needledrop of a 60s 45, but like how the EQ sounds on a CD comp that it is on. I can simply tell Ozone to learn the frequency curve of that CD track, and apply it to my needledrop. A tweak here and there after it's done, and there it is!
     
    Last edited: Mar 28, 2014
    back2vinyl likes this.
  18. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    That's awesome, darkmass - I think your experiences will be of great benefit to people experimenting with these techniques and it's very generous of you to have written them up and shared them at such length.

    A couple of initial thoughts:

    It's kinda obvious but perhaps worth mentioning - I think we should always run a comparison between the finished (EQd) product and the one we set out to copy, to see whether the EQ adjustment has indeed produced the desired result. Often the match is virtually perfect but sometimes there are discrepancies that need looking into. In the cases you're talking about, can you confirm that the finished product appeared to match the target (a) according to CurveEQ and (b) according to Har-Bal?

    It would be comforting to think that if the finished product is a perfect EQ match with the target, they must inevitably sound the same. But in fact, as your experiments have (I think) shown, it's not necessarily so. I've just had a similar experience - I was transferring the EQ of the David Bowie LP Hunky Dory to a rip taken from the EMI CD. The EQ match was good and so was the overall tonality but the track "Andy Warhol", for example, seemed to have a more aggressive edge to the guitar strumming on the EQd version than on the LP. On looking into this, I realised that the CD actually had more dynamic range than the LP and the transients on the CD version were still greater in amplitude on the CD version than on the LP version even after the EQ matching.

    So, there can be problems with using different sources. In the case of Blue, it seems you almost had a different mix, to the extent that the bass was differently presented, at ;east on that one track. Hopefully these sorts of problems are not too common and as long as we're aware of the pitfalls, we can avoid them.
     
  19. darkmass

    darkmass Forum Resident

    "
    Your "Andy Warhol" experience is quite similar to what I experienced when working with the Blues for Thought album. The equalization match was very good and coupled with the better dynamics of the DAD version, I had the sense that all the musicians were playing with "more muscle". Drum hits, guitar, everything, made it feel like they were performing in a small "chicken shack" type joint for a crowd they knew appreciated them. Though it wasn't a live recording, it had a "live" feel to it.

    As far as the "All I Want" finished product EQ spectrum appearing to match the target... Har-Bal has manual matching, and I used it in an iterative manner. I'd use the results from the match on a given cut, bring that back into Har-Bal, then if it still seemed different from the desired target I'd adjust those differences out and write out the fresh results. If it took a few times till I was satisfied, well, no problem. (The tracks on Blue seemed to take a few extra iterations, the tracks on Blues for Thought fell into place quickly--I now suspect because the left/right tracks on Blue had more than differences than EQ could account for.) As concerns the current work on "All I Want", using the Har-Bal result I had as of the time of your post, and comparing that to the EQ target, I show a perfect match from 18 Hz through to 22k Hz.

    My real working experience with CurveEQ has only come about during the current investigation of "All I Want". There's certainly the possibility I'll be making better use of it in the future. But, again, using the files I had at the time of your post, within CurveEQ I'm showing a pretty good match between my CurveEQ result and the target for "All I Want". The specifics are, a match within 0.7 dB from 20 Hz through to 20k Hz, with a match within 0.2 dB from 81.5 Hz through to 20k Hz. CurveEQ's developer has stated that he has purposely limited the matching below 200 Hz, but even so the match I got is not too shabby. I suspect the match would be even better if I had RMS normalized the result and target files to a common value before bringing them back into CurveEQ for the match check. (CurveEQ no doubt provides for internal gain adjustments for the match, but I'm still learning.)
     
  20. darkmass

    darkmass Forum Resident

    Now I've kind of fallen down the rabbit hole since the time of my last post in this thread as a result of thinking about ways I could better match the renegade HDTracks of Blue to the DCC rendition--though still limiting myself to working with "All I Want". The phase differences between the left and right channels were something that began to further ramp up my obsessive/compulsive nature. I started thinking along the lines of, "what if one side of the stereo pair was simply 'slipped' with respect to the other side"? That brought me to something I'd seen mentioned in the past, but had never really thought about--tape head azimuth alignment. If the tape head for playback was skewed differently than the recording head, that could phase shift some higher frequencies while leaving lower frequencies relatively untouched. (I previously reported phase errors up to 12k Hz, but I suspect the general fall off in the EQ curves above about 10k Hz tended to obscure phase effects occurring at higher frequencies than 12k Hz.) Anyway, I felt that azimuth effects were worth some investigation.

    I have a software tool that's probably rarely owned among audio "civilians"--iZotope RX3 Advanced. I myself would never invest the current $1199 USD in the software, but when RX3 was introduced late last year there was introductory pricing for a short while. On top of that introductory price break, I already owned the "standard" RX2, so I was able to take advantage of an "upgrade" price break. Those two breaks taken together brought my price for RX3 Advanced down to $499 USD--still pretty stiff, but I felt it was a workable price within the context of the workstation I was trying to build up. The point of this is that RX3 Advanced (and not the standard RX3) contains the ability to measure tape head azimuth errors...and correct those errors. That had to be worth a shot.

    Running the full HDTracks 196k rendition of "All I Want" through RX3's azimuth feature, it reported that the right channel was "+10.6 samples" (with respect to the left channel)--that's 55.2 microseconds. Converting 55.2 microseconds to Hz, that's one full cycle at 18115.94 Hz, or perfect half-cycle cancellation at, call it, 9058 Hz. I corrected that out with RX3, then reran EQ matching and looked at spectrum analysis results in Sound Forge Pro. Now that showed something really interesting, and something I almost cannot believe. Obscured by the HDTracks azimuth problems, there was evidence that there were also azimuth problems with the uprezzed to 192k DCC rendition of that same file! I think it can be stated without doubt that Steve would never let something like that occur, but I was still seeing what looked like the effects of a tape recording head / playback head azimuth misalignment. Bringing the uprezzed DCC version of the file into RX3 and running the azimuth feature, I got a reading for the right channel of "+5.0 samples"--that's 26 microseconds. Converting 26 microseconds to Hz, that's one full cycle at 38461.54 Hz, or perfect half-cycle cancellation at, call it, 19231 Hz. Okay, I'd say that's really on the fringes of things, and quite likely within the calibration tolerances of a mechanical screw adjustment (not that I at all know anything about how that adjustment is carried out in a high-quality professional environment).

    I think I should make one thing clear. When I say "perfect half-cycle cancellation" that is really a convenient mathematical construct of what graphically occurs when two sine waves exactly 180 degrees out of phase are added together. In the world of left/right stereo audio, that is not at all what happens. In the areas of the audio spectrum where proper left/right signals result in an instrument sounding like the instrument is precisely located in space, left/right signals exactly 180 degrees out of phase ("perfect half-cycle cancellation") result in a "diffuse" instrument location that the ears cannot precisely locate. What is lost is clear spatialization. While it is possible some listening locations could result in experiencing a decrease in audio level for the "cancelled" signal, that is certainly not the case for all listening locations. And when that "cancellation" frequency is 19231 Hz, I suspect few, few people would notice any difference from a signal that was completely in left/right phase at that frequency. At such a frequency the "cancellation" is a technical point--there is absolutely nothing amiss with the DCC rendition of "All I Want".

    Now with a bit of simple arithmetic and RX3, I could set the HDTracks azimuth mismatch to agree with the value of the DCC azimuth mismatch. (RX3 determines a "suggested" correction, which is what I've reported above, but in applying the correction to a file RX3 Advanced allows the user to override the suggested value with a desired value.) To match what Steve heard while mastering, I should probably align the HDTracks azimuth value to the DCC azimuth value--that should make both the Har-Bal and CurveEQ matching tools come up no spectrum analysis visible phase errors--but I felt an urge to zero-out all azimuth errors just to see how that worked and sounded. I could always form a fresh match to original DCC values at any time I wished.

    Running EQ matching through Har-Bal and Curve EQ on azimuth corrected HDTracks and DCC files, for the first time I got excellent matches using both equalization tools. For the first time, looking at both left/right stereo and mono spectrum analysis views in Sound Forge Pro, all plots were in very nice agreement. (Given, however, that the left channel of the HDTracks sourced files still showed some bass spectrum matching problems in the left/right stereo view spectrum analysis due to some of the right channel being folded into the left channel). Oh, and the 192k HDTracks file--with azimuth effects zeroed out followed by EQ matching with the DCC source--sounds very good. It may not exactly be Steve's take, but the HDTracks of "All I Want" has become something that sounds very much exactly like it should sound.


    In the mono views, within Sound Forge Pro's spectrum analysis, which correspond to the way the EQ matching tools operate, I saw the following...

    For CurveEQ results overlayed on the DCC target, there was an exact match from 127 Hz on up, but also a good match in the region of 77 Hz to 93 Hz (a graph crossover occurred in the center of that region)--at 26 Hz, the DCC plot was 2 dB higher than the CurveEQ plot.

    For Har-Bal results overlayed on the DCC target, there was an exact match from 43 Hz on up--at 24 Hz, the DCC plot was 1.3 dB higher than the Har-Bal plot.
     
  21. darkmass

    darkmass Forum Resident

    While "correcting" tape head azimuth misalignment seems to me to be a good thing to look at for any recording that once spent time on an analog tape recorder, tools which can provide insight or correction are almost nonexistent, except maybe at great expense. There may be more out there that I never discovered, but I did have a look to see if there was anything I could rationally suggest. I do not consider RX3 Advanced a rational suggestion (though if a person has the pockets...).

    There is something out there called "Stereo Tool" which some people like a lot. It's free, and it does tape head azimuth correction--though it requires a small payment to unlock VST plug-in operation. But if high resolution audio is a consideration, I could never recommend Stereo Tool. While it can take any resolution for an input, and output that same resolution, for its internal processing it first downrezzes the input to 44.1k or 48k, whichever evenly divides the input resolution. After processing at the reduced resolution, it uprezzes its result back to the input resolution. That clearly destroys all high resolution qualities. The downrezzing and uprezzing methods could be further called into question, but since it is working with even divisions and even multiplications, the up/down rezzing is probably reasonably benign. I've seen the tool's developer state that it was impossible to change the rezzing nature of the way Stereo Tool operates. Still, if a person is limiting themselves to working strictly with 44.1k or 48k, Stereo Tool may be a fine way to go. (I've seen no statements concerning bit depth.)

    But the author of Stereo Tool has suggested another program of his, "Tape Restore Live!", as an alternative for high resolution work. It too can work with tape head azimuth...and it's free. The downside is that Tape Restore Live! works only as a Winamp plugin. While Winamp is intended for stream output, I've seen that there are ways Winamp can be made to output wave files. So that could be worth investigating. Another downside is that as of late 2013, Winamp is no longer a product. It's probably still possible for a person to put their hands on a copy, however.

    Voxengo makes "Voxengo Sound Delay", a plug-in that's free, which its author reports can precisely delay one track of a stereo pair--though I suspect determining what a proper delay should be could be a problem. It may be that using Tape Restore Live! (or even Stereo Tool) as a tool to only determine what a proper azimuth correction would be, then using the "Voxengo Sound Delay plug-in to carry out the actual correction might be a feasible way to go.

    Now all this "azimuth" stuff is very interesting, at least to me, but in practice it has no equalization impact if a person is solely using CurveEQ. That is due to the phase independent way CurveEQ forms its "mid" curves for analysis and for equalization matching. If a person brings the results and target files into a secondary program for spectrum analysis, some things could be seen that are not really understandable. But in the case where the target is from a needledrop with a corresponding CD providing the file which will have its spectrum forced to match the needledrop target, no phase effects will be in the results file that were not present in the CD to begin with. A person using Har-Bal would have a more serious situation on their hands because it forms its mid curve in a more "standard" manner; however, a CurveEQ user gets away pretty clean.

    But. Many members of the audiophile community take some pride in systems and recordings that provide the means to hear all elements of the recording precisely positioned in space. Tape recording/playback head azimuth inconsistencies work against that precise positioning, due to induced phase differences between the left and right channels. Actually, that is true whether needledropping is concerned or not. However, if a person is inclined to delve into things, get their audio hands dirty, then there is the possibility that cheap and reasonably easy azimuth investigation and correction could actually bring some sound field positioning rewards.
     
  22. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    That's very impressive work you've done on phase differences. I think I follow it all but as you say it is quite a rabbit hole and it's one I'm not planning to follow you down just at the moment! That's for two reasons. First, from what you've said, it's not an issue as long as you're using Voxengo CurveEQ and since that's what I'm using, its interesting to me more in theory than in practice. Second, there's a limit to how far I'm prepared to go in pursuit of a quality needledrop - there has to be a limit on how much time you spend on each one or there'd be no time left to enjoy the music!

    It's interesting, though, to explore the limits of the possible and it's good to see you pushing the boundaries.

    Picking up on some of your comments about CurveEQ, you say the developer purposely limited the matching below 200Hz. I wonder why he did that? I have noticed that things can get a little erratic at the bottom end if you're comparing two sources that are wildly different. LPs for example always have a big hump at the bottom end from all the rumble and that can cause problems, though these are easily dealt with by manually adjusting the curve. Another issue is that if source A has a lot of bass and source B has very little, the bass doesn't seem to transfer well from A to B, probably because CurveEQ struggles to increase what's barely there to begin with. Again, you can manually compensate.

    Most of the time CurveEQ seems to do amazingly well but sometimes there are small discrepancies in the EQ curve between the finished product and the target. I think these may be caused by differences in dynamics or maybe by the sort of channel imbalances you've talked about - I haven't quite worked it out yet. But it's quite unusual in my experience so far for these differences to be big enough to be noticeable. Incidentally, you wondered whether you would get a better EQ curve match by level matching the two sources but CurveEQ isn't affected by gain so this shouldn't make any difference, in theory.
     
  23. back2vinyl

    back2vinyl Forum Resident Thread Starter

    Location:
    London, UK
    I don't know if you're still there, darkmass, but I've done a Voxengo needledrop on Joni Mitchell's Blue and wondered if you might be interested in the result. I took the standard issue HDCD, ripped it, and used dbPoweramp to decode it (though in fact I don't think there was any peak extension there in the first place). I then ripped Steve's well-liked vinyl mastering for Rhino from a few years back. Taking one track at a time, I then used Voxengo CurveEQ to apply the EQ of Steve's LP mastering to the HDCD.

    I haven't had time to give it a good listening yet but so far I'm very happy with the result. You talked about a discrepancy in the final section of the first track, "All I Want", where you said the bass guitar in your experiment wasn't as loud as it was on Steve's mastering. So far, I think my copy sounds OK, perhaps because I'm using the HDCD instead of the HDtracks download and because I'm using Voxengo instead of Har-Bal. But you be the judge! Here below, I've pasted links to the last minute of Track 1 of my finished product and also to the last minute of Track 1 of the LP - I'd be interested to know if you hear a difference:

    https://drive.google.com/file/d/0B4kRLGUeuQb-MHlSclNVVUh1Tzg/edit?usp=sharing

    https://drive.google.com/file/d/0B4kRLGUeuQb-OGFvSFBMTzNLUUU/edit?usp=sharing

    I deliberately haven't said which sample is which yet so you have a chance to do a sort of blind test, though you can probably work it out. I've chopped the end rather abruptly so you don't get any clues from the LP surface noise when the music fades!

    A couple of notes:

    1. I think the bass guitar is in fact the bottom string of a normal six-string acoustic guitar? But I'm not sure.

    2. My cartridge, an AT33PTG/II, is an MC type. The frequency response slopes very gently downwards from left to right nearly all the way through the frequency range, so the sound may be slightly more rolled-off than people with brighter cartridges are used to.

    3. The HDCD before EQing was very bright, far brighter than the LP.
     
    Last edited: Apr 15, 2014
  24. darkmass

    darkmass Forum Resident

    I'm still here. I've been fooling around with additional semi-meaningless stuff and have been pondering an additional post to this thread. Your latest post has brought me back to a form of reality. :)

    I'd believe you are correct about what I've been referring to as a "bass guitar", in fact I've been having similar thoughts for a while now. Still, "bass guitar" was a convenient shorthand for me.

    I've downloaded your files and look forward to hearing them. I'm going to wait a bit for the listen, however, because I want to prepare a similar segment from "All I Want" for you using my current best practices. I suspect I'll also pass on the similar segment from the well known DCC version.

    Stay tuned...
     
  25. darkmass

    darkmass Forum Resident

    Okay, let's see if I've been able to get this google drive stuff working...

    If so, in no particular order, there's the end segment from the DCC CD of "All I Want" and the equivalent segment originated from the HDTracks 192k/24 of the same selection. Though I uprezzed the DCC to 192k/24 for the EQ matching and other related work, the DCC segment here never made that journey. On the other hand, after I finished everything, I downrezzed the HDTracks result to 44.1k/16, then trimmed off the section here for your listening. Since I also want to make a regular CD of my results, I'm considering the downrez to CD specifications part of the process.

    I made no use of Har-Bal for this completion of the work, the EQ was done via CurveEQ...by a process that could be described as a bit less than straightforward. You are welcome to blind listen, but I'm hardly mandating that. However you listen, see what you think.

    https://drive.google.com/file/d/0ByoJu9I97JBmczVLMTZfSDNvSjA/edit?usp=sharing

    https://drive.google.com/file/d/0ByoJu9I97JBmX2k4NWtvQ1M1OXc/edit?usp=sharing

    I've listened to your examples, but I'll pass my observations on after you've had a chance to listen to my links.
     
    Last edited: Apr 16, 2014
Thread Status:
Not open for further replies.

Share This Page

molar-endocrine