Skip to main content
Topic: A new resampler DSP for foobar2000 (Read 639515 times) previous topic - next topic
0 Members and 3 Guests are viewing this topic.

Re: A new resampler DSP for foobar2000

Reply #50
What's the benefit of real-time up/down sampling? I thought people use these kind of plugins just for converting?
The benefit of real time is you don't have to remaster you entire music library to enjoy a high sample rate, because it's done on the fly.

And I dare say that enjoying a high sample rate resulting from upsampling is purely placebo effect. Then again, I dare say enjoying a high sample rate period is purely placebo effect. You're welcome to get your ears tested, but you'd probably be hard pressed to find any audio labs with equipment that peaks higher than 20 kHz. You could be like me, and find that at the ripe old age of 35, your hearing peaks out at around 14,800 Hz.

Re: A new resampler DSP for foobar2000

Reply #51
What's the benefit of real-time up/down sampling? I thought people use these kind of plugins just for converting?
The benefit of real time is you don't have to remaster you entire music library to enjoy a high sample rate, because it's done on the fly.

And I dare say that enjoying a high sample rate resulting from upsampling is purely placebo effect. Then again, I dare say enjoying a high sample rate period is purely placebo effect. You're welcome to get your ears tested, but you'd probably be hard pressed to find any audio labs with equipment that peaks higher than 20 kHz. You could be like me, and find that at the ripe old age of 35, your hearing peaks out at around 14,800 Hz.

I'm 38 and I peak out at 17.4 kHz but I think that still proves the point.

But hey, you never know...maybe at 47 kHz it vibrates the paint on the wall in such a way that it actually makes it microscopically flake off thus creating the additional "paint dust falling off the wall" noise that makes the music somehow feel warmer or better

 

Re: A new resampler DSP for foobar2000

Reply #52
Yes, but my point was that upsampling will not magically create frequencies that were lost in the original sampling process. It may create distortion, if the resampler is broken, but it won't bring back what was lost.

There may be hope for that sort of thing if one can train a neural network against original audio data and high quality downsampled or low pass filtered to half the original frequency response, using a wide variety of audio content. Similar to how the wonderfully named waifu2x filter works for drawn and sometimes even photographed image content.

It would probably be best trained against and trained to generate frequency domain sample data, using relatively small, overlapping windows, translated back to time domain using Fast Fourier Transform.

Re: A new resampler DSP for foobar2000

Reply #53
waifu2x
I don't understand why image and video upsampling can really improve perceived quality in many cases, for example madVR (NGU works for everything in my opinion) and SVP. Audio upsampling is not funny at all, even for nontransparent sample rates like 16kHz.

Re: A new resampler DSP for foobar2000

Reply #54
waifu2x
I don't understand why image and video upsampling can really improve perceived quality in many cases, for example madVR (NGU works for everything in my opinion) and SVP. Audio upsampling is not funny at all, even for nontransparent sample rates like 16kHz.

You can perfectly represent a bandlimited signal by way of discrete samples if you respect the Nyquist rate. When you are dealing with an already sampled signal it also means it has already been bandlimited (properly or not, that's another question), so it has what it has. What was there before the signal was bandlimited is gone for good. Up sampling will just attempt to represent the same bandlimited signal (or very close) with more samples than necesary.

Now, these techniques you mention are NOT resampling. They are not sampling an already existing signal which is reconstructed from the existing samples. Instead, they are GUESSING what might have been there before the bandlimiting process, or, more properly said, they attempt to fill the gaps with what a human observer would expect to perceive, or, at least, something that doesn't seem too much out of place as to be perceived as broken or unpleasant.

Resampling is straight math, while these reconstructive processes are more of a creative process. Not saying they have no merit or scientific basis, nor that they don't produce satisfactory results, but their aim isn't, and can't possibly be, taking an already sampled signal (and thus, inherently bandlimited, properly or not), as input and somehow roll time back and sample the original non-bandlimited signal at a higher sampling rate.
Despite all the math involved, it's still inherently an art of guesswork. The programmers try codify with a programming language what an impossibly good human reconstructor/restaurator would do if it was practical to spend years adding the perceived missing content to bandlimited signals

It's a VERY different approach and principle.

Re: A new resampler DSP for foobar2000

Reply #55
Hi Developer!
Can you update the plugin for  352.8 kHz and 384 kHz frequencies, please.

Re: A new resampler DSP for foobar2000

Reply #56
Can you update the plugin for  352.8 kHz and 384 kHz frequencies, please.
Plugin already supports 352.8 kHz and 384 kHz (it supports up to 2822400 Hz). Just enter needed frequency manually using digits on keyboard.

Re: A new resampler DSP for foobar2000

Reply #57
Thanks for the update. Love this component.

 
SimplePortal 1.0.0 RC1 © 2008-2018