Feature request related to sound artefacts at very high compression

[fraggy]

Hi there,

First of all: autoGK rulez! It took the pain out of putting one DVD on one CD as DivX. A big two thumbs up for this wonderfull application!!!

I've been using autoGK to compress my DVDs to very small files: 128Mb to 256Mb files, with a resolution of 320x240 or 160x120 to watch them on my handheld. It's a Zaurus SL-5500 Linux handheld with a strongarm 206Mhz processor and it plays those movies FULLSCREEN @ 25 fps without a hitch!!! And the visual quality is plain excellent because of autoGK's 2 pass compression! So now of course I want to use autoGK to compress my whole movie collection, hehe :D

Sound is less important for me, so I tried to compress it to the maximum: 16 kbps mono mp3, and that's where my problems arise!!!!!

At this extreme level of compression, the sound gets a metallic ring to it and other sound artefacts appear (with speakers you don't notice them too much, but with a headphone, it sounds plain awfull!).

Somebody told me that those artefacts can be reduced by the following 3 consecutive steps:

1 - First use an equaliser application to remove high frequencies,
2 - then compress the sound gently (low compression factor),
3 - then compress the sound more agressively

Since autoGK is all about automatically choosing the best settings for a given desired situation, it would be nice if autoGK would take this sound issue in account too and implement those sound processing steps when a very high sound compression is selected.

I'm quite sure you'd make a lot of handheld owners happy with this feature. Our storage media comes in SD (secure digital media), MMC (Multimedia Cards) and CF (CompactFlash) cards of 64, 128, 256 and 512Mb, so you can imagine the necessity of such a feature. A few lucky handheld owners have 1Gb, 2Gb or 4Gb memory cards or Microdrives, but that's a minority. Most of us will need to run our stuff from 256Mb cards... Being able to store 2 (or more) movies on one 256Mb card to watch movies on the road when boredom strikes, that's just great, and that would be even better if 16kbps mono sound would be made (almost) acceptable :)

I got Shrek 1 compressed at 320x190(widescreen)@25fps into a 128Mb file, including the subtitles! Too bad about the audio artifacts though, because the visuals are spectacular :D I'm really looking forward to see those sound artifacts disappear in a future version of autoGK :p

Anyway, keep up the good work. autoGK is a jewel.

[BigDid]

Originally posted by fraggy
...Sound is less important for me, so I tried to compress it to the maximum: 16 kbps mono mp3, and that's where my problems arise!!!!!
Hello fraggy and welcome to the forum :)

You don't tell how you get 16kbp/s mono mp3?
AGK can encode to 32kb mp3 mono directly; have you try it?
If not enough, out of the scope of AGK maybe for audio forum?

Did

[fraggy]

In special options I choose 32kbps and in hidden options I choose force mono.
That automatically results in 16kbps (because 32kbps is for 2 channels and since mono is only one channel, this is automatically half the value :D) and nop, it is absolutely not enough... It's plain horrible: the strong compression results in a horrible metallic sound.

I tried to increase the quality a bit, but the artifacts are too strong. At 192Kbps it's acceptable (96kbps for one channel), but this is increases datasize up to 6 times.)
I don't remember my numbers exactly, but at 16kbps mono mp3, my Shrek had a soundfile of about 15Mb... Let's round it up at 28Mb for simplicity's sake, that would mean:

100 Mb of video + 28Mb of sound, that would be a nice "ratio"...

Now these 100Mb of video are excellent and I'd like to keep it that way. To make the sound acceptable however, I would have the following ratio:

100Mb of video + 90Mb of sound and the sound quality still isn't great. So it's a bit sad that the sound would start to occuppy more space than the video, no?

[fraggy]

Let's just put it this way: if autoGK allows to encode to 32kbps mp3 directly, but it results in horrible sound quality, then this is not really a very usefull option, is it????

It's a bit against the principles of autoGK :P
The beauty of autoGK is that it generate complex batch jobs for you with the best possible parameters!

In order to make 32kbps sound encoding useable, you'd need to include extra processing steps.

For video, autoGK does quality comparison, chooses the best resolution, etc... so the video department does a perfect job. The sound department does a perfect job, just as long as you don't go below the 192kbps stereo mp3 "limit". For lower bitrates, the sound department is letting us down. Which is logical, because autoGK is geared towards filesizes of 1 CD (650Mb) or more, so most people will never ever use the option to encode to 32kbps mono sound.

However, the option is there and it yields unacceptable results... So the logical options (in my point of view) would be eighter to remove that option, or ameliorate the functionality...

Right now, to achieve good results for handhelds, I would have to let autoGK only compress the video... then with another tool I would have to equalize and compress my sound in 2 passes, and then mix the video and the sound file back together in a third application, but eh... then I'm back where I started: in the complex world of virtual dubbing, drowning in settings, jobs and parameters...

So hereby I plead for a feature request, that makes 32 kbps mono sound encoding usuable.

[len0x]

If you have exact workflow including only LAME that is producing nice low bitrate MP3 sound - we'll be happy to hear it.

[Sharktooth]

faac...

[fraggy]

I guess I could experiment a bit with LAME (but only the audio part), and will play around with different settings to compare the quality using different settings and compressing in multiple passes.

Then there is the equalization step itself... I doubt if the LAME encoder supports it by itself... Does LAME support the use of plugins? or otherwise have a method to filter out high frequencies?

I know Virtual Dub supports sound filters, that can be chained up...

Note that I'm not an expert in sound processing / compression at all (and I don't want to be :p), so if anybody want to help me with this, that would be very much appreciated.

I'll keep you informed of my findings.

[fraggy]

Sharktooth doesn't talk too much hey? :D

Thanks, for that tip. I've found some usefull info on FAAC already. I'll include it in my "research".

[BigDid]

Originally posted by fraggy
Sharktooth doesn't talk too much hey? :D ...
;) Off topic, it seems there is a contest going on to give shortest possible meaningful answers...
I already lost and I consider myself happy when I can make myself understand ( in english), I still try from time to time but cannot approach the shortness mastering of sharktooth :D

[Sharktooth]

I need the shortness coz i answare questions while i'm at work.

[stephanV]

just don't do it.

32 kbps mp3 is gonna sound bad anyway you look (uhm... hear) at it. LAME isnt particulary tuned for these sort of bit rates. The suggestions made by your friend (chopping of the high frequencies) are already done because 32 kbps has probably a maximum sampling frequency of 24 or 32 kHz (thus a maximum frequency for sound of ~12-16 kHz). Vorbis and AAC codecs (or even WMA) are generally better and more tuned for low bit rates, but you cant use them with autoGK AFAIK.