Reservoir Dogs (1992): Tarantino's 35mm Print Reconstruction

[jts19840]

Hey Aluminum Falcon,

I know I'm pretty late to this party but I just wanted to say that I have seen Tarantino's Print of Reservoir Dogs a few years ago at the New Beverly and from the screen captures It look like your regrade is spot on.
I was pretty stunned when I saw RD in the theater and the movie looked that dark and rich but can totally report that this is indeed accurate (and better).

I would love to check out this project in motion and am still a new member so I'm not quite how that works.
Any info that you could PM me would be much appreciated.

best,
j

[pipefan413]

I've also uploaded a little side-project... Reservoir Dogs (1992): Open Matte Sundance Premiere Reconstruction!

I love the look of both of these and have had them downloaded for ages. However, I had never actually watched the upscaled version until now, and it turns out the audio is way out of sync. Is this true of the copy others have as well, or is it an issue that has been fixed somewhere along the line? The audio is roughly 2 full seconds ahead of the video.

This is not true of the "Tarantino's 35 mm print" version, which looks bloody marvellous by the way. I haven't had the pleasure of 35 mm Reservoir Dogs but I did get to see a super beat up 35 mm print of Pulp Fiction and it was a great experience.

EDIT: OK I lined up the two projects in Audacity and AviSynth and confirmed what I suspected: the audio is basically in sync between the two projects but the video isn't, so it might be that it was assumed that the two videos were in sync but they aren't. In Audacity, the audio appears to be only very slightly different, at The "Tarantino's 35 mm" project has 48 fewer video frames at the start than the "Open Matte" version, which makes sense given my earlier eyeballed estimate of a 2-second sync discrepancy given that in terms of duration 48 frames is...

48 / (24000/1001) = 2.002 seconds

Which in samples is

48 / (24000/1001) * 48000 = 96,096 samples

To fix it, I've therefore just trimmed 96,096 samples from the audio by applying a "delay" of -2002 ms.


Quick and dirty comparison script:

Code:

tinoprnt=FFMPEGSource2("Reservoir_Dogs_Downscaled.mov").showframenumber()
opendvd=BlankClip(48,640,360,"YV12",24000,1001,0) + FFMPEGSource2("Reservoir_Dogs_OpenMatte_Downscaled.mov").showframenumber()
StackHorizontal(tinoprnt,opendvd)

[allldu]

I've also uploaded a little side-project... Reservoir Dogs (1992): Open Matte Sundance Premiere Reconstruction!

I love the look of both of these and have had them downloaded for ages. However, I had never actually watched the upscaled version until now, and it turns out the audio is way out of sync. Is this true of the copy others have as well, or is it an issue that has been fixed somewhere along the line? The audio is roughly 2 full seconds ahead of the video.

This is not true of the "Tarantino's 35 mm print" version, which looks bloody marvellous by the way. I haven't had the pleasure of 35 mm Reservoir Dogs but I did get to see a super beat up 35 mm print of Pulp Fiction and it was a great experience.

EDIT: OK I lined up the two projects in Audacity and AviSynth and confirmed what I suspected: the audio is basically in sync between the two projects but the video isn't, so it might be that it was assumed that the two videos were in sync but they aren't. In Audacity, the audio appears to be only very slightly different, at The "Tarantino's 35 mm" project has 48 fewer video frames at the start than the "Open Matte" version, which makes sense given my earlier eyeballed estimate of a 2-second sync discrepancy given that in terms of duration 48 frames is...

48 / (24000/1001) = 2.002 seconds

Which in samples is

48 / (24000/1001) * 48000 = 96,096 samples

To fix it, I've therefore just trimmed 96,096 samples from the audio by applying a "delay" of -2002 ms.


Quick and dirty comparison script:

Code:

tinoprnt=FFMPEGSource2("Reservoir_Dogs_Downscaled.mov").showframenumber()
opendvd=BlankClip(48,640,360,"YV12",24000,1001,0) + FFMPEGSource2("Reservoir_Dogs_OpenMatte_Downscaled.mov").showframenumber()
StackHorizontal(tinoprnt,opendvd)

Yeah, you got that right. The open matte is out of sync, while the widescreen version isn't - that was mentioned in the comments on MySpleen. But I actually found another problem with the LD 2.0 pcm track - during the infamous ear cutting scene, right when Mr. Blonde turns on his favorite tune on the radio and start dancing to it, the balance shifts heavily to the left (or right, can't remember for sure now), and it goes this way until the very end. I can vouch for it, since I use earphones watching my movies nowadays...

[pipefan413]

Yeah, you got that right. The open matte is out of sync, while the widescreen version isn't - that was mentioned in the comments on MySpleen. But I actually found another problem with the LD 2.0 pcm track - during the infamous ear cutting scene, right when Mr. Blonde turns on his favorite tune on the radio and start dancing to it, the balance shifts heavily to the left (or right, can't remember for sure now), and it goes this way until the very end. I can vouch for it, since I use earphones watching my movies nowadays...

Oh, really? Is that in both versions then I take it, if it's the LD audio that's affected? I pretty much never use headphones for films! I'm a projector and surround speaker setup kinda guy.

[allldu]

Yeah, you got that right. The open matte is out of sync, while the widescreen version isn't - that was mentioned in the comments on MySpleen. But I actually found another problem with the LD 2.0 pcm track - during the infamous ear cutting scene, right when Mr. Blonde turns on his favorite tune on the radio and start dancing to it, the balance shifts heavily to the left (or right, can't remember for sure now), and it goes this way until the very end. I can vouch for it, since I use earphones watching my movies nowadays...

Oh, really? Is that in both versions then I take it, if it's the LD audio that's affected? I pretty much never use headphones for films! I'm a projector and surround speaker setup kinda guy.

Yeah, I used to be a surround speaker guy, but then a kid happened, and now I'm a Sony surround headphones guy for quite a while... Anyway, not sure about the open matte version - I never gave it a watch since I knew it was out of sync. But I assume, if the same PCM LD track was used for this one as well, then it should have the same problem. And it's really frustrating, because I love the work the guy did with the colors! So, I'm curious if this is the problem with this particular track, or it's on every RD LD out there?

[pipefan413]

Yeah, I used to be a surround speaker guy, but then a kid happened...

Hahaha! I'm with you.

Anyway, not sure about the open matte version - I never gave it a watch since I knew it was out of sync. But I assume, if the same PCM LD track was used for this one as well, then it should have the same problem. And it's really frustrating, because I love the work the guy did with the colors! So, I'm curious if this is the problem with this particular track, or it's on every RD LD out there?

OK so I'm just on a lunch break but I had a quick run through of both the 35 mm and Open Matte projects to see if I could reproduce what you were talking about but I can't seem to hear a problem. It might be that I'm hearing it fine because I'm listening in straight stereo but something's off with the steering when it's de-matrixed and you're hearing that because you've got "surround" headphones, I'm not sure. But in plain old stereo on fairly basic USB headphones, the sound appears to be coming straight down the middle for the most part with some nice separation here and there e.g. when music is prominent. Hmm...

[allldu]

OK so I'm just on a lunch break but I had a quick run through of both the 35 mm and Open Matte projects to see if I could reproduce what you were talking about but I can't seem to hear a problem. It might be that I'm hearing it fine because I'm listening in straight stereo but something's off with the steering when it's de-matrixed and you're hearing that because you've got "surround" headphones, I'm not sure. But in plain old stereo on fairly basic USB headphones, the sound appears to be coming straight down the middle for the most part with some nice separation here and there e.g. when music is prominent. Hmm...

Yeah, weird. I should probably check it with my plain stereo headphones too.

[pipefan413]

Yeah, weird. I should probably check it with my plain stereo headphones too.

If it sounds bizarre when de-matrixed then it might be something that's endemic to the Dolby Stereo / Dolby Surround / whatever the hell you want to call it. I'm wondering if maybe it's something to do with the alignment of the channels or something; I know someone who was experimenting with matrix decoding on optical RETURN OF THE JEDI audio found that it didn't seem to decode properly without some tweaks to the alignment of the channels. I don't know what it was that was wrong with it before the tweaks, though, so it could be similar to this or it could be completely unrelated.

But yeah, in plain stereo it sounds OK as far as I can tell without actually watching / listening to the entire thing!

EDIT: Listened again to the specific scene you mentioned rather than jumping about after that point and... oh yeah, there's definitely something going on there.

Loaded into Audacity and... whaddayaknow:



I bet I can tell you why it goes wrong at that point too: it's probably the side change on the LD, because it goes awry at the 1h mark.

Presumably the capture is spot on for side 1 but side 2 for some reason has unbalanced L/R with R being significantly quieter for... some reason.

Because I've only ever seen this resync mentioned in connection with this project I'm not actually 100% certain where it came from or who actually did the capture (I see various people talking about it in the thread: @spoRv, @Koopa Luath). What I'm thinking is that this was probably not a bit-perfect capture and there's a levels issue that for some reason doesn't manifest until the cap for side 2. Part of my reasoning is that the audio's presented as 48 kHz rather than 44.1 kHz, which it seems used to be the done thing up to a certain point and then people got more picky and/or better at captures so started archiving the 44.1 kHz with the 48 kHz being optional (which, incidentally, I am very happy about). I could be wrong though, I suppose it is possible that this is how the audio actually is on the LaserDisc; it just feels unlikely.

In fact, looking closer at the waveform, I think the levels are slightly out on side 1 as well, it's just much less pronounced. L is consistently louder than R throughout, it's just that it's not really very noticeable until somewhere around the 1h mark (and LD68993WS is an NTSC CLV with 60 mins per side so that makes a certain amount of sense).

I reckon I can fix this though. If I split the audio back into the 2 sides and then normalise L and R independently of one another, theoretically, they should match a lot closer than they currently do. It might not be bang on because this isn't mono and L and R don't contain the same data so there will be moments when the peaks on L will be higher or lower than R, which might throw this off somewhat. But I just gave it a very quick try and it sounds much better already. Only trouble is I'm about to go on holiday for a week so I might not get it done before I go. I really want to now that I know I can do it though, hahah...

EDIT: Found the side change. It's this wee blip in the waveform. After that, R drops way down and is really quiet.



PLAN:

1. Split the audio off the main "Tarantino's 35 mm print" project back into the 2 sides at the identified mark
2. Normalise the channels and sides independently to make all 4 chunks of audio more consistent with one another (more specifically: amplify R on side 2 to better match side 1, then amplify/normalise R to better match L (which as far as I can tell, seems to be consistent throughout, at least)
3. Rejoin
4. Apply a tiny crossfade at the side change to cover the audible pop that's currently there (presumably due to the caps of the 2 sides having been joined directly to one another without any kind of adjustment to cover the gap)
5. Re-adjust the fixed audio to the "Open Matte" / "Premiere" version as previously described (-2002ms)

This may or may not get done before I leave the house to go away on a brief "holiday" (as far as anyone can safely have a holiday during COVID-19 anyway). If it doesn't, I might attempt to take a laptop and the files with me, or I might decide that it can wait until I get back in a week. Either way, I'll get it done, unless TAF objects / would rather do it himself.

UPDATE: Right OK so I'm now pretty confident this is either a capture problem or maybe something wrong with the actual LD source itself, however unlikely that may seem. I split the audio back into 2 halves then split the channels as well, and compared their peak amplitudes. Here's how that looks:

Side 1 L peak -7.2963 dB
Side 1 R peak -7.6095 dB (+0.313 dB = -7.2963 dB)

Side 2 L peak -7.7708 dB
Side 2 R peak -9.7100 dB (+1.939 dB = -7.7708 dB)

The trouble is, boosting R by those amounts is not sufficient to fix the issue. In order to boost R sufficiently, it would cause clipping, and it isn't just different in the peaks, but also in the areas where the level is lower; it's as if L has compressed dynamic range but R doesn't, somehow. So something a little more sophisticated is probably in order here.

I broke out the fancy stuff in RX to see how it handled it without much manual intervention from me. It can do "adaptive level matching" which automatically does exactly what I'm trying to achieve: it compares the level of the right channel to the level of the left channel, and brings it up or down to match. What it does if this causes clipping I don't yet know, but I guess I'm about to find out. Also, this will manipulate the right channel more than if I'd just done manual adjustments, which is not what I would personally prefer (e.g. what if L or R is *supposed* to be louder in a particular moment, e.g. when not much is happening in one channel but is in another?) but then I don't think this is really going to be "fully" fixable otherwise unless the audio is recaptured from scratch.

Which brings me to the final point then: if anybody still has the LaserDisc (I don't think it necessarily matters if it's LD68993WS or LD68993) and is able and willing to do a bit-perfect recap of it, that would probably be the best solution here by a country mile. It would of course need to be resynced again which is a pain but it's better than having funky levels across the whole track.

[pipefan413]

Right I need to stop editing one giant post and this is probably more likely to be seen if posted separately so here you go... a quick side by side of the track as it appears in the TAF projects and then (track 2) after RX's automatic, adaptive channel level matching: https://mega.nz/file/NhtnTDaT#ZG73KZ_pB2...ZHOM09WiKU Notice around the 7 second mark the right channel significantly drops and remains much lower, as @allldu says. Flip to track 2 and compare the same bit and everything afterwards and it's much better.

I reckon for the time being I'll encode this to DTS-HD MA 2.0 with surround flag for both the 35 mm and Open Matte versions and worry about uploading them muxed into anything once I'm back from a well-needed break and change of scenery. I'm sure the fix isn't perfect but it's almost certainly better than it was beforehand. PM if you're one of the regulars 'round these parts and want me to send what I've got (don't want to be posting public links here, beyond the sample above).

[The Aluminum Falcon]

Oh my. Hey guys. Following the chat here, I'm going to look back at my original files and perhaps figure out what went wrong! Sorry for being AWOL.

I have a suspicion I know what happened to make the open matte so out of sync. I think I accidentally muxed the track synced to the main 35mm recreation project to that as well. With luck, I'll look through my original project files this weekend and see if I can find the properly synced version.

Regardless, I think that I ripped the Reservoir Dogs LD myself if I recall correctly but it was definitely not bit-perfect, so not sure what caused the glitch with the Side 2 audio.

PM'ed you, regardless pipefan413! Thanks to you and allldu for spotting these things!