Hi, folks!

It's my first post, so be gentle!

I recently bought an iAudio M5L and was dismayed to find that it doesn't want to play any OGG files that were encoded with "Floor 0" [i.e. those encoded with libvorbis 1.0 beta 4 or lower, for the technically-minded among you].

The big problem there is that I've been ripping and encoding since the early days and have built up quite a collection of these. Plus "Floor 0" OGGs are still part of the Ogg Vorbis standard, and until it becomes deprecated/obsolete [which seems unlikely] it's a matter of stubborn principle that I purchase and use a digital audio player that fully supports the OGG standards.

I mean, every other decoder I've encountered can play them without problem: Winamp, Foobar, XMMS, even Mortplayer on my little i-mate JAM mobile phone/mini-PDA can play them. Heck, even iAudio's own PC-based JetAudio software can play them! Just not the M5L.

As I see it, I have several options, some more desirable than others:

1. Transcode - Not desirable, and I've read that OGG>OGG transcodes fair particularly badly with all manner of artifacts.

2. Hope that Cowon fix this in a firmware upgrade - Though I'm not particularly optimistic as they seem more keen to focus on their newer models...

3. Somehow convert Floor 0 to Floor 1 - My ideal solution is where I discover that some wonderful people have found a way of converting one to the other without the need to transcode and lose quality. I've not found one so far, and the more I research, the less viable it seems. Worth mentioning though, just in case.

4. Discover a DAP that can handle Floor 0 OGGs - This is where you lovely and wise people come in. If you know of a quality hard-drive based DAP that definitely copes with Floor 0 OGGs, I'd love to hear about it!

Thanks for reading if you made it this far! Any suggestions would be very welcome, believe me

You could try installing Rockbox on it. It should handle Floor0 files. I've at least fixed some of bugs that prevented old files from working.

Rockbox and the M5 don't [or at least aren't going to] mix:



So that's one avenue of possibility barricaded

Thanks, however, Lear!

[EDIT: Though there is a Rockbox port for the X5... if someone with that combination were to confirm that it solves the Floor 0 problem, that'd be verrry interesting...]

Perhaps post a "Floor 0" sample? I guess those old Vorbis betas are not common.

Rockbox uses Tremor, the official integer decoder from xiph.org. So it should support any valid Ogg Vorbis file. If it doesn't that's a bug that should be reported to xiph..

Good point, Alex!

The trouble is I have no way of creating a short sound clip example - all my encoders are newer. I would post a link to a full ogg, but this looks as though it would breach one of the Terms Of Service [understandably].

How about cutting a sample? A recent HA topic: Is there an Ogg Vorbis Editor like mp3 Direct Cut?

Excellent stuff - thanks again!

Okay, here's one:

floor-0-example-libvorbis10beta4_cut.ogg

[EDIT: Example file changed to one where audible quality differences are likely to be more noticeable!]

Floor 0 has been deprecated. The newer libvorbis compiles no longer use it. For those of you who are reading it's levinson-durbin LPC algorithm that was originally meant to be used with low-bitrate stuff. It can't do channel coupling. There was a legitimate reason for using it to begin with when Vorbis was first publically released. That's why most decoders now only now how to decode Floor 1. Floor 1 is the standard to all libvorbis compiles. It's step-wise implementation based upon a linear frequency model. Basically try decoding the files and re-encoding them or else you are out of luck sorry.



If you make small changes to the encoder that are not frozen with the decoder specs, you are going to have some issues. That's why a lot of newer changes can't be made to the encoder, with the exception of the code that already exists in there.

The person in charge of Telechips' firmware told me floor0 was working on his build, but I haven't checked with new firmware.

I don't know if any companies make HDD players based on Telechips' core.

Floor 0 really should not be part of the main branch. It should be removed or moved over to an experimental branch. Through my understanding you can compile Vorbis and make it use Floor 0 for experimental purposes, but other than that it's not worth it.

Sorry, didn't read your post closely enough, apparantly. One char makes a big difference...

Btw, I tried that sample file on Rockbox. It did play, but it sounded horrible, nowhere near the sound from Foobar2000. Must be a bug somewhere (I do get a crash in a simulator build, which should help); I'll see if I can find anything.

Are you even reading the posts you are replying to? A compliant Vorbis decoder is REQUIRED to have floor0 support.




It's an LPC based system. Levinson-Durbin *may* be used to calculate it, but that's just one approach. The spec will certainly not say anything about that. Neither was it meant to be used for low bitrate only.



It can. It's just harder, so it was never implemented.



Yes: nobody had thought of a better alternative, so it was the only thing that was supported.



No, no, no, completely wrong. It's piecewise linear based upon a Bark frequency model

Having the spec in mind: No, it's not possible to do this without slightly altering the signal (you can just approximate a floor 0 curve with a floor 1 curve). Whether it's better or worse compared to transcoding I'm not sure (Also, there's a trade-off between "conversion quality" and bitrate). Anyways, that's theory. Chances are very low someone will try this, I suppose.


Let me quote a part of the specification's introduction to avoid confusion:

It's still part of the spec (thus a "fully Vorbis compatible" decoder has to support it). But as you pointed out, it's not used within the encoder anymore.

Just for your information, I just downloaded the file and uploaded it to my iAudio G3 (firmware 1.55) It does not play it either.

Yes, your right my mistake. I knew it incoporated the bark frequency model into the code somewhere. I was just just citing what I remembered from memory. The only thing I that I remembered when I was reading through the code was an autocorrelation sequence. I had assumed the purpose was to use it for speech, but other then that I am not sure what the intentions were.



Right it should be, but it's not used within the encoder. That's what I am getting. I meant for computational purposes I would think it would be somewhat of a burden?

If you have an autocorrelation sequence, then you can use Levinson-Durbin on it to get an LPC filter.

Makes sense now?

Blimey, this has been a more active thread than I imagined! Thank you, everyone, for your contributions so far

Lear and Hanky - Thank you both for downloading and trying the sample file. Lear, your findings in particular are intriguing.

As SebastianG observed, it must surely be an incomplete decoder that makes a dramatic distinction between two files with one subtle difference [and that difference being perfectly within the standard Ogg specifications].

As an early adopter of an open source standard on reasons of principle as well as technical merit, I am loathe to accept or 'endorse' an incomplete implementation by changing my standard-conforming files or my way of listening to them. Similarly, I am keen to support and promote those manufacturers and developers who employ a complete implementation, which is why I'm disappointed with Cowon as I really thought they'd got one.

But as I step down from this small soapbox allow me to say thanks again for all your input. And I would still like to hear from anyone who can get that audio sample I provided above to work on their DAP. Anyone?

I've tried the sample on a rockboxed iPod and it didn't play correctly. Then I tried it on a Rio Karma where it did play correctly.

A-ha! Excellent news, drezon - cheers!

ok... so there are many different ways you can compute an autocorrelation sequence not just using a Levinson-Durbin approach? interesting.

No.

The autocorrelation sequence is the input to the Levinson-Durbin algorithm to produce the filter's coefficients. You don't compute the autocorrelation sequence via Levinson-Durbin at all.

Anyway, isn't it a stupid idea to calculate the "time domain representation" of a finite-order all-pole IIR filter via Levinson-Durbin even when you do work in the frequency domain all the time and need the filter's frequency response to work with? Why not coding the frequency response directly without approximating the desired response with a finite-order filter first? (That's pretty much the difference between floor0 and floor1)

IIRC my iRiver H120 (the original firmware) also didn't play floor0 files (some of the early Kahvi Collective Vorbis releases are floor0-style). I havn't checked any floor0 files lately (using Rockbox now) -- actually, I really don't care.

Sebastian

I found what caused the bad sound; Rockbox didn't do a full decode of the file, because a buffer was too small (and that buffer was added to avoid stack overflows). But making that buffer larger isn't enough, because the file needs quite a bit of memory to decode, more than what is set aside for codecs in Rockbox at the moment. I'll see if there's anything that can be done, without increasing the codec memory buffer.

Well, it looks as though getting my hands on a new or used-but-working Rio Karma is a no-go

I don't suppose anyone with another DAP that purports to support Ogg Vorbis would be so kind as to try out the example file linked to in my earlier post, and report back as to whether it plays okay? It would be appreciated, believe me

Didn't see this post from last week... anyway I understand what you are saying now. I am a little rusty on the process, bare with me. I am not going to pretend to understand the statistics behind this, just basically how it works.