You're welcome!

You're welcome!

You're welcome!!

Alex, I'm not completely sure what your question is asking. I think the short answer is that sampling and reconstruction use filters that 'smear' over time and also, often in quantization. The longer answer follows: According to Shannon's reconstruction theorem, periodic samples can reconstruct the audio waveform (providing the sampling rate exceeds the highest signal frequency by a factor of 2 or more). For this reason, PCM systems use low-pass bandwidth-limiting in the A/D (to avoid aliasing) and in the D/A (to minimise or eliminate upward-images). The reconstruction filter, as you say, smooths the waveform between the sample instants. However, the filters at each end do smear the signal over time. Slide #41 @ 44:16 illustrates this and shows the 'ideal' low-pass filter. Of course, there is no such thing as a perfect filter and in practice, they tend to be different as you can see in the lower-right of slide #42. These linear-phase filters turn out to be poorly matched to the human listener because they are non-causal - content comes out before the event. Put another way, the envelope of the reproduced sound (which we also detect) is changed to have pre-responses that cause 'brightness' and degradation. Later on in the talk (slide #81 @ 1:13:30), I show that the kernel does not have to be a 'sinc' and in fact, a causal filter (such as the one in the middle) is a better match if the sample rate is higher than 48 kHz. We can end up with an end-end impulse response that is more natural (slide #84 @1:14:30) where there are no pre-responses and a very short post-response. As I also showed in slide #51 (@ 53:23), that real-world converters use cascaded DSP to perform these operations and, unless the part is perfectly designed (and that is rare) inadequate dithering can result in correlated noise that follows the filter (slide 57).

🤣

I was ready for a good squeeze. Never have I been so disappointed.

iykyk