Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Why only near nyquist sampling is done in audio? (Read 4863 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Why only near nyquist sampling is done in audio?

Many of the voice codecs i see cut it very close when it comes to sampling rate vs the bandwidth they code upto. For example a 8khz coder requires input audio sampled at 16khz. This is exactly at nyquist. Why dont we have a margin here( in other words why are we assuming perfect reconstruction)?

Why only near nyquist sampling is done in audio?

Reply #1 –
Quote from: fullband on
Many of the voice codecs i see cut it very close when it comes to sampling rate vs the bandwidth they code upto. For example a 8khz coder requires input audio sampled at 16khz. This is exactly at nyquist. Why dont we have a margin here( in other words why are we assuming perfect reconstruction)?


The "bandwidth" in this case probably just refers to half the sampling rate, not the actual output of the decoder.

Why only near nyquist sampling is done in audio?

Reply #2 –
also the main purpose of a voice codec is lightweigth.

Why only near nyquist sampling is done in audio?

Reply #3 –
Because of efficiency.

Sure, you could store <8 kHz low-pass filtered voice at 48 or 96 kHz, but all you will store in those additional samples is redundant information / noise. Even if a codec didn't store >8 kHz content but still output a higher sampling rate all it would do is interpolation.


Also, there is some room for the filters to work. For example, telephone systems using a 8 kHz sampling rate actually have a usable bandwidth of roughly 3.2 kHz, not 4 kHz.
"I hear it when I see it."

Why only near nyquist sampling is done in audio?

Reply #4 –
Quote from: fullband on
Many of the voice codecs i see cut it very close when it comes to sampling rate vs the bandwidth they code upto. For example a 8khz coder requires input audio sampled at 16khz. This is exactly at nyquist. Why dont we have a margin here( in other words why are we assuming perfect reconstruction)?


We presume near-perfect reconstruction because most experts believe that with digital filters and modern design techniques we can produce near-perfect reconstruction at a very reasonable cost.

If you think otherwise, download one of the software DBT comparators, and some good near-SOTA reconstruction filters, and prove that they are deficient from a sound quality viewpoint. In this case the main main criteria are probably intelligibility and usability.

Why only near nyquist sampling is done in audio?

Reply #5 –
Quote from: Arnold B. Krueger on
Quote from: fullband on
Many of the voice codecs i see cut it very close when it comes to sampling rate vs the bandwidth they code upto. For example a 8khz coder requires input audio sampled at 16khz. This is exactly at nyquist. Why dont we have a margin here( in other words why are we assuming perfect reconstruction)?


We presume near-perfect reconstruction because most experts believe that with digital filters and modern design techniques we can produce near-perfect reconstruction at a very reasonable cost.

If you think otherwise, download one of the software DBT comparators, and some good near-SOTA reconstruction filters, and prove that they are deficient from a sound quality viewpoint. In this case the main main criteria are probably intelligibility and usability.


Thanks for that. I guess that is a good experiment i can do.

Why only near nyquist sampling is done in audio?

Reply #6 –
Audio is about the only discipline where near-perfect anti-aliasing and anti-imaging/reconstruction are feasible and thus are done. This is not the case in video/imaging, where such filtering is rudimentary or inexistent, and in metrology, where traditionally one has no filters and compensates with a very wide guard band. Getting your digital education from a non-audio centric source may indeed cause some confusion.

I don't know about telecom. Not meaning the baseband, but the intermediate modulations. I can imagine that lack of filtering there spoils the proceedings. But neither can I imagine near-perfect Sinc filtering in the receiver of a mobile phone.