Log in

View Full Version : Speex in ogm/mkv files


perki
1st September 2003, 19:03
Hi!

We are just producing Multilanguage Screen Capture Tutorial Videos.

We would like to release multilanguage videos, that are playable on Windows, Linux and Mac OSX, at least.

Using ogm or mkv as the container seems to be the best aproach, but the vorbis codec is not very optimized for voice encodings. ~30 kBit/s is really a lot for voice w/o music.
So the Speex codec would be a very good alternative, so we could go down to ~10kBit/s for one audio stream. But my question is, how these streams could be muxed into a mkv or ogm file. I only got the ACM version of this codec and I can mux the stream via a wav file into a avi, but this will not help an Macs and Linux stations. Perhaps you got an idea,

thanx in advance, Perki.

ChristianHJW
1st September 2003, 20:21
Using the ACM codec, it should work out of the box in both containers .... just test it in VirtualdubMod and report if it does ....

Sirber
1st September 2003, 21:23
"Windows, Linux and Mac OSX"

I would go for RMVB+RMs with a SMIL structure. RealPlayer is avalible on all platforms you need :) RealAudio can go down to 6 KB/s IIRC.

robUx4
2nd September 2003, 09:39
You should contact the speex developper and tell him about your problem. He will probably tell you only Linux + ogg is worth ;)

Making a Directshow decoder will probably not be too hard, or maybe the ACM already does coding/decoding. On Linux I'm not sure MPlayer or VLC can read speex yet. But when they will, it will work on Mac too !

perki
2nd September 2003, 12:40
Originally posted by Sirber
I would go for RMVB+RMs with a SMIL structure. RealPlayer is avalible on all platforms you need :) RealAudio can go down to 6 KB/s IIRC.I have tried Helix Producer, but I have to say, I know nothing about Real Media stuff. The Producer, doesnt let me change the settings of Audiences - it wants me to upgrade to a plus version, which costs $199,-. But I guess, my low knowledge of this topic is causing my irritations using Real Media technology. In the end, real would be the best format. Variable fps, a feature which is VERY helpful for screen capturings! If it support multilang, I still have to find out.

If you want to try, what I got working using xvid and vorbis, you can try: [da].ogm']this Link ('http://bau2.uibk.ac.at/perki/typo3/testvid/shortcuts_[en). This is a multilanguage ogm file.

Sirber
2nd September 2003, 12:56
RealVideo supports VFR, if you use DropDupe :). RM supports also multilanguage, but in separates files:

video.rmvb
english.rm
french.rm
english.smil
french.smil

IMO you should invest a little in Real, coz it will save you troubles :) RealPlayer works on all the platforms you need, whitout much troubles. There is a free version of producer that works in commandline, fellow the "Real" link in my signature.

Good luck! :D

Belgabor
2nd September 2003, 14:36
Matroska also supports vfr (see the respective thread in this forum :))

perki
3rd September 2003, 10:48
hmm... I just muxed a xvid video with a .rm voice encoded audio stream into a matroska file and this plays perfectly on my pc. Since VirtualDubMod doesnt offer to mux real media streams, this feature has been untested before (I used Sir Elvis mmg to do the job now). So it would be possible to create multilang videos, using the powerful real voice encoder and mkv.

I didnt manage yet to get good image quality, by using helix producer. But This I dont understand, cause it supports vfr and this is a killer feature for these kind of videos. Perhaps I still make mistakes - perhaps Sirber, could you encode a little demo avi, I send you, into a real video stream, so I at least know, whats possible, if I spent for time on that. (This is for the GPL CMS Typo3 and has no commercial background).

This (http://markus.oh-my-god.dk/t3dl/shortcuts_RLE.avi) is the footage to be converted.
[da].ogm']This ('http://bau2.uibk.ac.at/perki/typo3/testvid/shortcuts_[en) is an ffmpeg version I made, with vorbis sound and muxed into an ogm container (this has 3.9 mb) and
[da].wmv']This ('http://bau2.uibk.ac.at/perki/typo3/testvid/shortcuts_[en) was a MS WMV Screen Capture Sereis 9 version, which has only 1.5 MB.

Sirber
3rd September 2003, 12:51
RealMedia in Matroska can't work on Linux and MacOS IIRC.

[edit]

I'm downloading it. What is your target bitrate?

[edit 2]

Done. Q is nice on them all. Can someone do a 20kbps XviD clip (video only) from that sample? :D

45 kbps video, 5 kbps sound:
http://sirber.no-ip.com/~sirber/50kbps.rmvb
20 kbps video, 5 kbps sound:
http://sirber.no-ip.com/~sirber/25kbps.rmvb
18.5 kbps video, 6.5 kbps sound:
http://sirber.no-ip.com/~sirber/25kbps_BS.rmvb

Settings:

2-pass VBR
EHQ @ 85
DropDupe @ 999 MAX dropped

[edit 3]

Brave people of doom9, here I challenge you. Create a better quality same filesize clip than mine. I aim for 25kbps (the second one, 25 kbps.rmvb). No resizing allowed. Good luck to you all :D

celtic_druid
3rd September 2003, 15:36
Here's an XviD version with your audio 2pass, probably could have made it smaller to (It's 1.18MB's).

http://members.optushome.com.au/cdromguide/test.mkv

Sirber
3rd September 2003, 16:56
can you post your settings?

celtic_druid
3rd September 2003, 17:27
This (http://members.optushome.com.au/cdromguide/test2.mkv) one is under one meg.

Here (http://members.optushome.com.au/cdromguide/test3.mkv) is one that actually matches your filesize, well almost.

As for settings:

Motion 6
Quant MPEG
VHQ 4
Max I frame 250
Min I frame 5 (suppose I should have changed them, I mean the clip hardly has more than 250 frames)
qpel
chroma mot
bframes 2,150,75,0 DX50 BVOP

Sirber
3rd September 2003, 18:16
hum... for the same filesize, you can have higher bitrate in matroska. I'll convert them into matroska to see.

[edit]

I've seen test3.mkv, and there is wierd mice effect at the beginig, and noticable muskito noise arount the texts.

a new one, at 30kbps, with a more time between keys:
http://sirber.no-ip.com/~sirber/30kbps.rmvb

If I use less kbps, il will produce more artefacts like bluriness and muskito noise.

APF_Gandalf
4th September 2003, 04:32
Originally posted by Sirber

Brave people of doom9, here I challenge you. Create a better quality same filesize clip than mine. I aim for 25kbps (the second one, 25 kbps.rmvb). No resizing allowed. Good luck to you all :D [/B]

since you haven't asked for a particular codec, I tried with something else than Xvid/divx :D
here is the codec:
http://www.getafile.com/cgi-bin/merlot/get/techsmith/TSCC.exe

and here is the video file (hope it won't be deleted too fast by the free provider)
http://www.angelfire.com/wizard2/gandalf11/test-TSCC.avi

enjoy.

EDIT:
sorry, I haven't tried to reach a specific filesize but I used "maximum quality" for this codec.

BoNz1
4th September 2003, 04:47
Ah, you all need to improve your googling, :eek: I found a speex directshow here, http://www.speex.org/download/speexDS_setup.zip and source http://www.speex.org/download/speexDS_source.zip I have no idea how it works though, I've never tried it.

celtic_druid
4th September 2003, 05:05
Nah, tried that some time ago, could never get it to render properly back then and I still can't.

Kinda hard to compete with a lossless screen capture codec, but here is another try with xvid anyway test4 (http://members.optushome.com.au/cdromguide/test4.mkv)

Can you decode techsmith under OSX and Linux though?

APF_Gandalf
4th September 2003, 05:41
Min I frame 5 (suppose I should have changed them, I mean the clip hardly has more than 250 frames)
iirc, Xvid just ignore this setting when B-frames are enabled.

of course you can't decode Techmith on linux/macOS, this was just to show Sirber that in this particular case, there is a codec that can perform at least as well as RV9 ;).

btw how do you encode Xvid @2fps? it seems impossible for me, the file is unreadable once encoded even in Vdubmod.
could you try with
VHQ: 4
chroma motion: on
Q-pel: on
max B-frames: 5
B-frame Quant ratio: 150
B-frame offset: 75
B-frame threshold: 500
chroma optimiser: on
everything else by default.

iirc these settings gave me quite impressive results on LOTR extended edition ending credits. really good quality with ~25 MB for ~25 minutes.
I can't remember the thread in Xvid section where I found them.

celtic_druid
4th September 2003, 09:37
Actually the file is hopefully more like 1.4fps.

VirtualDubMod doesn't seem to like these files either, although at least they playback using directshow, which to answer your question is how they were created.

Anyway, here (http://members.optushome.com.au/cdromguide/new.mkv) is one done with the settings you suggested.

Sirber
4th September 2003, 14:10
Originally posted by APF_Gandalf
since you haven't asked for a particular codec, I tried with something else than Xvid/divx :DYou win :D

[edit]

Is this codec good for movie encoding?

celtic_druid
4th September 2003, 17:50
I doubt it, it would be I would think like saving complex photo's as GIF's instead of say JPG's... on that note I suppose it might be ok for cartoon/anime if the source was clean.

Edit: Ok around 90MB's for 10secs 640x272 25fps compared to around 55 for RGB huffyuv or 30MB's for YV12 VBLE keeping in mind that the movie was originally YV12 anyway.

RathO
12th September 2003, 17:14
Nah, tried that some time ago, could never get it to render properly back then and I still can't.
Hum.. i found it interesting to use speex for like director's commentary... i haven't tried it yet (i will as soon as i get home)

But why they released a DSfilter that dosen't work... :confused:
Any other solution? Alternative?

Regards

darkenergy
8th December 2003, 01:39
got also a problem with the speex DS :-(

MSVCRTD.dll is missing (--> error msg while installation)

is there maybe another possibility?

celtic_druid
8th December 2003, 06:19
That would be because it is a debug build. Not that having a release build would make any difference, well other than the fact that it should actually install.

The solution I would think would be to have an OGG splitter filter which also handles speex as that would appear to be where the problem lies as the splitter simply won't connect to the decoder, 0.9.9.6, 0.9.9.5 and Gabest's.

darkenergy
8th December 2003, 20:15
thx for your answer.

i found the needed dll on the net and was able to install the filter.

then i saw that it's really useless 'cause there comes no splitter with this ds-filter package. and because i'm using the core vorbis filter there was no other vorbis splitter in my system.
but also after installing the latest oggDS it didn't work (i think that was what you tried to explain) graphedit tells me that the oggsplitter can't connect to something if the input is a spx file. :-(


i'm wondring why they release a not working DS-filter.

Atamido
9th December 2003, 07:38
Originally posted by darkenergy
i'm wondring why they release a not working DS-filter. I have a lot of respect for Xiph and what they are trying to do, but sometimes they aren't to bright. I've heard them say a lot of things that make me wonder if they even want wide spread acceptance of their formats. Heck, they never even made a Vorbis DS filter. There was OggDS that was screwed up, and finally Toff spent a few days and made a good one (CoreVorbis).

Sirber
9th December 2003, 13:13
there is an ACM version of Speex, and it's working :)