The Audio Preservation Thread — Page 7

I agree with PDB: for those of us not savvy enough or who don't have the appropriate software, a 48/16 bit sample rate is preferable for BD format compatibility.

I keep 44.1 versions as well, but I doubt if there's any noticeable quality difference when using the right software and settings to resample, I always use the highest quality setting in Sound Forge (without applying an anti-aliasing filter since I assume it's not needed when going to a higher rate).

PDB said:

After along time, finally finished something:

Movie: Batman

Format: Laserdisc 12000 A/B

Input Soundtrack: PCM 2.0 Dolby Stereo Surround 44.1 khz, 16-bit, bit perfect 

Output Soundtrack: PCM 2.0 Dolby Stereo Surround 48 khz, 16-bit

Synced To: 2010 Blu-ray Release Region A 

Ripped/Synced by: PDB 

Notes: Contains the original Dolby Stereo soundtrack.

:O that's awesome!

borisanddoris said:

Just downloaded Batman LPCM. Is it me or is something really off with the audio?  It's like some odd compression artifacts. Noticed it in VLC, Audacity, as well as into my Pioneer processor via the Oppo 93. 

Really noticeable on the cymbal crash after Tim Burton's director credit. 

That's my fault. I was rushing to upload a new copy and exported the entire session instead of just the LD track. That's what happens when you rush. Thanks to Jonno for pointing that out. I already fixed it, tested it and uploading it now. I will send you a new link boris.

Just curious, what is the difference between "bit perfect" and "not bit perfect"?

jbaker111 said:

Just curious, what is the difference between "bit perfect" and "not bit perfect"?

 Read here:

http://originaltrilogy.com/forum/topic.cfm/Please-explain-Bit-Perfect-Audio-Capture-for-LaserDisc-PCM/topic/15749/

On the subject of Batman, how well is the audio preserved on DVD/BD for Batman Returns? Given that it was first to use Dolby Digital. 

NeonBible said:

On the subject of Batman, how well is the audio preserved on DVD/BD for Batman Returns? Given that it was first to use Dolby Digital. 

 I was thinking of doing Batman Returns LD's PCM Dolby Surround (Batman already done and all) and Jurassic Park LD's PCM Dolby Surround. Kind of a last gasp of Dolby Stereo set since both were the first movies to have digital theatrical soundtracks (BR was Dolby Digital/AC-3/SR-Dand JP was DTS) but both were still heard by the majority of people in Dolby Stereo.

That reminds me of when I first saw Jurassic Park. It was in DTS and boy even as a kid I could tell the difference. When that disc and the digital experience sign exploded, people in the theater got up and cheered.

I also am working on the VHS soundtracks of Frankenstein and Dracula now that I have a copy of the 1999 DVDs to do also. That way you will be able to compare the changes from VHS to the first (1999) DVDS to the BD.

If you're not using iZotope or Weiss Saracon for sample rate conversion, you really should be.

All SRC processes unavoidably degrade the audio quality of the original source.  SRC algorithms require very steep lowpass filters to reject unwanted high frequency energy at greater than half the sample rate, which is introduced into the signal regardless of whether the target rate is higher or lower than it started out.  This level of steepness is necessary in order to retain the full signal within the audible bandwidth while falling to a sufficiently low level at the Nyquist frequency to prevent aliasing, meaning that signals greater than half the sample rate are folded back down into the audible range, creating inharmonic distortion.

Unfortunately, the steeper the filter, the greater the amount of ringing (time-smearing) that is introduced into the processed signal.  This ringing can be either pre or post: that is, it can manifest either forwards or backwards in time relative to its original position.  Linear phase filters are usually used in SRC in order to preserve the phase relationships of the signal, and this type of filter generates both pre- and post-ringing in equal measure.  Such ringing obscures the clarity of transients (high-level, short duration peaks) and gives a general sense of blurriness in the sound.

From these inherent problems with SRC (and it should be noted that the quality of analog-to-digital and digital-to-analog converters is affected by these exact same issues), we can see that there are three parameters the algorithm designer must contend with to obtain the best possible sound after conversion: filter steepness, which determines both the amount of aliasing that is allowed through and the amount of ringing that is introduced; filter phase, which determines the time relationship of the ringing to the original signal; and cutoff point, which is the frequency at which the filter begins rolling off the audio.  Higher steepness allows a greater frequency range to be represented and rejects most aliasing, at the expense of the ringing becoming far more severe.  Lower steepness avoids problematic ringing artifacts, but at the expense of greater aliasing.  Lowering the cutoff point can alleviate both issues, but restricts the bandwidth of the resulting audio by cutting off some of the high frequency range.

Optimal quality can only come from achieving a balance between these parameters, and there aren't many conversion algorithms out there that can be said to preserve the sound quality of the source signal while minimizing distortion.  That's really the best you can hope for with SRC—it's never going to be a perfect result, but a well-designed filter can keep the distortion products small enough that they won't affect the quality of what you hear.  The same is also true of ADCs and DACs, as well.  This is one of the caveats of digital audio: it's only as good as the math that was programmed into it, and outside the usable range of those equations it will fall apart completely, generating extremely nasty inharmonic distortions that analog audio never had to contend with.  So if you're going to do digital processing on your sound, you'd better be sure that the programmers knew what they were doing before entrusting your work to their knowledge of mathematics.

I should also point out that SRC should be performed at the highest possible bit depth your system can handle, in order to avoid quantization error (another form of digital distortion, resulting from the bit depth being too low to represent the signal with complete accuracy).  Afterwards, you can dither it back down to your target format.  The noise floor will be increased slightly by doing this, but the sound quality will be far better preserved.

My recommendation for sample rate conversion is to use iZotope's solution, which is the most affordable high quality converter out there.  Some implementations of it allow the user to control the parameters, so you ought to understand what they mean before changing them—hopefully the above explanation is helpful for this.  The settings I personally use are: Filter Steepness 32; Cutoff 0.95; Pre-Ringing 0.99.  To me, this represents an ideal result, with a relatively shallow filter in order to minimize ringing, a frequency response up to about 19.8 khz before filter rolloff begins, and the filter being nearly linear phase but not quite, in order to push the ringing slightly later in time (since post-ringing is less noticeable than pre-ringing).  A small amount of aliasing is allowed through, but it is at a very low level compared to the actual signal and is entirely confined to the range above 20 khz, where it is inaudible to anyone who isn't a dog or a liar.  ;)

I hope this clears up some confusion about SRC.  It is a complicated subject, and there's a lot of misinformation about it out there, but I'll be glad to answer any further questions.  If you follow the advice I've given, your converted audio will be difficult to distinguish from the source, which is the best that it can be.

This website has comparisons of many SRC algorithms and is a great resource for seeing which ones are usable and which are not: http://src.infinitewave.ca

I see that Sound Forge has elements of iZotope RX, Ozone, and Nectar built into it, which is awesome.  I've recently started using these myself as Pro Tools plugins, and they really do sound great.  You can get some very good results from them if you know what you're doing, so if this is what's being used for laserdisc audio preservations, so much the better.

PDB said:

That's my fault. I was rushing to upload a new copy and exported the entire session instead of just the LD track. That's what happens when you rush. Thanks to Jonno for pointing that out. I already fixed it, tested it and uploading it now. I will send you a new link boris.

 I've just muxed the new version and checked out a few scenes - it sounds terrific so far. Thanks for putting in the hard work!

hairy_hen said:

All SRC processes unavoidably degrade the audio quality of the original source.  SRC algorithms require very steep lowpass filters to reject unwanted high frequency energy at greater than half the sample rate, which is introduced into the signal regardless of whether the target rate is higher or lower than it started out. 

I see that Sound Forge has elements of iZotope RX, Ozone, and Nectar built into it, which is awesome.  I've recently started using these myself as Pro Tools plugins, and they really do sound great.  You can get some very good results from them if you know what you're doing, so if this is what's being used for laserdisc audio preservations, so much the better.

Just to be clear, is this in reference to Sound Forge with anti-alias filtering, or without?  I've been using Sound Forge without anti-aliasing since that's what I was recommended a long time ago on another forum (and so far I haven't noticed any audible difference with the original 44.1kHz LD audio) , but I can re-do the resampling with another method if it will produce better results.

As far as I understand it, from reading posts by Dan Lavry (a designer of high-end hardware converters) and iZotope's Alexey Lukin, anti-aliasing should always be used, for any kind of conversion between sample rates.

The need for anti-aliasing is easier to comprehend for downsampling, since higher sample rates may contain ultrasonic frequencies that get pushed back down into lower regions if not first filtered out when going to a lower sample rate.  At 44.1 khz, the highest possible frequency that can ever be represented (commonly referred to as the Nyquist frequency) is half the sample rate, or 22.05 khz.  If you try to record anything higher than this, the DAC cannot distinguish between the actual signal and a lower frequency version that would fit into the same time relationship, and mathematically this works out to the difference between the ultrasonic frequency and Nyquist being 'mirrored' back down in the opposite direction.  For example, if you tried to record a 32.05 khz signal at a sample rate of 44.1, it would create an alias at 12.05 khz instead.  This aliased signal has no harmonic relationship to the original recording, and manifests as an unlistenable garbage tone.  Aliasing was quite a significant problem in the early days of digital audio, when more primitive converter design prevented filters from being as effective at rejecting ultrasonics as they needed to be, so it is likely that most digital recordings or transfers from those days contain a hashy, distorted top end to some degree.  (This may, indeed, include some of the early laserdiscs that are being preserved here).

We can see, then, that the need for anti-aliasing when downsampling is quite clear.  What is less clear, however, is why it is necessary when the sample rate is increased.  I don't claim to fully understand it myself, but apparently increasing the sample rate creates 'ghost images' of the original recording up in the ultrasonic range, and these too must be filtered out to maintain optimal quality.  They aren't as problematic as going the other way, since the higher the target sample rate, the shallower the filter needed to keep these ghost products inaudibly low, meaning the ringing introduced by the filter will therefore be less severe.  But they should still be filtered out rather than ignored—so in answer to the question posed above, I would say yes, anti-aliasing should always be enabled for SRC.  It's just a matter of determining the appropriate filter slope to correspond to which rates are being used.

The ability to use shallower anti-aliasing filters is actually the true benefit of using higher sample rates.  Contrary to what some would have you believe, there is really no mystical voodoo sound benefit that comes from recording ultrasonics, since not only can we not hear them, most microphones can't even pick them up, and most speakers and amplifiers will distort if forced to try to reproduce them.  The sound quality within our audible spectrum of hearing is no less good at 44.1 khz than it is at 192—good quality converters will capture all available detail within that range regardless of what rate is used.  Higher rates do not equate to greater time resolution at lower frequencies; in fact they are actually more prone to timing errors since the digital clock has to work so much harder to deliver each sample.  Those who claim to hear sound quality benefits to higher sample rates may indeed be hearing a difference, but the cause of this is simply that there is less ringing and/or aliasing due to the filters not needing to be as steep.

From all this it is evident that the quality of anti-alias filtering is one of the most important considerations in all of digital audio, and that the best sound can only be achieved when it is properly implemented.  Designing good quality filters at lower sample rates is quite a bit more difficult than at higher ones, which is why choosing the right converters (for recording, playback, and SRC purposes) is such a big deal.  It affects the clarity of everything you hear throughout the entire signal chain; and of course the better your analog section, the more noticeable any problems on the digital side will become.

So to sum all that up in one sentence: always use tools with good quality anti-aliasing enabled, and you'll be far ahead of the curve when it comes to delivering high fidelity results.

hairy_hen said:

So to sum all that up in one sentence: always use tools with good quality anti-aliasing enabled, and you'll be far ahead of the curve when it comes to delivering high fidelity results.

Great, thanks for the detailed reply.

Will there be any degradation from applying the anti-aliasing filter twice (once for upsampling 44.1kHz to 192kHz and once for downsampling down to 48kHz)?  Sound Forge allows one to set the interpolation accuracy from 1 to 4, and setting it to 4 takes a good while for the resampling processing to complete. So might it already be internally doing something similar to the 44.1>192>48 process?  I'd have a bit more peace of mind if I only had to apply the filter once and could go directly from 44.1 to 48, but if 44.1 to 192 to 48 will still be better despite applying the filter twice, then that's what I'll do.

Movie: Blade Runner - International Theatrical / Criterion Cut

Format: Laserdisc NJL-20008 (same master as Criterion LD)

Input Soundtrack: PCM 2.0 Dolby Stereo Surround 44.1 khz, 16-bit, bit perfect 

Output Soundtrack: PCM 2.0 Dolby Stereo Surround 44.1 khz, 16-bit

Synced To:  Blu-Ray - International theatrical version

Ripped/Synced by: Buster D/NeonBible 

Notes: Contains the original Dolby Stereo soundtrack.

NeonBible said: Output Soundtrack: PCM 2.0 Dolby Stereo Surround 44.1 khz, 16-bit

 If the figure given for the output sampling rate is accurate, I would appreciate a copy.

Well that's a hell of a read, hairy_hen. I appreciate a good technical read and clearly, you have much more knowledge on audio theory and mixing then I do. I just downloaded the Weiss Saracon Resampler to play around with. For my next project I might try to use it for resampling before editing.

I use Adobe Audition for all my projects since I have the Adobe suite. What I do know is that when I set the project's sampling to 48,it converts the captured laserdisc's PCM from 44.1 to 48 while also converting depth from 16-bit to 32-bit. I had previously read up as to the reasoning behind changing the bit depth for editing and I saw what you wrote about it in your second post. Fascinating stuff. Now thanks to your post I know bit more on sampling part too. I am unsure as to what processes Audition uses to preform said sample rate conversion. In audition even if I keep the project to 44.1 to match the LD's audio, it would still convert the bit depth to 32 for editing. So even at its original 44.1, the audio is modified on some level (in Audition).

I have to say, I have never noticed a difference between the original 44.1 and the edited 48 that I've done. I've done a couple of A/Bs myself and failed to notice a difference both on headphones and speakers. The new, edited 48 always seems transparent to the original 44.1. 

So I guess we get back to the heart of the matter, is it worth all the trouble to convert to 48? Does changing it to 48 affect the quality? Kind of the argument that Chewtobacca was making the last page. And is the audio inherently different when transformed to 48, similar to the discussion _,,,^..^,,,_ and I had about the DTS Cinema CD-ROMs. Keeping 44.1 would be "truer" to the original rip. Like I said, I can use a HTPC and play the Blu-ray with 44.1 but if anyone else wants to play it in a blu-ray player, they are out of luck. I'm going to try your recommendations, hairy-hen. See what Weis Saracon offers. Thanks for the insightful post.

borisanddoris said:

What a mix everyone. Thanks to all those involved for making this available. How fitting for its 25th Bativersary!

Jonno said:

PDB said:

That's my fault. I was rushing to upload a new copy and exported the entire session instead of just the LD track. That's what happens when you rush. Thanks to Jonno for pointing that out. I already fixed it, tested it and uploading it now. I will send you a new link boris.

 I've just muxed the new version and checked out a few scenes - it sounds terrific so far. Thanks for putting in the hard work!

Thanks guys!

borisanddoris said:

This makes me wonder, what did we use to convert that original Jurassic Park project from 44.1 to 48.  It sounds perfect to my ears and on the handful of systems I've tested it on, including an actual cinema through a CP-650.

 That sounds pretty cool. How did you have the opportunity to use a CP-650? How did it sound?

NeonBible said:

Movie: Blade Runner - International Theatrical / Criterion Cut

Format: Laserdisc NJL-20008 (same master as Criterion LD)

Input Soundtrack: PCM 2.0 Dolby Stereo Surround 44.1 khz, 16-bit, bit perfect 

Output Soundtrack: PCM 2.0 Dolby Stereo Surround 44.1 khz, 16-bit

Synced To:  Blu-Ray - International theatrical version

Ripped/Synced by: Buster D/NeonBible 

Notes: Contains the original Dolby Stereo soundtrack. Release format - FLAC.

 That's great Neonbible! I love Blade Runner. Hopefully, someone can follow your work and get both the theatrical and '92 director's cut done. Thanks to you and Buster D for that track.

PDB said:

So I guess we get back to the heart of the matter, is it worth all the trouble to convert to 48? Does changing it to 48 affect the quality?

It inevitably affects quality.  The high-quality resamplers minimise the damage.

Converting to 48 is unavoidable if one must have a disc playable in a stand-alone player, which is an understandable requirement; however, if it's done after editing, two tracks can be released: one at 44.1 and one at 48.  This way, nobody loses: those who wish to remux to MKV have the option of keeping the original sampling rate; those who want a resampled version can have it; the only change in the editor's workflow is that he resamples the guide track, rather than the track that he wishes to sync.

Again, I'm sincerely sorry if it looks like I'm pushing to have things my way, but it doesn't make sense to me to insist (justifiably) on bit-perfect rips and then to resample the result, especially in a less-than-ideal way.  It's a bit like insisting on buying paper of the highest quality and then giving it to a two-year-old to doodle on it (though obviously that analogy's an extreme one).