OGG playback quality

[Sykes]

TL;DR: OGG -q 10 (~ 500 kbps) sounds noticably worse than FLAC. Should it?

Long version: I know OGG is a lossy format etc etc, still, I wouldn't expect to hear a difference between OGG and FLAC. Especially since it's supposed to be way better than mp3.

I used to convert my FLAC/WAVs to ogg -q8 (~ 256 kbps). I've always realized they sound a little weird, but I've never done any 1:1 comparison. Weeell now I did. The best example is my 96/24, and binarual to boot, FLAC album (btw: it's Ottmar Liebert - Up Close; highly recommended, it's on hdtracks). Of course the Clip+ can't handle 24b, but it can handle 96 kHz all right. So I converted the files to A) 96 kHz/16b FLAC, B) 48 kHz/16b OGG. And there's definitely a difference - the OGG sounds harsh and the space is flat. Then I loaded the player with more FLACs instead of the original OGGs and now everything sounds way much better.

I wouldn't expect to hear such a difference on a small player (Clip+) and much less on Porta Pro headphones, which I've never considered to be very good.

I wonder if it's supposed to be like that, or the Clip+ isn't that good at playing OGG files? For the record, I'm not a fan of 320 kbps LAME mp3s either.

(BTW any idea how can I convert 96 to 48 kHz FLACs? No luck finding this option in Foobar2k.)

[CrowdAds]

[FreeZ5]

I'm no expert on OGG/Flac, but I believe that I've read that the Original Firmware's OGG decoder is not up to snuff. Have you tried Rockboxing your Clip+?

Steve

[Sykes]

Nope, it's about to leave for RMA on Monday actually, so maybe when it comes back.

Might Rockbox improve the quality of MP3 playback as well?

[saratoga]

Originally Posted by Sykes

Of course the Clip+ can't handle 24b, but it can handle 96 kHz all right. So I converted the files to A) 96 kHz/16b FLAC, B) 48 kHz/16b OGG. And there's definitely a difference - the OGG sounds harsh and the space is flat. Then I loaded the player with more FLACs instead of the original OGGs and now everything sounds way much better.

Might be related to resampling. Probably a good idea to resample everything on the PC using foobar rather then trust that the player is doing it properly.

Have you checked to see if the difference persists using foobar's ABX plugin? To be safe, I would convert the flac to 48k first, then encode it to Vorbis. Finally compare the two versions and see how they stack up.

Originally Posted by Sykes

(BTW any idea how can I convert 96 to 48 kHz FLACs? No luck finding this option in Foobar2k.)

Yes, under DSP there should be one called "resampler".

[Sykes]

Originally Posted by saratoga

Have you checked to see if the difference persists using foobar's ABX plugin? To be safe, I would convert the flac to 48k first, then encode it to Vorbis. Finally compare the two versions and see how they stack up.

Didn't try that, my PC audio is crap (onboard sound card, cheap 2.1 speakers). Now that I've ripped some CDs to FLAC instead of OGG, I think they sound better on the PC as well, but I could be imagining things. I've already deleted the OGGs, but I'll try the ABX thing later.

Anyway, could resampling 96 -> 48 kHz cause some noticeable loss in quality? 96 -> 44 I'd understand, but 96:48 is 2:1, and in theory shouldn't be obvious on lower than the highest-end systems, or am I wrong?

Originally Posted by saratoga

Yes, under DSP there should be one called "resampler".

Ah thanks, I'll try that.

[saratoga]

Originally Posted by Sykes

Anyway, could resampling 96 -> 48 kHz cause some noticeable loss in quality? 96 -> 44 I'd understand, but 96:48 is 2:1, and in theory shouldn't be obvious on lower than the highest-end systems, or am I wrong?

I would expect simple devices like portable mp3 players to do equally badly in both those cases.

[Sykes]

That's what I thought, but now I've tried something else. I've converted the 24b/96k FLAC to 16/96 FLAC, 16/48 FLAC and 16/48 OGG with -q 10 quality. I loaded the Clip+ with them and here's the funny thing.

What I thought is the difference between FLAC and OGG, actually seems to be between 96k and 48k. 16/48 FLAC and OGG sound exactly the same, while 16/96 FLAC sounds better. There's a bit more airiness and the sound is more relaxing overall.

So I guess OGG is fine after all, just needs to be used with -q 10 instead of -q 8.

But Clip+ can harvest some extra quality out of 96k apparently, for some reason.

For the record, I did ABX testing on my PC (crappy audio) with Port Pros and I can't hear any difference between either of those. 320k mp3 is easy to spot however (3 out of 3 right).

Anyone wants to try it out?

[saratoga]

Originally Posted by Sykes

But Clip+ can harvest some extra quality out of 96k apparently, for some reason.

Its probably just the crappy resampling used by the sandisk software for 96k tracks. People usually decide that the bigger numbered file sounds better, even if its just some extra distortion.

[Sykes]

Possibly, but I think the 96k tracks sounds with less distortion (48k sounds more 'harsh', like when the person playing the guitar isn't so well trained) and also there's a bit more space/airiness/whatever.

Note that this is a binaural recording and my only 96k album, so YMMV. (And I've only checked one track actually.)

Maybe the resampling is crappy either way and 96k just provides some more data to work with. Like when working with digital pictures - you may only need 1000*1000 px, but it's always better to start from a higher resolution and downsample later. Or maybe the player always downsamples to something like 44k and 96k is better for that. Than again, I think I read somewhere that the Clip+ can output frequencies way above 20k Hz (I think something like 40k Hz or so), so maybe it can use the 96k after all.

[TackyTiger]

Originally Posted by Sykes

Like when working with digital pictures - you may only need 1000*1000 px, but it's always better to start from a higher resolution and downsample later.

This analogy is better suited to the bit depth rather than the sample rate.

Sampling rates way above the range of human hearing would be analogous to a digital picture with a gamut far into the ultraviolet.

[Sykes]

Not really. First, gamut isn't related to bit depth at all. Unlike in sound, where digital sampling causes the frequency range to be related to the sampling rate.

Actually, resolution in digital imagery can be described as 'frequency', as in - a frequency too high (the tiniest details) cannot be recognized by the eye. Just like with sound - a frequency too high is lost to the ear.

Bit depth in digital imagery is more related to bit resolution in digital sound.

But that's just digital. In the analogue world, of course the visible gamut is related to frequency range (as in, both are dependent on wave lengths), but that relation doesn't translate to 2D digital imagery and digital sound. To find a closer relation, we'd have to compare the ear to one photosensitive cell in the eye, not the eye as a whole.

[saratoga]

Originally Posted by Sykes

Possibly, but I think the 96k tracks sounds with less distortion (48k sounds more 'harsh', like when the person playing the guitar isn't so well trained) and also there's a bit more space/airiness/whatever.

Note that this is a binaural recording and my only 96k album, so YMMV. (And I've only checked one track actually.)

I was curious so I just tested, and the Sandisk firmware doesn't actually resample. A 30kHz tone encoded to 96khz flac plays back at 30khz as expected. My guess is that the difference is here is the resampling used to generate the 48k test file.

Originally Posted by Sykes

Maybe the resampling is crappy either way and 96k just provides some more data to work with.

To be clear, 96k is completely uesless. It doesn't provide more data to work with because theres nothing of value gained in going from 48k to 96k. So the most likely explanation is that something has gone wrong in the preparation of the samples, or some other experimental error.

[Sykes]

Originally Posted by saratoga

To be clear, 96k is completely useless. It doesn't provide more data to work with because theres nothing of value gained in going from 48k to 96k. So the most likely explanation is that something has gone wrong in the preparation of the samples, or some other experimental error.

I've converted them using foobar2k, just straight conversion. Both 16/48 and 16/96 come straight from 24/96. Any better way to do it? I can't really imagine how a 96 -> 48 conversion could go wrong.

And while I know that in theory, there shouldn't be audible difference... Who knows. I'd really like someone else to try it out.

[saratoga]

Originally Posted by Sykes

I've converted them using foobar2k, just straight conversion. Both 16/48 and 16/96 come straight from 24/96. Any better way to do it? I can't really imagine how a 96 -> 48 conversion could go wrong.

And while I know that in theory, there shouldn't be audible difference... Who knows. I'd really like someone else to try it out.

Is this actually a blind comparison?

[Sykes]

No, but the difference is quite obvious. Surprisingly so, actually.