3 ms·

This is a bit of an interesting side discussion (I use sox indeed to down sample) but I've recently changed my stance on this after reading the results of some

by derriz 8y ago

This is a bit of an interesting side discussion (I use sox indeed to down sample) but I've recently changed my stance on this after reading the results of some blind testing of MQA (24-bit at 192KHz or something like that) recordings. My knowledge of analog and digital signal processing is superficial but I think I can at least understand the arguments. I used to buy your argument on why sampling at more than 48KHz is a waste of time. The argument goes something like: humans cannot hear frequencies above 20KHz (less for older people) so as long as we sample above the Niquist frequency (40Khz) then the audible frequencies can be perfectly represented. 48KHz > 2*hearing range so QED, higher frequency sampling is at best a waste of time and may even make the sound worse. The credible counter to this argument that I've read is that this argument requires being able to perfectly filter out the >20KHz frequencies before you sample. A perfect filter with an exact 20KHz cut-off does not exist in the real world - some of the audible range gets affected at the high end and also some >20KHz signal gets through. One of the resulting artifacts is pre-imagining (like a subtle-reverse echo).

KozmoNau7 8y ago
If the blind testing of MQA compared MQA encoded tracks with and without MQA decoding, they were actually comparing MQA to 13-bit audio, as the (lossy) MQA data is encoded in the last 3 bits of "noise". That can definitely be heard on dynamic tracks. But CD-quality audio is 16-bit, which does not have an audible noise floor for music. >"The credible counter to this argument that I've read is that this argument requires being able to perfectly filter out the >20KHz frequencies before you sample. A perfect filter with an exact 20KHz cut-off does not exist in the real world - some of the audible range gets affected at the high end and also some >20KHz signal gets through." You don't need to cut off sharply at 20kHz, you have a band of 2.05kHz (for 44.1) or 4kHz, even with a non-oversampling DAC. Piece of cake for digital filters even 20 years ago, and reasonably possible even with analog filters. But it's not even relevant for most people, since most DACs oversample to at least double sample rate, and can thus implement smooth gentle filters with no issues. That gives you the benefit of smooth filters you mention for high sampling rates, without wasting any storage space, making "hi-res" music pointless. Read this: https://www.xiph.org/~xiphmont/demo/neil-young.html https://www.xiph.org/~xiphmont/demo/neil-young.html
- derriz 8y ago
  My understanding was that the blind tests were not as naive as that and the comparisons were with full 16/44 recordings and playback. Funny enough I have read that article on xiph.org - which is a good read - and was completely convinced. But listening to guys like this https://youtu.be/geaoEt-9V-w https://youtu.be/geaoEt-9V-w has made me question this. He claims that in practice it is impossible to perfectly band-limit a signal (whether with digital or analog filters) and this gives rise to artifacts in the reproduced signal which can be observed by oscilloscope. And yes the guy seems to know what he's talking about but I'm happy to be corrected on that. Having said all that, I've little interest in the top-end audio market so it's somewhat irrelevant to my needs. Support for hi-res audio is pretty down my list of irritations with the Sonos, to be honest.
  - KozmoNau7 8y ago
    Oh jeez, not that guy. He's a former hifi salesman, with only a very tenuous grasp of the technical details. I've experienced this countless of times, the salespeople completely buy into the "high end" mumbo-jumbo and don't bother to actually learn the technical aspects. It's very common for them to show some minor effect on an oscilloscope, but ignore the fact that any artifacts from ringing and such would be close to the Nyquist frequency and below -80dBFS in the signal. In other words, utterly inaudible. For all their talk about trusting their ears, they seem to conveniently forget that maxim the moment it doesn't suit their made-up arguments. Yes, in theory you cannot 100% perfectly band limit a signal. However, in practice you can band limit it well enough that any artifacts are either very close to or actually below the noise floor, and thus completely irrelevant. On the complete opposite end of the scale of salesperson:tech guru, the article I linked was written by none other than Chris "Monty" Montgomery (xiphmont on this very site), who among other things founded the Xiph.org free codec foundation, and either was fully responsible for creating the Ogg container format, the Vorbis, FLAC, Speex and Opus audio formats, and countless other technical achievements, or at least had a big hand in them. Free and open digital audio owes a gigantic debt to him, and if anyone knows their way around digital audio, it's him.

by derriz 8y ago

KozmoNau7 8y ago

If the blind testing of MQA compared MQA encoded tracks with and without MQA decoding, they were actually comparing MQA to 13-bit audio, as the (lossy) MQA data is encoded in the last 3 bits of "noise". That can definitely be heard on dynamic tracks. But CD-quality audio is 16-bit, which does not have an audible noise floor for music. >"The credible counter to this argument that I've read is that this argument requires being able to perfectly filter out the >20KHz frequencies before you sample. A perfect filter with an exact 20KHz cut-off does not exist in the real world - some of the audible range gets affected at the high end and also some >20KHz signal gets through." You don't need to cut off sharply at 20kHz, you have a band of 2.05kHz (for 44.1) or 4kHz, even with a non-oversampling DAC. Piece of cake for digital filters even 20 years ago, and reasonably possible even with analog filters. But it's not even relevant for most people, since most DACs oversample to at least double sample rate, and can thus implement smooth gentle filters with no issues. That gives you the benefit of smooth filters you mention for high sampling rates, without wasting any storage space, making "hi-res" music pointless. Read this: https://www.xiph.org/~xiphmont/demo/neil-young.html https://www.xiph.org/~xiphmont/demo/neil-young.html

derriz 8y ago
My understanding was that the blind tests were not as naive as that and the comparisons were with full 16/44 recordings and playback. Funny enough I have read that article on xiph.org - which is a good read - and was completely convinced. But listening to guys like this https://youtu.be/geaoEt-9V-w https://youtu.be/geaoEt-9V-w has made me question this. He claims that in practice it is impossible to perfectly band-limit a signal (whether with digital or analog filters) and this gives rise to artifacts in the reproduced signal which can be observed by oscilloscope. And yes the guy seems to know what he's talking about but I'm happy to be corrected on that. Having said all that, I've little interest in the top-end audio market so it's somewhat irrelevant to my needs. Support for hi-res audio is pretty down my list of irritations with the Sonos, to be honest.
- KozmoNau7 8y ago
  Oh jeez, not that guy. He's a former hifi salesman, with only a very tenuous grasp of the technical details. I've experienced this countless of times, the salespeople completely buy into the "high end" mumbo-jumbo and don't bother to actually learn the technical aspects. It's very common for them to show some minor effect on an oscilloscope, but ignore the fact that any artifacts from ringing and such would be close to the Nyquist frequency and below -80dBFS in the signal. In other words, utterly inaudible. For all their talk about trusting their ears, they seem to conveniently forget that maxim the moment it doesn't suit their made-up arguments. Yes, in theory you cannot 100% perfectly band limit a signal. However, in practice you can band limit it well enough that any artifacts are either very close to or actually below the noise floor, and thus completely irrelevant. On the complete opposite end of the scale of salesperson:tech guru, the article I linked was written by none other than Chris "Monty" Montgomery (xiphmont on this very site), who among other things founded the Xiph.org free codec foundation, and either was fully responsible for creating the Ogg container format, the Vorbis, FLAC, Speex and Opus audio formats, and countless other technical achievements, or at least had a big hand in them. Free and open digital audio owes a gigantic debt to him, and if anyone knows their way around digital audio, it's him.