Comments by "Lawrence D’Oliveiro" (@lawrencedoliveiro9104) on "Nyquist-Shannon; The Backbone of Digital Sound" video.

  1. Particularly since PC hard drives of the period were only on the order of tens of megabytes. This was a particular issue to those trying to create CD-ROMs. Even when DVDs first came out, they held more than the common hard drive capacity, though the ratio was smaller. By about the ’00s, hard drives had overtaken optical media and left them in the dust.
    11
  2. One thing to consider is the difference between raw capture quality and final delivery quality. If you are going to mess around with the sound (filter it, apply other effects, mix it down), then you want the final result to make full use of the capacity of human hearing. Since all the processing will involve some kind of degradation, it makes sense to start with raw quality higher than human hearing, to give you room to play around. The same reasoning applies to “lossless” versus “lossy” image encoding, and the use of depths greater than 8 bits per pixel component.
    3
  3. In the analog realm, you have this thing called “generation loss”, where the quality degrades through repeated copying and processing. A similar thing happens even in the digital world, though to a lesser extent, because while straight copying is effectively 100% perfect, actual processing and filtering involves finite-precision computer arithmetic, which is subject to rounding errors. So yes, it helps to start with a higher-quality set of raw samples, to give yourself more headroom to play around.
    2
  4. True, this is an omission in his video. Quantization artifacts are more prominent for simple tones than for complex mixtures of sound. Dithering helps, because the random-sounding noise is less noticeable than repeating distortion.
    2
  5. CDs (and DVDs) use lasers. As did good old “laserdisc” (hence the name).
    2
  6. That’s an analog definition. And what about higher-order filters?
    2
  7. Beware of somebody mixing up THD, quantization noise and aliasing, and then claiming to be an “actual signal processing engineer”.
    2
  8. The car I’ve been driving lately (newer than my own) has no CD player, but it does have Bluetooth and USB ports.
    2
  9. You are trying to produce a quality result at the final delivery sample rate and bit depth. You have to leave room for processing degradation so you don’t end up producing a worse result than that. Consider the multiply-add operation that is heavily used in FFT and convolution computations. In modern hardware, it is common to have an instruction that can perform it with one rounding error, instead of separate ones for the multiply and the add. So that cuts the rounding errors in half, but it still does not reduce them to zero. And remember you typically have to perform thousands of these operations for each sample.
    2
  10. On Vimeo you can re-upload.
    2
  11. Yawn How is what you prefer relevant to this video?
    2
  12. I can only see tiny thumbnails, with no option to make them bigger.
    1
  13. Hearing is a sense. If you are not “hearing” it, then what “sense” are you using?
    1
  14. Digital filters are quite capable of implementing a “brick wall” cutoff.
    1
  15. This idea that inaudible frequencies can make an audible difference ... the only rational explanation I can come up with involves some non-linearity of response of the ear. Perhaps not an explanation that audiophiles would like ...
    1
  16. You didn’t mention THD? Yes you did.
    1
  17. 14:58 Analog filters cannot have a sharp cutoff (without introducing phase-shift artifacts), but digital filters can. Oversampling can give you a perfectly good cutoff at a 20kHz × 2 = 40kHz sampling rate, or any other rate you want. The reason for the 44.1kHz audio CD sampling rate has to do with the original use of adapted VCR machines for the audio mastering, and the need to able to adapt both PAL and NTSC machines for the purpose. Source: John Watkinson, The Art Of Digital Audio -- I have an old edition of this and his Art Of Digital Video -- both are still well worth reading.
    1
  18. 16:03 Ah, you did know at least part of that.
    1
  19. 3:28 Fourier is not complicated! (The main trouble I have with it is remembering where the π factors come in.) The wonderful symmetry between time/space-domain and frequency-domain representation of signals is just breathtakingly beautiful to behold. So there.
    1
  20. 5:15 See, nothing complicated about that, is there? ;)
    1
  21. That could be just the “ringing” he already alluded to.
    1
  22. But if you let them loose, where do they go?
    1
  23. You were talking about “loosing” them. If you didn’t mean “let them loose”, what did you mean?
    1
  24. 4:19 Yeah! (Thumbs up for getting that point across.)
    1
  25. 1:48 s/Kotelkinov/Kotelnikov/
    1
  26. We call it “hum”. What other word should we use?
    1
  27. Which MP3 encoder do you use? LAME seems to be the best.
    1
  28. Sigh He was so careful to keep emphasizing “band-limited”, and yet the ill-informed responses are so predictable ...
    1
  29. Remember that audio CDs came out soon after the whole quadraphonic fad went bust. So I think they were concentrating primarily on delivering high-quality 2-channel stereo sound, because that’s what the market clearly wanted. CDs were designed for audio first, and general computer data second. Whereas with DVDs it was the other way round. Thus, it was easier to add new audio and video formats on DVDs, just by adding new file format options.
    1
  30. Surround sound is more realistic. I think the problem was that 4 channels wasn’t enough; that’s why you have 5.1, or even 7.1 or more nowadays.
    1
  31. Yes, there is a point. Raising the amplitude of a quiet signal, or resampling a signal to a different sample rate -- digital numeric operations all have rounding errors.
    1
  32. No, it won’t be OK. You cannot magically recreate information that was never captured in the first place. And you cannot escape rounding errors. This is all basic Computing 101.
    1
  33. It matters when you do any kind of processing to the audio.
    1
  34. And what do you fill the extra bits with? Where do you magically create the information from, out of thin air?
    1
  35. OK, let’s use a simple example. Say you have captured a signal at 16 bits per sample. That gives you a noise floor of -90dB. Supposing the signal actually has a level of -45dB. You want to boost that to something closer to 0dB. But at the same time you end up boosting the noise floor to -45dB. How do you fix that?
    1
  36. Remember he actually said “perfectly reconstruct original band-limited signal”. That’s the key point.
    1
  37. If the D/A converter is not band-limited, then it cannot be reconstructing the original signal, can it?
    1
  38. You did notice he was careful to keep emphasizing throughout that the signal was “band-limited”, did you not?
    1
  39. Bravo! Good explanation. But to preempt the inevitable disbelievers, and paraphrase Michael “Brexit” Gove: “I think the audiophile public has had enough of experts”. Tell them to go back to their oxygen-free cabling ... maybe that lack of oxygen has gone to their heads ...
    1
  40. The effect of that is to lower the amplitude of the higher frequencies close to the Nyquist limit. The same thing happens with scanned digital images, because of the nonzero size of the pixel samples, and the cure for this is the well-known “unsharp mask” filter. The same kind of equalization filtering can be done to sound.
    1
  41. So what you are saying is that the response of the human ear is not linear, that it is subject to distortion.
    1
  42. Hearing aids have to do with sensitivity and frequency response.
    1
  43. Have you thought that it’s not just your ears, but also your mind -- i.e. a psychosomatic phenomenon? Something that you can only rule out by doing blind A-B tests.
    1
  44. Very possibly. Though DAT was effectively sabotaged for consumer use anyway, since the record labels were still terrified over the prospect of a recordable medium which could make perfect copies. I think DAT machines still gave you the option of selecting a 44.1kHz sample rate if you wanted.
    1
  45. Yeah, the ideal reconstruction involves sinc functions, which have an infinite extent both before and after the peak, and only slowly fade away from that peak. So in order to be causal (not respond to a signal change before it happens!), it would require an infinite amount of buffer lag.
    1
  46. Once the filter cutoff descends below the noise floor from other sources, that’s as good as zero, because you can never measure the difference.
    1
  47. Frequency-independent time delay happens with real-life sound all the time. Think of the difference between listening to a concert from the front seats as opposed to the back ones: every change in distance of 3m from the sound source corresponds to a difference in timing of 10ms. So really, digital sound is nothing special in regards to this.
    1
  48. I loved NICAM stereo. Your TV signal could go completely to crap, and the NICAM would be just about the last thing to cut out -- after the TV picture had become just about unwatchable.
    1