Is there any way to encode VBR 1 pass at 16 bit depth? I just feel that i am waisting space if i chose 24bits, why did they not include 16 bit mode?

You can't choose bitdepth because lossy files don't have a bitdepth. They just call it 24bit because it looks cool.

So why on VBR 2-pass and CBR i can use 16 bit?

I think it may be decoder: 24 bit setting may allow decoder to output at 24 bits. But, yes, lossy formats cannot have bitdepth; this means that you don't waste your space by choosing 24 bits.

Well if it doesnt matter, why does LAME MP3 and WMA both let you chose? Wouldnt they just remove the option if it didnt matter?

Which LAME switch allows you to set the bitdepth?

--bitwidth w / input bit width is w (default 16)

http://www.hydrogenaudio.org/forums/index....showtopic=47759

All the settings for foobar2000 WMA Conversion.

Lossy files have no bit depth, so whatever you're reading is either wrong or being misunderstood.

Although I (literally) have no idea how it would relate to the OP, I can easily imagine using knowledge of the input PCM resolution to set things like minimum quantization thresholds and the like, in an encoder.

But I have no information on the OP, sorry. I'll poke somebody who might, but don't count on it.

Do you guys know what I'm talking about? Ill show you an example:

If i chooe CBR i can chose 64kbps, 44Hz and 16bits, 2 channel

But i change it to Quality VBR i have to chose something like:

Quality 25, 44Hz, 24bit, 2 channel

See how the 16 bit changes to 24?







Oh and im talking about bits per sample and not bitdepth

It's the same.

Most of the people here at HA have little use for WMA, so you may not get a satisfactory answer.

Just take a WAVE editor like Audacity, zoom in so you can see the separate samples. The bit depth tells you, how many values can a sample have (2^16 = 65536 for 16bit). Does it make sense? O.K., then almost all lossy formats use DCT conversion, which will transform a wave to sum of trigonometrical functions. So during DCT you're loosing any bit depth, the data are stored as sinus waves.

You can (or must) define bitdepth of input data, but the actual compressed file doesn't have any bit depth. So if WMA is using DCT (and I suppose it is), it can't have any bit depth. It is some kind of misinformation.

New Topic Title : Let's talk about bit depth

Sure, but theres scale factors and all that, so while the average sample might only get 4-6 bits or similar, you can't actually say what the bit depth of any individual sample will be. IMO its less misleading to say theres no actual bit depth.

I think you miss the point. If I know that the input is 16 bits, there's no point in having scale factors or quantizers that describe something 18 bits down, now, is there?

I'm saying that knowing the original resolution can easily lead to coding efficiency, which is, I'd think, enough incentive to ask for it.

There is no way on the planet earth to assign any "bit depth" to a perceptual coder's output that means much of anything useful.

I'm not a coding expert, I just know some mathematics. IMO once the input data are converted using DCT to frequency domain, there's absolutely no use for the bit depth information. Even a single sinus wave has infinite bit depth. That's the important point - bit depth of WMA, MP3, AAC, Ogg Vorbis etc. is infinite.

It can be that the reason why it has the 16bits option in CBR is being it a legacy option.
I doubt there is any reasonable fact to require 24bits of input for it to work, i just assume that they worked with 24bit data in order to improve the encoder, and *maybe* a marketing campaing around the lines of "See! we encode at 24bits!"

Edit: Remember that the codec was developed with the VC-1 video codec in mind, not as an audio CD encoder. So assuming the masters are on 24bits, it could seem a direct option.

This is to specify bit depth of input data when input is raw data (ie just in order to be able to correctly interpret the input data).






But on the other hand, we probably all have some adaptative noise floor in our encoders (don't we?), so knowledge of the input bit depth is not that interesting anymore.

So you'd let your system adapt below the actual noise floor? Why?




Err. No. It does not. I fyou start with a 2 bit sine wave, you'll see the quantization noise in one form or another (let's hope you dithered), not "infinite bit depth". When you quantize to some bit depth, you irrevocably add noise. (not necessary the same power at all frequences, there is indeed noise shaping, but you can't eliminate noise)

When you quantize to any fixed (integer) bit depth, you add noise. You don't ever get rid of it. Them's the roolz. Ask Dr. Shannon.

Most of the time, the instantaneous dynamic range handled by the encoder is quite lower than the one of the initial data. (usually a human can not make use of a +90dB dynamic range instantaneously)
But I think that you've got a point there, as in long silent parts, we (LAME) might reduce our noise floor quite below the noise floor of the original data.

Boy, do I have a synthetic signal or two for you

Seriously, look at things like chimes, glockenspiels, triangles...

Since the WMA Pro decoder supports output of 24 bit/sample, there must be an option to encode with 24 bit input samples...as easy as that. And since the platform usually is a PC, there is also no issue with processor wordlength restrictions (floating point anyway). So assumed that all the internal processing is adapted to support the 24 bit resolution (like the pyschoacoustic model etc, which I don't know of course), it would well make sense.
Which doesn't mean that it will "sound" better then... marketing sure sounds like a good reason to claim support for 24 bit

Dunno, I've been using it lately for the lossless codec and it seems to work very well for me. I haven't messed with FLAC or any of that I'm just too lazy... Lossless comes with Windows haha.