OptimFROG DualStream, better than 192kbps MP3? Options

[Feltzkrone]

I played around with OFS at highest compression (--quality 0 --optimize best) with advanced noise shaping enabled (--ans). Compression results in an average bitrate of ~230kbps for some of my black metal albums.

Generally I tend to use OFS instead of MP3 because of those many reasons given on the OptimFROG DualStream page, mainly because these ones:
- no clipping
- no preeche/postecho
- suitable for further processing and transcoding
- no frequency domain artifacts

In fact if someone wants to transcode from OFS to MP3 it's not much worse than using the original WAV file for MP3 encoding, right? This is a very nice thing as OFS could be used for listening to music at home and can be transcoded to MP3 for mobile or car players.

Yes, I know MP3 isn't comparable to OFS as they use totally different techniques to achive lossy compression. When generating a delta-wave file from the original WAV and a decoded OFS I got a flat waveform containing noise at -36db (as shown in Cool Edit) mainly appearing at higher frequency appeareance such as on highhats or cymbal crashs.

So to all you audiophiles: Is OFS "better" than 192kbps MP3s regarding quality, regardless having other advantages like the possibility of near lossless transcoding?

[guruboolez]

There's audible noise with --quality 0. Noise shaping reduces it, but this might introduce some artifacts (IIRC, Den noticed some problems with --ans). I don't know is this noise is audible on metal.

But I wonder: why taking mp3@192 kbps as reference? Why not compare something like mpc or aac to dualstream? Answer is probably easier

[Feltzkrone]

The main reason for taking 192kbps MP3 as reference is because I actually use LAME to encode sound files and I'd like to know if OFS would be a good candidate for switching to other compression than MP3.

I know OGG, AAC etc. already are good reasons to stop working with MP3 files but you know what happens when trying to transcode from OGG/AAC/MPC/WMA... to MP3 for use in some mobile/car MP3 player.

Therefore MP3 as a reference, because it can be used in mobile players unlike OGG/AAC/MPC which can't be played everywhere and not even used for transcoding, at least not as good as OFS.

EDIT: Without ABXing I acutally can't hear any audible artifacts. You know, (black) metal in most cases already sounds "noisy" on original CD.

I've seen more problems on MP3 with metal than pop, techno or common electronical music styles (not speaking about electronic music with many many hard-to-encode effects and filters). Metal MP3s in some cases sound some kind of "empty" or are glitchy at toms and cymbal rides, something I definetly can't find on OFS.

This post has been edited by Feltzkrone: Mar 3 2004, 19:00

[guruboolez]

How do you know that dualstream is a better source than MPC, AAC or Vorbis for transcoding? Den found wavpack lossy to perform very well on transcoding, but it was:

- one or two samples (bluemonday)
- atrac3 as output format.

I don't know if dualstream is 'cleaner' for black album transcoding with lame mp3. I only know that the transcoding is much more slower with dualstream than with mpc (DS decoding speed is very slow).

[Feltzkrone]

I've once got some >200kbps WMA (don't know which version, but some older one) files and they even didn't sound clean as 320kbps-encoded MP3s, i.e. transcoding introduced artefacts which weren't there (or just not audible) in the WMA files.

My opinion is Transform Coder -> Transform Coder always introduces new audible artefacts, regardless which bitrates are used. Basically I think (correct me if I'm wrong) TC1 cuts out some "inaudible" signals and TC2 cuts out "inaudible" signals other to those which TC1 would leave out. As OFS just introduces some additional white noise at -36db I don't think it would matter that much when transcoding to MP3. Please correct me if I'm wrong, it's just what I think and I'm not sure of.

[guruboolez]

Honestly, I've no idea about it. Sorry.

[Feltzkrone]

Just for reference I will transcode some OFS files to MP3, then the original WAV files to MP3 and then open both in Cool Edit Pro and create delta waveforms just to find out how big the "differences" are.

OFS->MP3 (192kbps) already introduced some audible artifacts on cymbal rides. Let's see if WAV->MP3 does better...

[Gecko]

and create delta waveforms just to find out how big the "differences" are.

... and how do you know whether the "delta" is audible?

Only your very own ears (not eyes) can tell you how OFS sounds compared to 192k mp3s. You have to compare them directly.

Please see these two threads:
http://www.hydrogenaudio.org/forums/index....showtopic=18794
http://www.hydrogenaudio.org/forums/index....t=ST&f=1&t=5838

[music_man_mpc]

My opinion is Transform Coder -> Transform Coder always introduces new audible artefacts, regardless which bitrates are used.

Ogg and AAC are transform encoders. Musepack is a subband encoder, and (theoretically) much better for transcoding. OFS should be better, but the only way to tell for sure is ABX testing.

[tigre]

Another problem with OFS transcoding to mp3 can be the noiseshaping. High-pitched sounds (-> shaped noise added by Optimfrog) require many bits, so the encoder you use could run out of bits because of reaching the upper bitrate limit more often. Threads containing more information about this:
Darknoise Copy Protection
High pitched noise = copy protection

[Feltzkrone]

Thanks for your info!

What the eye does see is some delta-waveform at -15db. What my ear does hear is white noise with heavy glitches, which are totally inaudible when comparing WAV->OFS->MP3 to WAV->MP3 but audible when createing a delta-waveform.

So that's the "visual" aspect concerning the difference between OFS->MP3 transcoding and direct MP3 encoding. Funny what our ears can't hear that (loud delta-waveform) when it's mixed with the right second waveform.

EDIT: What I can say so far is that OFS is better than 192kbps MP3 on further signal processing like channel mixing (e.g. inverting one channel and mix it with the other should cut out the mid sound, on MP3 it sounds awful).

This post has been edited by Feltzkrone: Mar 4 2004, 14:20

[2Bdecided]

Another problem with OFS transcoding to mp3 can be the noiseshaping. High-pitched sounds (-> shaped noise added by Optimfrog) require many bits, so the encoder you use could run out of bits because of reaching the upper bitrate limit more often. Threads containing more information about this:
Darknoise Copy Protection
High pitched noise = copy protection

Surely the optimfrog noise shaping is totally different from the noise shaping used by CD mastering? I don't know, because I've only tried wavpack lossy.

"Noise Shaping" in CD mastering means pushing the noise to high (supposedly inaudible) frequencies - and quite moderately so. The peak energy due to this noise shaping will usually be dramatically lower than the peak energy in the actual audio signal.

"Noise Shaping" in near-lossless or hybrid codecs aim to push the "coding noise" into spectral or temporal regions where it is less likely to be heard. It's like psychoacoustic coding, but without a proper model, and without sub-bands or a transform. (Please correct me is optimfrog is doing something different).


The suggestion that noise shaping on CDs can adversely effect psychoacoustic (e.g. mp3) encoding comes from a paper on noise shaping and dither (unfortunately I've lost the reference, but it's been posted here). It's true (for one thing, the VBR bitrate goes up), but it's subtle (it usually sounds fine). The problems with encoding undial.wav are quite different - only stupid levels of noise shaping (never used on CD) are going to drive an encoder into clipping like this!

Cheers,
David.

[tigre]

I've checked dual stream with --quality 0 --optimize best --mode best --ans. With the samples I tested the high frequency boost is very small, not at all comparable to what I've done here, so this will probably not cause any problems with transcoding to lossy formats.

Out of curiosity I've tried to ABX, with success (1st and only sample I tried):
Quizas sample from recent 128kbps AAC listening test.
At ~ 14 seconds there's audible high-pitched noise added (ofs settings as described above), ABXed 7/7 with fb2k ABX tool. After knowing it's there, it's audible for me even under non-ABX conditions (listening to the whole sample), but not annoying at all. Transcoded to lame (3.90.3) APS it's ABX'able as well.

From what I've read at Optimfrog homepage, --quality 0 can't be expected to be transparent anyway.