16 bit files almost unlistenable now...

[mansr]

10 minutes ago, adamdea said:

It's a risky business trying to dump quantisation noise in such a small, potentially audible band. Especially when the only point in it is to ensure that you couldn't hear the quantisation noise  in a silent passage with the volume on max.

At 44.1 kHz (and 48 kHz) there's really no safe place to put noise. Dither shaping at these sample rates relies on our hearing being less sensitive at higher frequencies. Allowing the noise level to rise a bit beyond, say, 12 kHz gives a little more dynamic range in the lower region. The trouble is that the 10 dB or so difference in sensitivity between 5 kHz and 10 kHz is barely enough to allow any useful shaping before the increased noise level at high frequencies becomes audible. With higher sample rates, you can readily raise the noise level by 20 dB or more in the 40 kHz region without anything bad happening. Even there, though, if the level is too high, IMD will shift the noise back into the audible range.

[Summit]

One of the pros of oversampling is that the noise-shaping, filtering etc can be done at much higher frequency than what the recording has natively.  

[danadam]

2 hours ago, mansr said:

Shaped dither at 44.1 kHz is tricky as it easily becomes audible. At 88.2 kHz and up, there's a wider range of completely inaudible frequencies where noise can be dumped without issue.

AFAICT shaped filter for >48kHz is not available in sox. Do you know why?

[mansr]

1 minute ago, danadam said:

AFAICT shaped filter for >48kHz is not available in sox. Do you know why?

Nobody made a suitable one.

[adamdea]

1 hour ago, Summit said:

One of the pros of oversampling is that the noise-shaping, filtering etc can be done at much higher frequency than what the recording has natively.  

Do you mean oversampling in a dac? If so I'm not sure I follow about the noise shaping. Or do you mean oversampling in an adc (ie using a higher sampling rate than the final distribution format). If so I follow.

[Summit]

4 hours ago, adamdea said:

Do you mean oversampling in a dac? If so I'm not sure I follow about the noise shaping. Or do you mean oversampling in an adc (ie using a higher sampling rate than the final distribution format). If so I follow.

 

Yes I meant oversampling in a DAC, although filtering and noise-shaping can be made in both an ADC, DAC or in computer software like HQPlayer.

 

Any signal, can be frequency shaped using filters. This is what filters do – they change amplitudes of frequency components of a signal that passes through them. They can make high frequency components weaker (low-pass filter), or can make them stronger (high-pass filter) or can frequency-shape a signal any way you want.

 

We already used some frequency shaping in our oversampling chapter – to cut off high frequencies from a quantized signal (we weakened high-frequency components down to zero).

 

The quantization noise can be shaped to improve signal-to-noise ratio of a quantized signal. But how do we do it?

 

We cannot simply place a filter at the output of our quantizer because the quantizer must be the last element in our digitalization chain – if we place a filter after it, the output signal from filter will not be quantized any more and we will not be able to use it for a digital computer.

 

Imagine that you oversampled your signal and now you have a lot of frequency space at higher frequencies that are not used by your actual signal. Only the quantization noise occupies this region. Now we could simply cancel that noise portion by filtering out these frequencies as explained in the oversampling chapter. However we can do much better – let’s re-shape the quantization noise so it will mostly be pushed away from lower-frequency regions and forced into higher-frequency regions and just then we cancel it out. That would be really charming.

 

http://charming.awardspace.com/delta_sigma/delta_sigma.html

[fas42]

In experiments I did with severely reduced bit depth material, the type or lack of dither makes a huge subjective difference - results ranged from unlistenable, to pleasantly "noisy". But with conventional recordings I have never heard an issue, unless the engineers made an outright mistake.

[adamdea]

14 hours ago, Summit said:

 

Yes I meant oversampling in a DAC, although filtering and noise-shaping can be made in both an ADC, DAC or in computer software like HQPlayer.

 

Any signal, can be frequency shaped using filters. This is what filters do – they change amplitudes of frequency components of a signal that passes through them. They can make high frequency components weaker (low-pass filter), or can make them stronger (high-pass filter) or can frequency-shape a signal any way you want.

 

We already used some frequency shaping in our oversampling chapter – to cut off high frequencies from a quantized signal (we weakened high-frequency components down to zero).

 

The quantization noise can be shaped to improve signal-to-noise ratio of a quantized signal. But how do we do it?

 

We cannot simply place a filter at the output of our quantizer because the quantizer must be the last element in our digitalization chain – if we place a filter after it, the output signal from filter will not be quantized any more and we will not be able to use it for a digital computer.

 

Imagine that you oversampled your signal and now you have a lot of frequency space at higher frequencies that are not used by your actual signal. Only the quantization noise occupies this region. Now we could simply cancel that noise portion by filtering out these frequencies as explained in the oversampling chapter. However we can do much better – let’s re-shape the quantization noise so it will mostly be pushed away from lower-frequency regions and forced into higher-frequency regions and just then we cancel it out. That would be really charming.

 

http://charming.awardspace.com/delta_sigma/delta_sigma.html

This is about A/D and what they are talking about it shaping the quantisation noise from the delta sigma conversion from analogue. 

I don’t think it applies to a dac except in relation to requantisation- eg the additional noise in the delta sigma conversion when turning the (usually) 16 bit pcm data into low bitrate data. 

But this does not afaik affect the shape of the quantisation noise implicit in the original data. 

Happy to be corrected if I have misunderstood. 

[Summit]

1 hour ago, adamdea said:

This is about A/D and what they are talking about it shaping the quantisation noise from the delta sigma conversion from analogue. 

I don’t think it applies to a dac except in relation to requantisation- eg the additional noise in the delta sigma conversion when turning the (usually) 16 bit pcm data into low bitrate data. 

But this does not afaik affect the shape of the quantisation noise implicit in the original data. 

Happy to be corrected if I have misunderstood. 

 

“Modern delta-sigma converters rely on a combination of internal high-speed upsampling and noise-shaping algorithms (digital feedback wrapped around the switch array) to attain their performance. Without dither + noise-shaping algorithms, you only get 12-bits to 14-bits of resolution, and we'd all still be using ladder converters.”

 

"Single-bit converters (and the SACD/DSD system) rely on dither and noise-shaping even more. Without noise shaping, SACD/DSD at 64fs (2.8224MHz) would only have 6-bits, or 36dB, of dynamic range, about the same as a not-very-good telephone connection."

 

[R2R DAC]…: “The ideal noise spectra is a "triangular" spectra centered on the Nyquist frequency of the physical converter—in other words, if the converter is running at 352.8MHz, the center of the dither-noise spectra should be 176.4kHz.”

 

https://positive-feedback.com/Issue66/dsd.htm

 

“A Sigma-Delta based DA converter realizes a high SNR with the use of a DAC with few quantization levels and noise-shaping techniques. In the digital domain the input signal to the DAC is shaped, such that the quantization noise of the DAC is moved to high frequencies. In the analog domain a passive low-pass filter removes the quantization noise, resulting in a clean baseband signal. The structure of a Sigma-Delta DAC is, except for some special PWM systems, a feed-forward solution, i.e. there is no feed-back from the analog output into the noise-shaping filter. Because the noise-shaping feed-back signal is not crossing the analog-digital boundary, the name Sigma-Delta DAC is confusing and misleading. A Sigma-Delta DAC is the combination of a DD converter and a high-speed few-bit DAC. In Fig. 2.6 the complete Sigma-Delta DAC structure is shown. The digital n-bit input signal is passed to a DD converter which upsamples the input to N × Fs before an all digital SDM reduces the word-length. The noise-shaped m-bit signal is passed to the mbit DAC which converts the digital signal to the analog domain. Finally the analog signal is filtered to remove the out-of-band quantization noise.”

 

file:///C:/Users/AS/Downloads/9789400713864-c2.pdf

[esldude]

4 minutes ago, Summit said:

 

“Modern delta-sigma converters rely on a combination of internal high-speed upsampling and noise-shaping algorithms (digital feedback wrapped around the switch array) to attain their performance. Without dither + noise-shaping algorithms, you only get 12-bits to 14-bits of resolution, and we'd all still be using ladder converters.”

 

"Single-bit converters (and the SACD/DSD system) rely on dither and noise-shaping even more. Without noise shaping, SACD/DSD at 64fs (2.8224MHz) would only have 6-bits, or 36dB, of dynamic range, about the same as a not-very-good telephone connection."

 

[R2R DAC]…: “The ideal noise spectra is a "triangular" spectra centered on the Nyquist frequency of the physical converter—in other words, if the converter is running at 352.8MHz, the center of the dither-noise spectra should be 176.4kHz.”

 

https://positive-feedback.com/Issue66/dsd.htm

 

“A Sigma-Delta based DA converter realizes a high SNR with the use of a DAC with few quantization levels and noise-shaping techniques. In the digital domain the input signal to the DAC is shaped, such that the quantization noise of the DAC is moved to high frequencies. In the analog domain a passive low-pass filter removes the quantization noise, resulting in a clean baseband signal. The structure of a Sigma-Delta DAC is, except for some special PWM systems, a feed-forward solution, i.e. there is no feed-back from the analog output into the noise-shaping filter. Because the noise-shaping feed-back signal is not crossing the analog-digital boundary, the name Sigma-Delta DAC is confusing and misleading. A Sigma-Delta DAC is the combination of a DD converter and a high-speed few-bit DAC. In Fig. 2.6 the complete Sigma-Delta DAC structure is shown. The digital n-bit input signal is passed to a DD converter which upsamples the input to N × Fs before an all digital SDM reduces the word-length. The noise-shaped m-bit signal is passed to the mbit DAC which converts the digital signal to the analog domain. Finally the analog signal is filtered to remove the out-of-band quantization noise.”

 

file:///C:/Users/AS/Downloads/9789400713864-c2.pdf

 

Is there some point to your post related to the thread topic?

[Summit]

20 minutes ago, esldude said:

 

Is there some point to your post related to the thread topic?

 

Yes it’s an explanation that noise-shaping, also can, and is often used for Red book because of oversampling to higher frequency. If the OP thinks my posts are OT I will ask Chris to erase them. Ok?

[adamdea]

 

2 minutes ago, Summit said:

 

Yes it’s an explanation that noise-shaping, also can, and is often used for Red book because of oversampling to higher frequency.

 

Well sort of... It's just a pair of quotes explaining how a ds dac works. Of course the oversampling  and noise shaping is "used for red book" in the sense that it used to convert redbook to analogue but it has nothing to do with noise shaping the 16 bit quantisation noise.  TBH the distinction between quantisation noise in the recording and requantisation noise caused by the conversion process is a bit tricky and is not clearly explained in most generalist texts.

It's easy to hop from a/d conversion noise to 16 quantisation noise to d/a conversion noise in an explanation because of  course the chain is only as strong as the weakest link. Without da conversion noise shaping,  (most?) ds dacs would not achieve 16 bit performance in the audible band. Equally if you really want to see the a/d converter noise floor then you need to buy 24/96 or ideally 24/192 (in fact that's pretty much all you get going from 24/96 to 24/192)

[esldude]

29 minutes ago, Summit said:

 

Yes it’s an explanation that noise-shaping, also can, and is often used for Red book because of oversampling to higher frequency. If the OP thinks my posts are OT I will ask Chris to erase them. Ok?

I am still searching for the point.  You referring to noise shaping in sigma-delta DACs, or noise-shaping during mastering before being sent to the redbook format?  Does this have something to do with noise shaping making redbook unlistenable or sound worse?

 

 

[Jud]

On 7/13/2018 at 10:24 AM, adamdea said:

 

Have a go at downloading the test files here and see whether you have a preference for each pair. If so note them down.

https://archimago.blogspot.com/2014/04/internet-test-24-bit-vs-16-bit-audio.html

There is no haze or compression with 16 bits relative to 24 bits, just a higher quantisation/dither noise level. [this was covered in a thread recently] However for most recordings the background noise level is much higher than the 16 bit quantisation noise. If you want to (actually) hear the difference its best to use stupidly amplified digital silence. 

 

There was an article by Bob Stuart years ago, supposedly arguing for hi rez where he showed that if you ignored masking [which is absurd] you could just hear triangular pdf dither noise in a 16 bit recording if you played it at 120dB peak level [which you can't or you would be deaf]. Even then you wouldn't be able to hear the quantisation noise if you used noise shaped dither.

 

I of course agree with most of this (can't do otherwise, since it's factual).  The single piece I'd like to have a little conversation over is the thing about ignoring masking being absurd.  Certainly, ignoring masking from ambient sound is absurd, and that is what we mostly are talking about regarding noise floor.

 

But I also have read people saying that music masks noise/distortion from the equipment or recording.  While also obviously true, since we are trying to hear the music rather than the noise, I would think we'd be concerned about noise/distortion masking (interfering with clearly discerning) low level details in the music, rather than vice versa.

[esldude]

19 minutes ago, Jud said:

 

I of course agree with most of this (can't do otherwise, since it's factual).  The single piece I'd like to have a little conversation over is the thing about ignoring masking being absurd.  Certainly, ignoring masking from ambient sound is absurd, and that is what we mostly are talking about regarding noise floor.

 

But I also have read people saying that music masks noise/distortion from the equipment or recording.  While also obviously true, since we are trying to hear the music rather than the noise, I would think we'd be concerned about noise/distortion masking (interfering with clearly discerning) low level details in the music, rather than vice versa.

Well we hear some 10 or 15 or maybe even 20 db into noise, but we don't hear noise 20 db into music?  Why?  Look at how masking works. 

 

One of things our ear does is break up things into maybe 30 or 32 bands of frequencies.  So some music can be at a level so that SPL wise it is 20 db below the noise level over the whole band we can hear, but the music content is still above the part of the noise in that band our ears are filtering with.  Something like the way a tone can be unmasked in an FFT even though the signal level of the tone is below the noise level for the entire band.   The ear is more advanced than a static FFT in that it moves those filters around.  Still it moves those around looking for tones which within the ear's filters is above noise in that filter band.  

 

The reverse with our ear hearing noise above music doesn't work.  No doubt that is why our ears work like they do.  Eventually as the signal level of music descends far enough into noise yes the noise interferes with us hearing it at all much less details of it. 

 

The noise levels in 16 bit even with dither and noise shaping are low enough it is difficult to come up with a situation where it can interfere with low level detail.  Not impossible, but very difficult.  Very, very few recordings have a noise floor low enough for 16 bit to become an issue in terms of noise. And by very few I mean like probably not even .1% maybe not even .01% so basic noise issues with 16 vs 24 bit isn't why someone is dissatisfied with 16 bit. 

[mansr]

2 minutes ago, Jud said:

I of course agree with most of this (can't do otherwise, since it's factual).  The single piece I'd like to have a little conversation over is the thing about ignoring masking being absurd.  Certainly, ignoring masking from ambient sound is absurd, and that is what we mostly are talking about regarding noise floor.

 

But I also have read people saying that music masks noise/distortion from the equipment or recording.  While also obviously true, since we are trying to hear the music rather than the noise, I would think we'd be concerned about noise/distortion masking (interfering with clearly discerning) low level details in the music, rather than vice versa.

Masking simply means one sound becoming indiscernible in the presence of another dominating sound. In the present context, we're talking about low-level noise being masked by louder sounds, either in the recording or from the surroundings. While 16-bit TPDF dither noise is generally not audible at normal playback levels, older recordings sourced from tape often have a background hiss easily discernible until the music starts, at which point it is no longer noticeable.

[PeterSt]

3 minutes ago, mansr said:

have a background hiss easily discernible until the music starts, at which point it is no longer noticeable.

 

Nice theory.

[mansr]

Just now, PeterSt said:

Nice theory.

What's that supposed to mean?

[PeterSt]

1 minute ago, mansr said:

2 minutes ago, PeterSt said:

Nice theory.

What's that supposed to mean?

 

That practice is different.

[Jud]

4 minutes ago, mansr said:

Masking simply means one sound becoming indiscernible in the presence of another dominating sound. In the present context, we're talking about low-level noise being masked by louder sounds, either in the recording or from the surroundings. While 16-bit TPDF dither noise is generally not audible at normal playback levels, older recordings sourced from tape often have a background hiss easily discernible until the music starts, at which point it is no longer noticeable.

 

Yes, but...

 

When I ask you what you want to listen to, I don't anticipate the reply "Some TPDF dither please, or maybe some lovely IMD."  What we're trying to hear is the music, so whether distortion is audible doesn't seem so important to me as whether it makes some of the music harder to hear. 

 

However, for purposes of this thread, @esldude's point that this isn't going to be a big practical distinction between 16 and 24 bit recordings makes sense to me.

[adamdea]

25 minutes ago, Jud said:

 

I of course agree with most of this (can't do otherwise, since it's factual).  The single piece I'd like to have a little conversation over is the thing about ignoring masking being absurd.  Certainly, ignoring masking from ambient sound is absurd, and that is what we mostly are talking about regarding noise floor.

 

But I also have read people saying that music masks noise/distortion from the equipment or recording.  While also obviously true, since we are trying to hear the music rather than the noise, I would think we'd be concerned about noise/distortion masking (interfering with clearly discerning) low level details in the music, rather than vice versa.

All I was getting at was that if you have music playing so that the peaks are at 120dB then your chances of hearing any noise (or anything else) at 90dB below peak is pretty slight. Your ear has a sort of variable dynamic range so when it hears loud things it reduces its sensitivity  to avoid damage. So the quietest sound can hear i the presence of a loud signal is very different from the quietest sound you can hear period. 

 

Putting it another way- masking is the main reason why perceptual codecs work.

 

Noise could mask signal in principle but

1) you can hear a tone 20 db below the overall noise level because it will still be higher than the noise level in the relevant bin. so 16 bit allows your to hear tones at -110 dB below (theoretical) peak

2) with a changing signal (aka music) and a constant noise won't your brain focus on the signal?

[esldude]

1 minute ago, adamdea said:

All I was getting at was that if you have music playing so that the peaks are at 120dB then your chances of hearing any noise (or anything else) at 90dB below peak is pretty slight. Your ear has a sort of variable dynamic range so when it hears loud things it reduces its sensitivity  to avoid damage. So the quietest sound can hear i the presence of a loud signal is very different from the quietest sound you can hear period. 

 

Putting it another way- masking is the main reason why perceptual codecs work.

 

Noise could mask signal in principle but

1) you can hear a tone 20 db below the overall noise level because it will still be higher than the noise level in the relevant bin. so 16 bit allows your to hear tones at -110 dB below (theoretical) peak

2) with a changing signal (aka music) and a constant noise won't your brain focus on the signal?

Instantaneous dynamic range of the ear is around 60 db.  The rest is what you are talking about.  How quickly other effects expand that and how audible things are gets complicated.  But yes, even if you had playback peaking pretty often at 120 db and noise 90 db down, which by itself would be audible, you'll never hear it with that signal as it masks that level even for a period of time after it drops away.  

[mansr]

6 minutes ago, esldude said:

Well we hear some 10 or 15 or maybe even 20 db into noise, but we don't hear noise 20 db into music?  Why?

That's not a meaningful question. Noise, in the sense we're discussing here, is a wide-band signal with a smooth spectrum. When noise level is given as a single figure, it is the integrated power, possibly with weighting, over the entire frequency range of interest. A tone or music, which is a collection of tones, with the same SPL has the power concentrated in a few narrow regions. The spectral intensity, measured in dB/Hz, in these regions is much higher than that of the noise. That is why we can hear the signal even in the presence of noise with higher total power.

 

I think you were saying something similar, if in a roundabout way.

[esldude]

14 minutes ago, mansr said:

That's not a meaningful question. Noise, in the sense we're discussing here, is a wide-band signal with a smooth spectrum. When noise level is given as a single figure, it is the integrated power, possibly with weighting, over the entire frequency range of interest. A tone or music, which is a collection of tones, with the same SPL has the power concentrated in a few narrow regions. The spectral intensity, measured in dB/Hz, in these regions is much higher than that of the noise. That is why we can hear the signal even in the presence of noise with higher total power.

 

I think you were saying something similar, if in a roundabout way.

Yes that is exactly what I was saying.  

[esldude]

Here is a quick example.  

Pink noise at about - 40 dbFS and less than that from peak levels in the music.  Noise alone for the first few seconds and then with the music. Bonus points if you can tell me when the noise goes away.  It isn't impossible, you can hear when it is removed partway thru the snippet, but I think it is harder than some might expect.  This level of noise would be like bad cassette tape with no Dolby noise reduction.

 

 06 Heat Wave.mp3