24 to 16 bit conversion theory

[Bob Loblaw]

Hello guys,

 

Using dbpoweramp to convert 24/96 flacs to 320 lame mp3, i wonder about what actually happens to bit depth, as in wondering if the extra 8 bits are cut off or if the entire "picture" is shrunken to 16/48.

 

Downloaded both versions of a Pearl jam bootleg (Atlanta 2012) in both 16/44 and 24/96 from their website to compare. Using Audacity, there was some noticeable difference between the 2 , where the 16/44 was indeed very bad and the 24 bit version was not that great, but at least it was not brick-walled.

 

Would you guys buy 2 versions for high-end playback and portability or just stick with the better mix and convert from there?

[tranz]

Why bother with mp3?

[Bob Loblaw]

Portability (ipad, phone)

[audiventory]

Hi Bob,

 

Depennd on:

 

1. Who use better conversion software: you or store?

 

2. Do you listen on used apparatus difference between home and store's conversions? (more attention here for quiet places of classical music and jazz)

 

If you trunk bit depth also need consider used dithering algorithm.

 

Best regards,

Yuri

[tranz]

Portability (ipad, phone)

 

With the amount of space on those I would keep it uncompressed.

 

Regarding dropping bits, it happens every day when a CD is mastered since most people record in 24bits. So there are dithering programs used to essentially create a white noise floor to mask any artifacts from dropping the bits. Not sure if you can hear that on mp3 on iphone/pad though so it is not something to worry about. Trust your own ears though and if you like it better go for it.

[tmtomh]

Unless something has changed, Apple iDevices can play 24/48 files but nothing higher (no 24/96 or 24/192 for example).

 

So for the best sonics, I would recommend just downsampling to 48kHz and keeping the bit depth at 24.

 

But if you're also looking to save storage space, then yes, high-bitrate mp3 is a viable solution.

 

LAME, considered the best mp3 encoding algorithm, reportedly can accept 24-bit files for encoding. Whether LAME spits out a 24-bit or 16-bit mp3 file as a result is unknown - and probably unknowable, since the variable compression of mp3 and other lossy formats means they don't really have constant bitrates like lossless formats do.

 

For this same reason, I would think dithering would be irrelevant too. (I've read stuff that says you should first dither the 24-bit file to 16-bit, and then convert to mp3 - but since the mp3 conversion is reducing the bit-depth again, I don't see the point of the initial 24-to-16 step.)

 

Finally, I would never purchase two versions of the same digital files (24 bit and 16 bit). I would always do the converting myself, because today's consumer tools usually are of good enough quality to produce excellent results, especially for portable listening.

[Bob Loblaw]

Thank you all for the great info, specially for the 24 bit mp3 suggestion... For some reason just never crossed my mind it would be possible.

 

My workstation/ music server has enough space to pretty much handle anything i need, but i will start getting my music soon to my flash based portables (ipad, phone, laptop) in which obviously storage space might be an issue. The less quality i lose the better, even if i just use a microstreammer with the Grado SR80s.

[Bob Loblaw]

Edit

[audiventory]

 

LAME, considered the best mp3 encoding algorithm, reportedly can accept 24-bit files for encoding. Whether LAME spits out a 24-bit or 16-bit mp3 file as a result is unknown - and probably unknowable, since the variable compression of mp3 and other lossy formats means they don't really have constant bitrates like lossless formats do.

 

For this same reason, I would think dithering would be irrelevant too. (I've read stuff that says you should first dither the 24-bit file to 16-bit, and then convert to mp3 - but since the mp3 conversion is reducing the bit-depth again, I don't see the point of the initial 24-to-16 step.)

 

Hi Tmtomh,

 

As I know mp3 format don't reduce bit depth as itself. If I don't mistaken, applied spectral and dynamical range truncating.

 

Howerer how it done I don't learn. Here need more information for estimate how impact dither to non-linear distortions of [encoded to mp3] and [decoded to wav] signal.

 

Here possibly check it via experiment.

 

Best regards,

Yuri

[Bill Lord]

Hi Tmtomh,

 

As I know mp3 format don't reduce bit depth as itself. If I don't mistaken, applied spectral and dynamical range truncating.

 

Howerer how it done I don't learn. Here need more information for estimate how impact dither to non-linear distortions of [encoded to mp3] and [decoded to wav] signal.

 

Here possibly check it via experiment.

 

Best regards,

Yuri

 

Yuri - You are right. Spectral and dynamical range truncating are a part of lossless compression.

 

 

Truncating 24bit to 16bit, with a proper dither, will result in a perfectly acceptable audio file.

 

Here is a video worth discussion, basically the video states the difference between 24bit and 16bit is in-audible noise.

 

[Bob Loblaw]

What software does this guy use to convert?

[tmtomh]

Thanks Yuri and others for your responses. Part of the problem is that I (and I suspect others) don't know the proper terminology to use when describing what mp3 compression does.

 

But when Bill Lord writes that "spectral and dynamic range truncation" are part of "lossless compression," that's not quite right. Lossless compression will preserve the dynamic range and spectrum - only resampling and reducing bit depth will impact spectrum and dynamic range.

 

Resampling truncates the frequency spectrum by reducing the ultrasonic cutoff - so a 96kHz file can contain frequencies up to 48kHz, but a 48kHz file can contain frequencies only up to 24kHz.

 

Dynamic range truncation refers to the fact that when you reduce the bit-depth, you reduce the theoretical signal-to-noise ratio. So 24-bit audio has a theoretical noise floor of about -144dB, while for 16-bit that noise floor goes up to -96dB. The reality is more complex, but that's the main principle. And dither is random noise added to the signal to (A) make the digital noise more random, and sometimes also (B) to "shape" the digital noise so that the noise floor is lower in the midrange where our ears are most sensitive to it, and higher in the upper frequencies where we can't hear it as well.

 

So lossless compression has nothing to do with any of that.

 

As for mp3, it certainly doesn't downsample - a lossless 48kHz file will become a 48kHz mp3. But the bitrate - bits per second - will certainly decrease because of the lossy compression. So IMHO one can say that mp3s have a variable bit-depth, and therefore a lower average bit-depth than the lossless original.

[audiventory]

Here is a video worth discussion, basically the video states the difference between 24bit and 16bit is in-audible noise.

 

Hi Bill,

 

Thank you for link. Interesting method of testing.

 

I bad understand English on hearing, however as I understand there noise was about -60 dB, isn’t it?

 

 

 

 

What software does this guy use to convert?

 

Hi Bob,

 

It look like to Reaper’s (used in video audio editor) builtin converter.

 

 

 

 

As for mp3, it certainly doesn't downsample - a lossless 48kHz file will become a 48kHz mp3. But the bitrate - bits per second - will certainly decrease because of the lossy compression. So IMHO one can say that mp3s have a variable bit-depth, and therefore a lower average bit-depth than the lossless original.

 

Hi Tmtomh,

 

If mp3 coder/decoder inside work in integer: for this case possibly have place bit depth truncation.

 

If it work with float point math inside: here impossibly talk about bit depth truncation. Here only multiply and dividing.

 

Dynamic range truncation refers to the fact that when you reduce the bit-depth, you reduce the theoretical signal-to-noise ratio. So 24-bit audio has a theoretical noise floor of about -144dB, while for 16-bit that noise floor goes up to -96dB. The reality is more complex, but that's the main principle. And dither is random noise added to the signal to (A) make the digital noise more random, and sometimes also (B) to "shape" the digital noise so that the noise floor is lower in the midrange where our ears are most sensitive to it, and higher in the upper frequencies where we can't hear it as well.

 

If be exact

 

24 dB has -146…-150 dB noise floor

 

16 bit has -120 … -130 dB

 

Dithered signal (for careful algorithms) about -90 … -110 dB

 

Such digits appear due random nature of noise and applying window/accumulation during spectrum drawing.

 

Best regards,

Yuri

[Bill Lord]

Thanks Yuri and others for your responses. Part of the problem is that I (and I suspect others) don't know the proper terminology to use when describing what mp3 compression does.

 

But when Bill Lord writes that "spectral and dynamic range truncation" are part of "lossless compression," that's not quite right. Lossless compression will preserve the dynamic range and spectrum - only resampling and reducing bit depth will impact spectrum and dynamic range.

 

Resampling truncates the frequency spectrum by reducing the ultrasonic cutoff - so a 96kHz file can contain frequencies up to 48kHz, but a 48kHz file can contain frequencies only up to 24kHz.

 

Dynamic range truncation refers to the fact that when you reduce the bit-depth, you reduce the theoretical signal-to-noise ratio. So 24-bit audio has a theoretical noise floor of about -144dB, while for 16-bit that noise floor goes up to -96dB. The reality is more complex, but that's the main principle. And dither is random noise added to the signal to (A) make the digital noise more random, and sometimes also (B) to "shape" the digital noise so that the noise floor is lower in the midrange where our ears are most sensitive to it, and higher in the upper frequencies where we can't hear it as well.

 

So lossless compression has nothing to do with any of that.

 

As for mp3, it certainly doesn't downsample - a lossless 48kHz file will become a 48kHz mp3. But the bitrate - bits per second - will certainly decrease because of the lossy compression. So IMHO one can say that mp3s have a variable bit-depth, and therefore a lower average bit-depth than the lossless original.

 

doh, You're right.. It's supposed to be lossy compression, I made a typo.

[tmtomh]

No problem - after I made my post I thought, "Oh no, what if he just made a typo?" :-)

[Bill Lord]

tmtomh- :-)

 

Also I found this white paper interesting.

 

http://lavryengineering.com/pdfs/lavry-sampling-theory.pdf