Audio desync with Dazzle + amarec

DJS

Edit history:

DJS: 2013-01-18 11:25:06 am

DJS: 2013-01-18 10:17:07 am

DJS: 2013-01-18 07:50:32 am

DJS: 2013-01-18 07:46:44 am

DJS: 2013-01-18 07:45:12 am

2013-01-18 07:42:25 am

torch slug since 2006

for my birthday, i decided to get a Dazzle DVC101, since my easycap (the one you sent me, nate) started having connection issues with the s-video port.

anyway, i got the dazzle, and im having some issues with it:
* 720x480 causes tons of frame drops and bobbing (640x480 seems to work good)
* audio desync. synced up at the begining but gets worse as it goes on...
see/hear for yourself: http://lambdan.com/dazzle (MQ only, since this isnt really a quality test.)

environment: windows 7 x64, amarectv 2.10b, lagarith codec, drivers from here
game/console: metroid prime on wii, s-video NTSC

i read on http://notesmash.org/krystal/vdub.html , and i think that it addresses "interleaving", which i think is my problem. i tried creating a .avs file with:

Code:

AVISource("amarec(20130118-1517).avi").DelayAudio(-0.210).Normalize()

* 210 was the relative latency value i got in virtualdub

this didnt solve the audio desync

so yeah, if anyone knows whats wrong, let me know. i know people do use dazzle + amarec to record their runs, so its possible. with virtualdub, the audio/video seems to sync (i say seem, cause it might be placebo) but i very much prefer amarec.

edit: audio is hooked up via line-in. i have the same desync issues with dazzle audio too.
edit2: dazzle dvc100 =/= dazzle dvc101; they are the same thing driver wise atleast

Thread title:

for my birthday, i decided to get a Dazzle DVC101, since my easycap (the one you sent me, nate) started having [url=http://www.youtube.com/watch?v=h4M80u-zD-Q]connection issues with the s-video port[/url].

anyway, i got the dazzle, and im having some issues with it:
* 720x480 causes tons of frame drops and bobbing (640x480 seems to work good)
* audio desync. synced up at the begining but gets worse as it goes on...
see/hear for yourself: http://lambdan.com/dazzle (MQ only, since this isnt really a quality test.)

environment: windows 7 x64, amarectv 2.10b, lagarith codec, drivers from [url=http://cdn.pinnaclesys.com/SupportFiles/Hardware_Installer/readmeHW10.htm]here[/url]
game/console: metroid prime on wii, s-video NTSC

i read on http://notesmash.org/krystal/vdub.html , and i think that it addresses "interleaving", which i think is my problem. i tried creating a .avs file with:
[code]
AVISource("amarec(20130118-1517).avi").DelayAudio(-0.210).Normalize() 
[/code] 
* 210 was the relative latency value i got in virtualdub

this didnt solve the audio desync

so yeah, if anyone knows whats wrong, let me know. i know people do use dazzle + amarec to record their runs, so its possible. with virtualdub, the audio/video seems to sync (i say seem, cause it might be placebo) but i very much prefer amarec.

edit: audio is hooked up via line-in. i have the same desync issues with dazzle audio too.
edit2: dazzle dvc100 =/= dazzle dvc101; they are the same thing driver wise atleast

nate

2013-01-18 04:22:41 pm

720x480 versus 640x480 doesn't matter since anri handles the aspect ratio. the rest i don't know about unfortunately ... it could be a driver issue or it could be something else entirely.

DJS

Edit history:

DJS: 2013-01-20 07:33:55 am

DJS: 2013-01-20 04:03:17 am

2013-01-20 03:59:37 am

torch slug since 2006

okay so i made a workaround. if you read the 60fps streaming topic, you know how:
amarec (dazzle as input, 29.97 fps, 640x480), live output enabled
virtualdub (using amarecs live output, 640x480, lagarith)

with this, audio actually syncs. so im wondering if this quality is good enough. it does not use anri-chan's deinterlacer, since amarec outputs it progressive (after deinterlacing it itself), so i hope that's ok. the pro's of using this setup is that virtualdub records it in 60fps (amarec caps capturing at 30 fps for some reason)

game is F-ZERO GX, NTSC played on Wii S-Video... very intensive game

also anyone that knows about intel vs amd,

do you know if intel cpu's perform better than amd's when encoding? reason why im asking, is that my mac mini has a i5 (a mobile one) at 2.3ghz , and my main rig has a phenom2 x4 @ 3.7ghz. it may sound stupid, but im wondering if the i5 would perform better for streaming and recording and rendering. like, if the encoders "scale" better with intel cpus or something. and dont mind that its on a mac, i would run windows via boot camp.

http://obsproject.com/estimator - even if its just a estimator, if you put a 2nd gen i5 on the cpu it recommends 30 fps, while if you put a amd processor, it recommends 25 fps.

accidently hit X next to the IQ.... F zero GX IQ

EDIT: for the heck of it, to try composite and pal-60, i recorded some gta4 from 360.
LQ MQ HQ
IQ
log

okay so i made a workaround. if you read the 60fps streaming topic, you know how:
amarec (dazzle as input, 29.97 fps, 640x480), live output enabled
virtualdub (using amarecs live output, 640x480, lagarith)

with this, audio actually syncs. so im wondering if this quality is good enough. it does not use anri-chan's deinterlacer, since amarec outputs it progressive (after deinterlacing it itself), so i hope that's ok. the pro's of using this setup is that virtualdub records it in 60fps (amarec caps capturing at 30 fps for some reason)

game is F-ZERO GX, NTSC played on Wii S-Video... very intensive game

also anyone that knows about intel vs amd,[spoiler] do you know if intel cpu's perform better than amd's when encoding? reason why im asking, is that my mac mini has a i5 (a mobile one) at 2.3ghz , and my main rig has a phenom2 x4 @ 3.7ghz. it may sound stupid, but im wondering if the i5 would perform better for streaming and recording and rendering. like, if the encoders "scale" better with intel cpus or something. and dont mind that its on a mac, i would run windows via boot camp.

http://obsproject.com/estimator - even if its just a estimator, if you put a 2nd gen i5 on the cpu it recommends 30 fps, while if you put a amd processor, it recommends 25 fps.[/spoiler]

accidently hit X next to the IQ....  [url=https://dl.dropbox.com/u/60071552/fzerogx_IQ.mp4]F zero GX IQ[/url]

EDIT:  for the heck of it, to try composite and pal-60, i recorded some gta4 from 360. 
[url=https://dl.dropbox.com/u/60071552/gta4qt/gta4_LQ.mp4]LQ[/url] [url=https://dl.dropbox.com/u/60071552/gta4qt/gta4.mp4]MQ[/url] [url=https://dl.dropbox.com/u/60071552/gta4qt/gta4_HQ.mp4]HQ[/url]
[url=https://dl.dropbox.com/u/60071552/gta4qt/gta4_IQ.mp4]IQ[/url]
[url=https://dl.dropbox.com/u/60071552/gta4qt/log_276645085.txt]log[/url]

Attachments:

2013-01-20 08:45:14 am

f-zero gx looks good. does anyone know what amarec's deinterlacer is? it's very good - can tell by looking at the smallest numbers as they change frame by frame. you could easily fool me and say this was originally progressive.

gta3 looks great too. man it's good to test launching both your car and yourself.

intel and amd are just brand names, the cpus aren't necessarily any better. but in this case, phenom ii is like a core 2 duo, so of course the amd is slower than the intel i5. if the number of cores is the same then the i5 will almost certainly be faster. unfortunately amd hasn't released anything quite as good as intel's latest chips. this situation is reversed from a few years ago, when amd had the fastest cpus.

DJS

Edit history:

DJS: 2013-01-20 09:08:33 am

2013-01-20 09:07:47 am

torch slug since 2006

wow, thanks. i was afraid you would reject them since they didnt use anri-chan's deinterlacer.

my mac mini has a i5-2415m (note the m for mobile), (2 cores, 4 threads, so fake quad-core). my phenom2 is a real quad-core. so, the i5 would be faster?

i think i might record some footage, and then see which one spits out a IQ fastest, unless there is a better way to benchmark this.

nate

2013-01-20 09:09:57 am

yeah i don't know. the amd may be slightly faster.

i noticed on the amarec site it says it supports hardware deinterlacing which makes me go hmmmmm.

DJS

2013-01-20 09:14:03 am

torch slug since 2006

yeah im not sure about the difference
for f-zero gx i think i had it set to Retro Game, and for GTA4 it was set to role-playing game.
im not sure if they make any difference, but im pretty sure that Retro Game makes it F1 (it bobs for F2 stuff), and Role-playing game is F2 (it doesnt bob), Action Game seems to be like Retro Game, but it only bobs at the top...

Attachment:

Skärmklipp.PNG

nate

2013-01-20 09:21:24 am

weird, i would expect retro game to just field split and output d4. but i guess that's not what this is. don't know about the other two either. it's possible it's motion compensation versus interpolation. interpolation has made a big comeback in the last 2-3 years.

DJS

Edit history:

DJS: 2013-01-20 09:25:29 am

2013-01-20 09:23:22 am

torch slug since 2006

if you want to, tell me what game you wanna see and i can record some samples with each setting. maybe you can see what is what.
oh, and if you want raw files or not.

nate

2013-01-20 09:32:21 am

if you have the time, it would be cool to see an actual old school 16-bit game (like sonic 2 if you have it) played on a mega drive with each of the four options (does the hw deinterlacing work for you?). that should probably be enough for me to figure out what each one is. can you capture to lagarith? that should keep the file size down. i only need about 20 seconds of you running around.

DJS

Edit history:

DJS: 2013-01-20 09:50:29 am

2013-01-20 09:34:51 am

torch slug since 2006

yeah i capture lagarith. basically i was thinking i set the deinterlacing option in amarec (amarec then outputs that option to the live output, virtualdub), and in virtualdub i just record to lagarith.

sadly, i dont have sonic 2 or a mega drive. i guess i could play them on the wii vc, but im not sure if wii vc but i dont know how you feel about that. i do have a nes and a snes though. on the nes i have smb1, smb3, zelda1, tetris, and my snes has mario world, yoshis island, zelda3, actraiser....

edit: hw deinterlacing option doesnt work (virtualdub gets the Amarec "no signal" picture). i played some sonic 2 on vc, but the video files got around 400mb each, so idk if you want them.

nate

2013-01-20 09:53:37 am

yeah vc is no good because it's deflickered. i guess try zelda 3 since that has quite a bit of text and you can run around pretty fast.

DJS

2013-01-20 10:04:27 am

torch slug since 2006

okay i got the 3 videos now. they're the raw files from virtualdub, so 640x480, lagarith yuy2, no audio (to keep file size down). they are 651mb in total.
how do you want them? can i attach them or are they too big.

nate

2013-01-20 10:12:45 am

cool. mediafire or whatever works.

DJS

2013-01-20 12:59:13 pm

torch slug since 2006

https://dl.dropbox.com/u/60071552/alttp/action_game.avi

https://dl.dropbox.com/u/60071552/alttp/retro_game.avi

https://dl.dropbox.com/u/60071552/alttp/roll_playing_game.avi

i hope the download speeds are acceptable, because the upload speed sure as hell wasnt.

nate

2013-01-20 03:54:24 pm

all three videos are f1.

action game seems to be a field split and an upscale with some kind of interpolation. i notice that the image is moving up and down two pixels every frame; this is probably the same problem anri's 1-pixel bob is meant to address.

retro game seems to be a field split and an upscale with no interpolation. this is the correct display for d4 material (minus the upscale, but it's harmless). it corresponds with norichan's "f1" display mode.

rpg seems to be some kind of smart bob. interlacing artifacts are visible in areas with motion. this deinterlacing mode is probably called rpg because there isn't much motion in rpgs and for d1 ones you can get the whole vertical resolution with the most obvious interlacing artifacts removed by the smart bob. this is equivalent to norichan's "deinterlaced" display mode (may not be in released versions of norichan). of course for d4 stuff you should use "retro game".

based on this analysis my guess is you actually used "action game" for that f-zero gx test. if not then you should have - it doesn't have the 1-pixel bob with d1 input, right? otherwise maybe the interpolation used by the smart bob ("rpg") is the same as with "action game" and so much was detected to be moving in your clip that it ended up equivalent to "action game," lol.

so, to conclude, if you are playing:

anything d4: use "retro game"
anything d1: use "action game"

i'm not convinced that the interpolation it uses matches up with the quality of mvbob on interlaced input. i think it will probably depend on the specific game. f-zero gx is very challenging for motion compensation deinterlacers like mvbob.

all three videos are f1.

action game seems to be a field split and an upscale with some kind of interpolation. i notice that the image is moving up and down two pixels every frame; this is probably the same problem anri's 1-pixel bob is meant to address.

retro game seems to be a field split and an upscale with no interpolation. this is the correct display for d4 material (minus the upscale, but it's harmless). it corresponds with norichan's "f1" display mode.

rpg seems to be some kind of smart bob. interlacing artifacts are visible in areas with motion. this deinterlacing mode is probably called rpg because there isn't much motion in rpgs and for d1 ones you can get the whole vertical resolution with the most obvious interlacing artifacts removed by the smart bob. this is equivalent to norichan's "deinterlaced" display mode (may not be in released versions of norichan). of course for d4 stuff you should use "retro game".

based on this analysis my guess is you actually used "action game" for that f-zero gx test. if not then you should have - it doesn't have the 1-pixel bob with d1 input, right? otherwise maybe the interpolation used by the smart bob ("rpg") is the same as with "action game" and so much was detected to be moving in your clip that it ended up equivalent to "action game," lol.

so, to conclude, if you are playing:

anything d4: use "retro game"
anything d1: use "action game"

i'm not convinced that the interpolation it uses matches up with the quality of mvbob on interlaced input. i think it will probably depend on the specific game. f-zero gx is very challenging for motion compensation deinterlacers like mvbob.

DJS

Edit history:

DJS: 2013-01-21 08:38:30 am

DJS: 2013-01-21 08:17:12 am

2013-01-21 01:21:08 am

torch slug since 2006

i wanna recall that using retro game for super mario 64 (d4 f2), makes it bob. but rpg mode works good with sm64, so are you sure all 3 are f1?
when i get up and feel like it, ill probably record some f-zero in action mode (just to confirm), and some sm64 in retro mode and show them to you.

edit: ok i recalled wrong, sm64 works with retro. now on to f-zero gx + action

edit 2 : heres fzero, a iq, recorded from Action game deinterlacing. i think there are some frame drops around 01:19 though, but im not sure, maybe you can tell better. https://dl.dropbox.com/u/60071552/fzero_gx_action_IQ.mp4

nate

2013-01-22 11:30:27 am

can you capture some video of that same track interlaced? that way i can compare this and the output of mvbob.

DJS

2013-01-22 11:47:27 am

torch slug since 2006

can i capture with amarec, without audio? or do you want virtualdub?

nate

2013-01-22 12:34:58 pm

however you can get raw interlaced video from the card. don't need audio.

DJS

2013-01-22 01:17:21 pm

torch slug since 2006

went with virtualdub, still not sure if amarec captures 30 fps only so... i hope 16secs is enough
https://dl.dropbox.com/u/60071552/asd.avi

nate

2013-01-22 04:41:16 pm

not a 100% fair comparison because the amarec one was compressed (iq) but some differences are obvious nonetheless: the loss of detail on the portraits at the left, the stairstep on the small map, and the loss of detail on the tower in the background. i guess mvbob is still the champion.

Attachments:

amarec_highlighted.png

mvbob_highlighted.png

ballofsnow

2013-01-22 09:22:53 pm

Quote from DJS:

went with virtualdub, still not sure if amarec captures 30 fps only so... i hope 16secs is enough
https://dl.dropbox.com/u/60071552/asd.avi

Man, what's with that video. Whether you do assumetff or assumebff and then separatefields, the frames keep jumping backwards and forwards.

DJS

2013-01-23 03:17:59 am

torch slug since 2006

Quote from ballofsnow:

Man, what's with that video. Whether you do assumetff or assumebff and then separatefields, the frames keep jumping backwards and forwards.

i have no idea, it might be fixed if i switch to amarec and just stick with 30 fps.

im thinking, can i record audio with audacity or something? and then just include the audio track in the video. i dont really wanna deal with syncing though, but maybe its worth it, if i wanna get a run up here on sda.

for youtube/not sda, ill probably just use twitch highlights, even if that is kinda sad but atleast the audio is synced.

nate

2013-01-23 08:35:00 am

snow, he captured interlaced material at f1. so you have to changefps(29.97) right off.

recording the audio separately and syncing it yourself is ok of course if you want to do the work. maybe time using a known sound effect at the start of every segment or something.

have you considered locating another easycap and using norichan? i'm not saying it would be easy to find a compatible one, just wondering.