But, if there is not an established method to measure transparency that is published and researched, where does one draw a line on audibility? Nothing is as clear cut, there is always a fuzzy boundary and the width of the boundary may depend on what is being tested and also the system as a whole and the listener and it is best looked at as possibility values rather than a hard threshold. For example when do you stop hearing a difference in the tonal characteristics? It will depend on a lot of factor including the system preproducing the sound not just the transducer. It is not as simple as saying if the change is below 1dB no one can hear it, what if it is 1.1dB or 0.9dB? What if the AC is running? trained listener vs untrained, and so on.
Thank you for your suggestion, I am already a member of AES and thanks for your time.
@KMann isn't saying anything controversial.
True, but of course that is just an explanation of how a filter could theoretically correct for some recording/ADC errors. It is conditional on an ADC actually having filters that are not good enough and actually having enough content in the signal we’re recording that is above the Nyquist freq to cause significant aliasing in the 20kHz - 22.05kHz band. The poster who we’re questioning (@HIPlayerDSD) has not yet provided any examples of recordings that actually fulfil these two conditions.
That’s where it get‘s complicated or at least the marketing does (!), because it uses semantics of what an anti-imaging/reconstruction filter should actually do. For me, it seems obvious that it should reduce images and reconstruct the audible band, which has been relatively trivial to achieve for many years. So, if there is some audible difference, it is because the filter is effectively faulty.
I agree with this
If you really narrow the scope of what Signal processing means to your previous statement then yes. Because applying a digital filter for the purpose of removing out of band images is still signal processing.
I agree with this. However, showing the relationship between reconstruction accuracy in the audible band and reducing the out of band images is not a trivial one. One could argue, NOS DAC is enough as you do not hear the out of band images and all one needs to do is a simple EQ to correct the droop in the higher frequencies in the audible band.
I didn’t mean trivial in the sense that it’s simple but in the sense that it’s been possible for around 30 years or more, is well established/routine and is accomplished even by cheap DACs.
I don’t think one could rationally argue that a NOS DAC is enough, a “simple EQ” to correct the droop, a high-shelf boost with a low Q for example would also boost the level of the images. Although the images are beyond the range of human hearing, they may well be of sufficient magnitude to cause IMD within the audible band and even more likely if you’re also boosting them with an EQ shelf. I would think there would/could be additional issues from just using “a simple EQ”, inter-sample peak levels for example.
If I ever discovered I had specific songs with this problem in my collection, frankly I'd rather apply the filtering to the files themselves with Audacity than select an apodizing filter in my DAC and dull the highs on all of my other tracks.
There are a couple of serious issues with the statement you quoted. Firstly, “ringing caused by the ADCs or DSP used in production/recording” is actually rare and when it does occur it’s both: Low magnitude and up near the Nyquist freq. So as it’s inaudible, how does that constitute a “problematic signal” and why would it be “arguably preferable” to change it? Secondly, “apodizing filter” is largely just a marketing term, it doesn’t necessarily tell us anything about the filter’s response, technically (AFAIK) most modern anti-alias or anti-imaging filters could be described as “apodizing”. Therefore, an “apodizing filter” may or may not (inaudibly) improve ringing, in fact it could actually make the ringing worse in specific production/recording cases (although again, inaudibly)! Lastly, despite several requests GoldenSound has not provided a single example of a mastered recording containing such “problematic signals” that have actually been improved by an apodizing filter, even baring in mind it would be inaudible anyway!
But it does, if you go back to the etymology of the word, like the Wikipedia article does. An apodizing filter was so named because it "removes the foot", i.e. in comparison to a worse filter that allows some egregious lobe or hump to pass right at the beginning of the stopband (or even centered on the cutoff frequency?), this one flattens that area out significantly and therefore "removes the foot" and therefore is a-pod-izing.
In this case there's very little to suggest it's "removing any foot", so it doesn't fit the name "apodizing" all that well.
“Apodizing” does mean “removes the foot” but then you appear to have invented what you think the “foot” is. My maths knowledge is not that great, so maybe someone like @71 dB can explain it better or correct my understanding but AFAIK: “Apodizing” in practice just means effectively “multiplying” a function/waveform data with a “Window Function” where only the data within the “window” interval is multiplied by the window function, IE. The window overlap is the “foot” that is removed. Apodization is used in various fields but in the case of digital filters it is used to convert an infinite impulse response, such as Shannon/Whittaker’s ideal sinc function, into a finite impulse response filter. Therefore AFAIK, all FIR filters in practice could be described as “apodizing” and as FIR filters can have numerous/various properties depending on how they’re designed, the term “apodizing” does not necessarily tell us anything about what properties the filter may have. For example, it does not tell us whether the filter has a slow or fast roll-off or whether the filter ringing is pre, post or only post (although commonly they are used to shift the ringing more towards post).
Again AFAIK: But a fast linear phase filter can only be a FIR filter and therefore is apodizing. By your own admission you are just guessing what the “foot” is, rather than going on the actual definition of a Window Function (with a zero-value outside the window interval).
