Critique my capture/post processing for YouTube upload?

[aramkolt]

Hello all, finally got past the stage of building and troubleshooting various aspects of my capture PC that would accept an AGP AIW 9000 Pro card and I'm trying to get an idea if there is much that I should expect to improve beyond what I've got here, or if there are some obvious issues to anyone can see with the capture itself.

Capture PC:

Runs XP
ASRock 4CoreDual-VSTA motherboard with Core2Extreme processor - Had to underclock to 2.4Ghz to get it to boot for some reason
4GB of ram (3.2GB visible to the system due to MOBO and 32 bit XP limitations)
New 700W modern ATX power supply and modern case from Amazon
512GB SSD for the OS to run on (SATA)
2TB capture HDD (SATA)
AIW 9000 Pro 64MB video/capture card - Recapped all SMT and through-hole capacitors
4ft custom 8 pin to s-Video cable (to keep S-Video cable as short as possible, extra wires are not needed as audio in goes directly to TBSC sound card)
Turtle Beach Santa Cruz PCI Sound Card

VCR
Personally refurbished AG1980P with TBC/YC card, power supply, front display board, control board, Head Amp, RF Amp all fully recapped, others recapped as needed.

SOURCE - This was a commercially produced promotional video handed out in 1998 at the 80 year anniversary of the John Deere Waterloo Works in Waterloo, Iowa. Tape probably has had minimal playthroughs and is in very good condition, though it is getting to be 25 years old.

Video chain - for this test, it was just VCR straight to AIW card via S-Video without a full frame TBC, line TBC within the VCR was turned on. No dropped frames were noted during the capture.

Captured under "DVD-High" profile in MMC as MPEG2 - Raw file and the deinterlaced post processing files are both attached. I have also tried capturing in lossless and that gives a bit different of a color profile mainly, however I believe that YouTube eventually re-encodes at the same 4:2:0 that MMC does for MPEG.

Post processing - Used the Mpeg2 raw capture file also attached as the input to Hybrid with Bob deinterlacing to 59.94FPS, Cropped 8 pixels on both sides and 10 pixels from the bottom where you can see the head switching noise. This was then resized to 960x720 and bitrate was 5000Kbps. Eventually I'd like to have it be YouTube compliant which to my understanding allows for 8000kbps for 60fps 720P video, but that would be a bit above the 100MB limit for attachments here. I didn't mess with any "deblock", "dehalo", or sharpening settings which I understand can be helpful, would appreciate any recommendation if it looks like those would be beneficial here and to what degree. It's kind of interesting that both the input and post-processed files are about the same size, yet the post processed one looks a lot cleaner due to modern encoding and better deinterlacing.

I didn't mess with the pixel aspect ratio, but I do realize that could be off a bit with my cropping and resizing I did there. I'm open to suggestions about best ways to crop and resize within hybrid if that isn't ideal - I know there's something that can be done with pixel aspect ratios (PAR) but seems like just resizing would accomplish the same thing if plan is to be eventually viewing on YouTube, laptop, or a digital TV more or less used as a display. For YouTube, you want the resolution to be at least 720 vertically as it allows more bandwidth and I believe that is the lowest vertical resolution that allows for 60 frames progressive.

Also is interesting that supposedly YouTube will directly accept MPEG2 as an upload, so I'll have to see if it does its own sort of transcoding/deinterlacing in some way that I suppose could even be better than what Hybrid does (but I doubt it haha). Also that would mean you're stuck with the lower image noise and side bars most likely which isn't ideal.

I will say that the colors on the post-processed file look oversaturated, but I also can't say that looks particularly bad. Not sure what setting in hybrid I'd use to change that? I figured it would look pretty similar to the input since I left most of the settings at defaults other than crop/resize, deinterlace. Could have something to do with a color space conversion maybe?

Any suggestions/comments appreciated - I'm a total novice, so I don't expect to have "gotten it right" first try. I do have some other fancy equipment that I'm not exactly sure how I'd integrate into the capture system including analog pattern generators, analog waveform monitor and vectorscope, as well as a couple of TBCs (DPS-235 and Hotronic AP41 with the S-Video in/out as well as DOC (dropout compensation) upgrades).

[latreche34]

720p is horrible for youtube, you may as well leave it at 480p, You can upscale to 1080p for less compression artifacts.

[aramkolt]

Gave it a try at 720p, here's the result:
https://www.youtube.com/watch?v=ByyXFIsFUAg&t=1s

It didn't take much time to process, so it might not have actually re-encoded it. The audio does seem to have been converted as it has a high pitched noise I didn't hear in the file playback

[latreche34]

Your aspect ratio is wrong though, 1440x1080 gives you the right aspect ratio.

[timtape]

Quote:

Originally Posted by aramkolt

Gave it a try at 720p, here's the result:
https://www.youtube.com/watch?v=ByyXFIsFUAg&t=1s

...The audio does seem to have been converted as it has a high pitched noise I didn't hear in the file playback

There's an intermittent high frequency whistle around 14 kHz, best seen on a spectrograph and (for those with challenged hearing) best heard by slowing down the playback. I believe HiFi VCR recordings, which yours sounds like, can contain frequencies higher than the nominal 20 kHz which may lead to problems when uploading to YT. Sometimes before uploading it's best to check this out with a spectrograph and if necessary filter out such frequencies before uploading or compressing.

[lordsmurf]

Technically, the ATI MMC MPEG-2 loses some quality, downconverting to 4:2:0, but Youtube will do that later anyway.

How is this being deinterlaced? Hopefully QTGMC, but tip = do not use "slower" because it softens. Use faster, then add switches if issues.

[aramkolt]

What tool is recommended for the audio histogram and then what tool to reduce the amplitude of any spikes? I did consider using a graphic equalizer at the time of capture and just turning the highest band down as low as it would go so that it wouldn't be there at the time of capture to save a step later, but guessing that isn't ideal, but perhaps that could save a step in the workflow later? My equalizer has a 16khz and a 20khz band, so turning down the 20k I feel wouldn't loose much.

It seems the audio whistle gets added when resampling the audio in hybrid rather than at the YouTube upload step, is there anything that can be done within hybrid alone to reduce high frequency noise? I saw an option for "DRC" which is dynamic range compression, but not sure if it would help with noise

I'll have to give 1440X1080p a try for YouTube and see how that looks.

I did use QTGMC within hybrid and I think it was just defaults which is the "fast" setting.

Is it better to mask or crop off the 10 pixels of noise at the bottom in the source and then just be ok with the black bar at the bottom, or is it ok to crop those 10 pixels and just allow that bit of vertical stretch to fit vertically? Seems like 10 pixels out of 480 would be relatively imperceptible since it's about 2% of the total height?

Thanks for all the feedback!

[latreche34]

The legal resolution of a 4:3 frame is 704x480 for NTSC captured from analog tapes, You can offset the bottom crop with more horizontal crop, for 10 pixels vertically it is about 13 pixels horizontally but doesn't have to be precise, you will not notice a deformation of less than 10 pixels. Crop and mask if you are going to keep it 480, If resizing to HD, crop and immediately resize while still lossless before encoding.

[timtape]

Quote:

Originally Posted by aramkolt

What tool is recommended for the audio histogram and then what tool to reduce the amplitude of any spikes? I did consider using a graphic equalizer at the time of capture and just turning the highest band down as low as it would go so that it wouldn't be there at the time of capture to save a step later, but guessing that isn't ideal, but perhaps that could save a step in the workflow later? My equalizer has a 16khz and a 20khz band, so turning down the 20k I feel wouldn't loose much.

It seems the audio whistle gets added when resampling the audio in hybrid rather than at the YouTube upload step, is there anything that can be done within hybrid alone to reduce high frequency noise? I saw an option for "DRC" which is dynamic range compression, but not sure if it would help with noise...

I've uploaded a screen shot of the standard Izotope display and included in black and white is the older type audio spectrum analyzer. Both show essentially the same thing but in different ways. I use both. But naturally so much more is revealed in the moving image displays as the audio is played back.

These sorts of displays are common amongst various competing audio recording, mastering, restoration brands these days. I get most use out of the Izotope "essentials" which is not expensive. Various special deals come up every now and then. Some offer a generous free trial period.

It's possible the recording has been "overcooked" at some stage in the process. Being such high frequency and inaudible to many, perhaps the problem wasnt even noticed in production.

Except for the man's opening speech which is muffled and with high background noise, the rest of the audio is to my ears well produced but too treble heavy and too bass light, which can mostly be fixed with careful EQ. I'd use broad, gentle sweeps, not narrow band. Good luck.

[aramkolt]

Ah, I'll have to take a look at Isotope.

Here's the two videos, one rendered to 1080p vs 720p on YouTube. The 1080p one is 13000kbps vs the 720p being 8000kbps as I believe those are the max bitrates it'll give to them.

Color looks a bit different on the 1080p one, but otherwise I can't say I'm seeing a whole lot in terms of quality improvement/artifacts.

This shows both videos which you can play simultaneously using the play button at the bottom:
https://viewsync.net/watch?v=ByyXFIs...&t=0&mode=solo

The AG1980 does have some SMT caps on the audio II boards that usually refurbishers don't touch since the audio almost never has significant issues, but could be contributing too I suppose. I can't recall if I changed them on this unit or not, but that could be contributing maybe.

I should be able to get an audio spectrum analysis at capture and compare it to playback of the digital file and try to chase down where that's happening - could also be that it was recorded that way - being a low budget promo video, probably didn't undergo super rigorous audio scrutiny haha.

Thanks again for all the suggestions!

[latreche34]

The color looks different is because you probably did not flag the 1080p footage for rec.709 for HD standard.