Why 24/192 is a bad idea?
- Thread starter audioguy
- Start date
As I said a number of people did this listening test, not just me. The majority recognised & described the differences that they found in the 20 samples file. What part of the blind testing do you not understand?
This waveform, with it's 0.2dB ripple is recognisable as different by these people. Therefore your stated criteria for audio transparency is wrong in this regard!
Thanks for clarifying. Yes, I had thought 0.1 dB total span was the accepted maximum for inaudibility.
--Ethan
Ethan, is that 100 db S/N for transparency attributed to you correctly states your position?
Update: After a private exchange with Monty (xiphmont) we understand each other's methodology on how the bit extension is presented and handled. When I calculate the way he did I too obtain no additional spurs, just a d.c. offset. And, of course, if he uses my method he sees spurs. For the curious, the three methods I tried after scaling 24-bit data to 16-bit data were:
Rounding to 16 bits = no extra spurs
Truncating to 16 bits = odd-order harmonics appear at low level (well above the quantization noise floor)
Ceiling or floor function (always "rounding" to the integer above or below) = no harmonics but additional d.c. offset term added
Note part of the difference in our methodolgy is that I (almost) always use relatively prime signal frequencies to avoid the need for FFT windowing, and that method will "hide" some of the spurs Monty (or anyone else) sees in a general case. The prime method, as presented in IEEEE Standard 1241 (and elsewhere), eliminates frequency "clumping" that causes quantization energy to be distributed more heavily in some FFT bins than others, creating harmonic spurs. These are a consequence of the signal frequency, sampling frequency, and FFT length. They are not caused by nonlinearities (distortion) in the amplifiers, filters, or thresholds of the data converters themselves. Since as a designer I have always been interested in nonlinearities of the converter (ADC or DAC) itself, and not quantization errors related to binning, I always use relatively prime frequencies. That and my rounding/truncation of unary data meant I did not see the same FFT result Monty saw.
Monty, I have no wish to intrude in this discussion, but wanted to clarify our stances. Please correct as needed.
All is well - Don
Rounding to 16 bits = no extra spurs
Truncating to 16 bits = odd-order harmonics appear at low level (well above the quantization noise floor)
Ceiling or floor function (always "rounding" to the integer above or below) = no harmonics but additional d.c. offset term added
Note part of the difference in our methodolgy is that I (almost) always use relatively prime signal frequencies to avoid the need for FFT windowing, and that method will "hide" some of the spurs Monty (or anyone else) sees in a general case. The prime method, as presented in IEEEE Standard 1241 (and elsewhere), eliminates frequency "clumping" that causes quantization energy to be distributed more heavily in some FFT bins than others, creating harmonic spurs. These are a consequence of the signal frequency, sampling frequency, and FFT length. They are not caused by nonlinearities (distortion) in the amplifiers, filters, or thresholds of the data converters themselves. Since as a designer I have always been interested in nonlinearities of the converter (ADC or DAC) itself, and not quantization errors related to binning, I always use relatively prime frequencies. That and my rounding/truncation of unary data meant I did not see the same FFT result Monty saw.
Monty, I have no wish to intrude in this discussion, but wanted to clarify our stances. Please correct as needed.
All is well - Don
Where did that attribution come from? Regardless, I sometimes say -80 dB is quiet enough not to hear. But what really matters is the spectrum of the noise versus the spectrum of the music. If I play a sustained low A note on my Fender bass at 55 Hz, and there's a fair amount of the third harmonic at 165 Hz, you'll need a lot more noise at those frequencies than -80 dB. But you might possibly hear a 3 KHz tone if it's at -80. Most people will hear 3 KHz at -60 in the presence of 55 and 165 Hz. This is a very basic principle, called the Masking Effect.
--Ethan
Originally from the NWAVGuy's article. I chased it and it came from this AES workshop presentation you made: http://www.google.com/url?sa=t&rct=...8O3qAw&usg=AFQjCNHKCz8k4mRLRtZLrqDImoaOzeExzQ (note: clicking on it downloads a file)
"22:10 Aside from devices that intentionally add “color” by changing the frequency response or adding distortion, it’s generally accepted that audio gear should aim to be transparent. This is easily tested by measuring the above four parameters with various test signals. If the frequency response is flat to less than 1/10th dB from 20 Hz to 20 KHz, and the sum of all noise and distortion is at least 100 dB below the music, a device can be said to be audibly transparent. A device that’s transparent will sound the same as every other transparent device, whether a microphone preamp or DAW summing algorithm."
From that video that you CONSTANTLY refer people to! If you have no idea of it's contents is it just stream of conciousness uttering, then?
Full credit to Amirm for sitting through the full 30+ mins of it - I couldn't afford to waste that much time out of what's left of my time on earth. Considering it's author isn't bothered, why should anyone else be? (EDIT: I see that Amirm just read (or maybe searched) the transcript of that video - good move
- I was thinking that transcribing it was a bit of effort)So, Ethan, are you now equivocating about your statements that you so bullishly put forth on that video & also here & elsewhere! Can you answer amirm's & my direct questions - are you standing over the 4 criteria for audio transparency quoted (now that you know they are from you
)?Edit: Let's just open up a new thread on this as it deserves some attention!
Last edited:
John, Amir is good too (his actual first name).
...The 'm' at the end stands for Madrona, his company, I believe.
* Am I right Amir? ...Or does it stand for your last name: Majidimehr?
Sorry, just thought of mentioning, that's all. :b
...The 'm' at the end stands for Madrona, his company, I believe.
* Am I right Amir? ...Or does it stand for your last name: Majidimehr?
Sorry, just thought of mentioning, that's all. :b
Actually my alias is the same as what I had at Microsoft. The convention there is first name plus first letter of last name. But yes, feel free to call me Amir.
Ah, first letter of your last mame. ...Microsoft, Madrona, Majidimehr. :b
Me, in French, it's Robertc. A is my middle name; Andre. ...From my sig (in English): Bob A CHarr, or BACH. :b
And what does it have to do with 24/192? ...Everything & nothing. ...C for Collateral.
Me, in French, it's Robertc. A is my middle name; Andre. ...From my sig (in English): Bob A CHarr, or BACH. :b
And what does it have to do with 24/192? ...Everything & nothing. ...C for Collateral.
Guys, you can call me anything you want just not early in the morning
Yes, John Kenny. :b
* Who's John Darko? ...I know, the audio writer.
High res audio 24/192, good or bad idea? :b
* Me, like I said before, I'm real happy with 20/88.2, or 24/176.4 is mighty fine as well. :b
...32/352.8? Why not. :b
* Me, like I said before, I'm real happy with 20/88.2, or 24/176.4 is mighty fine as well. :b
...32/352.8? Why not. :b
The topic is Purely Personal Performance (PPP), and the extent to which many audiophiles go to have their PPP become a Universal Performance Window.
Last edited:
Similar threads
