High-res audio comparison: Linn Records Free High Res Samples

[Tks]

Classial signals and systems. Run of the mill DSP + ADC/DAC functionality.

Here we go with the bait and switch.

No one is talking about oversampling for modern DACs and ADCs to work.

I'll try again. I am simply upsampling an already existing audio file. I don't care about oversampling DACs/ADCs do internally for the sake of spreading out quantization noise to reduce the effects, or reduce jitter in the analogue reconstruction for example due to poor clocks or whatnot. I also wanted to mention concerning your bit-depth claim, I don't care about increasing bit-depth for the sake of easier DSP application. None of these are of concern, and are tangential to my main question. (Though it would be interesting in hearing how increasing the bit-depth of a file and nothing else, would improve sound.. If you could prove this, I'll be impressed enough to simply accept your any claims about upsampling without contest). EDIT extra sentence: I really wish Amir didn't give away why the bit-depth increase comment was nonsensical. The reason I was so willing to accept any claims redshift makes if he could prove the bit-depth increasing senitments he holds, is because I know there's no way he, nor anyone else could satisfy such a proposition.

I want to know if I am simply upsampling an already existing audio track. What am I getting now that I have done this. I'm not going to be doing anything else to the file, there won't be any filtering that messes with sound to the the degree passband will exhibit different frequency response or whatnot (though starting with a 8kHz recording, this would be the entire original energy present within the recording itself).

Again, once more. I am only upsampling a file. What will be changing audibly here? Don't tell me it will "sound better", that isn't informative in the slightest; explain mechanistically.

 

[Blumlein 88]

Classial signals and systems. Run of the mill DSP + ADC/DAC functionality.

There were DAW plug-ins that worked better at 96 khz and it was worthwhile to upsample for that in days past. I don't know the state of such things today. I know many DSP processes in DAW software will internally upsample, do the processing and downsample the result back to whatever file you are working on. Basically I'd think most DSP takes care of this quite well and you upsampling or distributing files at 192 khz that originated at lower sample rates probably gains you almost exactly nothing.

If you have some information to the contrary, then tell us what it is or tell us how to do some testing of both conditions to uncover the difference.

 

[redshift]

That trades off sampling rate for bit depth. It gives no new information as there is none to give you.

If this were not so we would make all digital cameras 1 megapixel and then split out 100 megapixel images out of them. I hope you know that is an impossibility.

Pan and tilt (scan) your 1 googol hertz sampled RGB pixel with 1 bit resolution (for each color) and it is only a matter of processing speed to create an image of basically arbitrary bit depth and size.

 

[redshift]

There were DAW plug-ins that worked better at 96 khz and it was worthwhile to upsample for that in days past. I don't know the state of such things today. I know many DSP processes in DAW software will internally upsample, do the processing and downsample the result back to whatever file you are working on. Basically I'd think most DSP takes care of this quite well and you upsampling or distributing files at 192 khz that originated at lower sample rates probably gains you almost exactly nothing.

If you have some information to the contrary, then tell us what it is or tell us how to do some testing of both conditions to uncover the difference.

Well, my point being that it is better to record at, for example, 24/32bits 192k and distribute the record that way to avoid signal processing (DSP/EQ) artifacts (and processing overhead) from truncation and oversampling/up sampling/interpolation in the playback system.

The proper way of reducing the amount of data is by lossy/lossless compression and not that of truncation before distribution to the consumers.

However, I’m not suggesting that we should have a higher dynamic range in the record than what state of the art DAC’s and amps can resolve. I guess we are at about 130dB SnR now. That implies ~24 bits. Say 32 to make it a single precision float.

Now; if our ears, brains, speakers and rooms didn’t suck.

 

[Newman]

How does using DSP require "up sampling to maintain resolution"?

I have a side question: why does DSP do upsampling of CD quality music? Does it achieve anything?

 

[danadam]

What if I use 8kHz sampling rate for my audio? Would playing it back resampled 192kHz allow me to hear "more accurate detail"?

Yeah, you are effectively oversampling the signal. That way you could convert it to a higher res signal, perhaps go from a 16bit 8kHz to a 20bit 44.1kHz signal.

There is 8bit 8kHz sample in the attachment. Can you show us how to do this magic conversion that adds accuracy and detail?

 

[redshift]

There is 8bit 8kHz sample in the attachment. Can you show us how to do this magic conversion that adds accuracy and detail?

Wasn’t it a 192kHz 8bit track?

 

[redshift]

I have a side question: why does DSP do upsampling of CD quality music? Does it achieve anything?

It depends if you want to do some post processing on it. For example in a DSP or EQ. Every bit of binary arithmetic will cause some truncation. For sure there are some fancy ways of handling division remainders. But I doubt a run of the mill DSP handles that? It just divides and throws away the remainder.

For example 255/9 = 28.3333, assuming 8 bit representation would become just 28. That’s truncation for you. The more taps you got in your filters/effects, it becomes increasingly important to have hi res/up sampled data.

 

[dc655321]

I have a side question: why does DSP do upsampling of CD quality music? Does it achieve anything?

Upsampling cannot add new information (ignoring distortion).

However, resampling may be done in a DSP engine in order to make source sample rate match that of the filters provided by the engine.

 

[Blumlein 88]

It depends if you want to do some post processing on it. For example in a DSP or EQ. Every bit of binary arithmetic will cause some truncation. For sure there are some fancy ways of handling division remainders. But I doubt a run of the mill DSP handles that? It just divides and throws away the remainder.

For example 255/9 = 28.3333, assuming 8 bit representation would become just 28. That’s truncation for you. The more taps you got in your filters/effects, it becomes increasingly important to have hi res/up sampled data.

Most any modern DSP will run at least 32 bit float, most run more than that.

 

[redshift]

Most any modern DSP will run at least 32 bit float, most run more than that.

Yeah, it makes sense if you got some sharp filters with plenty of taps. Truncation is bad.

 

[danadam]

Wasn’t it a 192kHz 8bit track?

Unless I lost my reading comprehension, no. Tks specifically said about taking 8kHz source and resampling it to 192kHz.

Then for some reason you also added to that increasing of bit depth. That's why the sample is 8 bit. I am curious how the magic conversion will improve that.

 

[Newman]

So, what is the view on the notion that CD-quality audio will be diminished by DSP in the playback chain, to the point that audible issues might occur?

What about high-bitrate AAC/MP3?

And if so, do we come to the conclusion that something higher than CD quality is necessary for playback chains that involve DSP, if we want to avoid audible issues?

 

[redshift]

So, what is the view on the notion that CD-quality audio will be diminished by DSP in the playback chain, to the point that audible issues might occur?

What about high-bitrate AAC/MP3?

And if so, do we come to the conclusion that something higher than CD quality is necessary for playback chains that involve DSP, if we want to avoid audible issues?

Upsampling sort of already “solves” that.

But why first flatly truncate data to 16bit 44.1k and then upsample it to say 32bit 192k again before we shove it into our processing and amplification pipelines?

That being said; state of the art DAC’s and amps resolves better and processes faster than 16bit 44.1k. It’s time to put that format in the dustbin of history because we don’t listen to music that way anymore.

 

[Blumlein 88]

Upsampling sort of already “solves” that.

But why first flatly truncate data to 16bit 44.1k and then upsample it to say 32bit 192k again before we shove it into our processing and amplification pipelines?

That being said; state of the art DAC’s and amps resolves better and processes faster than 16bit 44.1k. It’s time to put that format in the dustbin of history because we don’t listen to music that way anymore.

But why kill the format? With computers and such it can keep serving that and newer ones transparently to us.

 

[dc655321]

Upsampling sort of already “solves” that.

What is upsampling solving here and how?
You've repeatedly asserted this with zero evidence.

 

[redshift]

What is upsampling solving here and how?
You've repeatedly asserted this with zero evidence.

It ”solves” the problem of truncation in the DSP. It is better to process hi res data than lo res data. At least if the signal integrity is of importance, which I assumes it is?

 

[dc655321]

It ”solves” the problem of truncation in the DSP. It is better to process hi res data than lo res data. At least if the signal integrity is of importance, which I assumes it is?

Upsampling is converting a digital signal to a higher sampling rate, yes?
This adds no new information and no increase in resolution.

A digital signal could also have its bit-depth changed, say from 16 to 24 bits.
That process could push the noise floor down.

Truncation occurs in the LSB, which leads to quantization distortion.
At what dBFS do you think truncation of the LSB manifests?

 

[Newman]

@dc655321 if upsampling is the wrong word, then what is the right, commonly used word for increasing bit depth?

Just wondering what common word you think he should have used.

 

[dc655321]

@dc655321 if upsampling is the wrong word, then what is the right, commonly used word for increasing bit depth?

Just wondering what common word you think he should have used.

I dunno - increasing bit depth or bit rate (latter is tied to sampling rate too) seem common enough.

I don't really care about the vocabulary, just trying to make sure we're discussing the same idea(s).
It's the misinformation that bothers me...