is ogg vorbis good to encode 96khz/24bit audio? which setting to make no difference with the original (same as for 44.1khz/16bit ?)
thanks
Full Version: encoding 96khz/24bit audio?
nobody has ever done it ?
The purpose of lossy compression schemes like mp3 and vorbis is to discard whatever parts of the audio data make no perceptual difference to the sound and compress what's left with specialized algorithms. Since you can't hear anything above 20kHz the encoder will discard almost all of it. Since there's no practical way to hear anything with a dynamic range larger than 16 bits it's also gone. If you want to encode 96/24 audio you have to use a lossless codec like FLAC or WavePack. Lossy 96/24 is an oxymoron of sorts.
QUOTE(Curtor @ Jun 26 2007, 14:35) 
The purpose of lossy compression schemes like mp3 and vorbis is to discard whatever parts of the audio data make no perceptual difference to the sound and compress what's left with specialized algorithms. Since you can't hear anything above 20kHz the encoder will discard almost all of it. Since there's no practical way to hear anything with a dynamic range larger than 16 bits it's also gone. If you want to encode 96/24 audio you have to use a lossless codec like FLAC or WavePack. Lossy 96/24 is an oxymoron of sorts.
Why is that ? Surely at higher settings like 320~384k it is near lossless. I tried keys sample from pcabx.com and it seems to works good at Q9 and over.
http://64.41.69.21/product/cardd_deluxe/index.htm
QUOTE(shadowking @ Jun 26 2007, 01:15)
Why is that ? Surely at higher settings like 320~384k it is near lossless. I tried keys sample from pcabx.com and it seems to works good at Q9 and over.
320 is nowhere near lossless and has nothing to do with the question. Perceptual encoding discards all of that material... period. That's what it does. The higher bitrate simply enhances the ability to encode the material that is left. It's irrelevant the bitrate when discussing trying to encode 96/24 material into a lossy codec.
It's not so simple Curtor.
I have to preface this by saying that I rarely use Vorbis, but understand the issue regarding lossy codec in general fairly well.
Taking the bit depth first. Lossy codecs don't have a fixed internal bitdepth - the effective bitdepth changes dynamically, frequency band by frequency band, moment by moment - that's how they work. Within a given frequency band, there are often only a few bits of resolution. However, the noise floor can easily be below that of a 16-bit signal. Many encoders can accept 24-bit samples as input. Just like 16-bit samples, most of the accuracy will be thrown away, but there will be moments (typically very quiet moments!) where some of the extra accuracy will be used in some way. Just because it's probably inaudible does not mean that a lossy codec will always destroy it.
As for sample rate: IIRC here vorbis is different from other common lossy codecs, in that it does support higher sample rates. mp3 doesn't go above 48kHz. Whatever the bitrate, any sensible psychoacoustic model is going to dump everything above ~20kHz, but it seems that, for very high bitrates, vorbis does not implement a fixed low pass filter. I've never checked to see if the psychoacoustics allow anything above 20kHz through at any point. I'd expect the absolute threshold of hearing calculation to cause it all to be dumped, but you never know.
What bitrate is required for transparency @ 24/96?
16-bit vs 24-bit makes no difference. If anything, a 24-bit source can in theory be slightly easier to encode because there's less quantisation noise to confuse the psy model, though in practice it rarely makes a difference (noise shaped 16-bit dither is an exception - you can often see codecs wasting bits on this, which won't happen with original 24-bit sources).
44.1kHz vs 96kHz - I don't know. Depending on design, the codec could be less efficient at 96kHz than at 48kHz. Most Hydrogen Audio members would argue that downsampling 96kHz content to 48kHz is a transparent process, so encoding 24-bit 48kHz at a similar bitrates to normal 16-bit 44.1kHz CD content should produce similar results. Encoding it at 96kHz probably needs a higher bitrate for no audible improvement in quality, or may work equally way at exactly the same bitrate.
Apparently CaptainTrips, you may be the first to try it, though I don't think enough vorbis experts have read this thread to be sure of that yet.
For compatibility, I think downsampling to 48kHz (or 44.1kHz) and encoding with your normal settings is a good option.
Cheers,
David.
I have to preface this by saying that I rarely use Vorbis, but understand the issue regarding lossy codec in general fairly well.
Taking the bit depth first. Lossy codecs don't have a fixed internal bitdepth - the effective bitdepth changes dynamically, frequency band by frequency band, moment by moment - that's how they work. Within a given frequency band, there are often only a few bits of resolution. However, the noise floor can easily be below that of a 16-bit signal. Many encoders can accept 24-bit samples as input. Just like 16-bit samples, most of the accuracy will be thrown away, but there will be moments (typically very quiet moments!) where some of the extra accuracy will be used in some way. Just because it's probably inaudible does not mean that a lossy codec will always destroy it.
As for sample rate: IIRC here vorbis is different from other common lossy codecs, in that it does support higher sample rates. mp3 doesn't go above 48kHz. Whatever the bitrate, any sensible psychoacoustic model is going to dump everything above ~20kHz, but it seems that, for very high bitrates, vorbis does not implement a fixed low pass filter. I've never checked to see if the psychoacoustics allow anything above 20kHz through at any point. I'd expect the absolute threshold of hearing calculation to cause it all to be dumped, but you never know.
What bitrate is required for transparency @ 24/96?
16-bit vs 24-bit makes no difference. If anything, a 24-bit source can in theory be slightly easier to encode because there's less quantisation noise to confuse the psy model, though in practice it rarely makes a difference (noise shaped 16-bit dither is an exception - you can often see codecs wasting bits on this, which won't happen with original 24-bit sources).
44.1kHz vs 96kHz - I don't know. Depending on design, the codec could be less efficient at 96kHz than at 48kHz. Most Hydrogen Audio members would argue that downsampling 96kHz content to 48kHz is a transparent process, so encoding 24-bit 48kHz at a similar bitrates to normal 16-bit 44.1kHz CD content should produce similar results. Encoding it at 96kHz probably needs a higher bitrate for no audible improvement in quality, or may work equally way at exactly the same bitrate.
Apparently CaptainTrips, you may be the first to try it, though I don't think enough vorbis experts have read this thread to be sure of that yet.
For compatibility, I think downsampling to 48kHz (or 44.1kHz) and encoding with your normal settings is a good option.
Cheers,
David.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.