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: Upsampling 44.1->88.2, 48->96  (Read 3819 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Upsampling 44.1->88.2, 48->96

I understand that my dac-amp can probably benefit from higher sampled input. I don't believe my device upsamples my audio from 44.1 to 88.2.

DBPoweramp (seems faithfully) to produce FLAC. Is there a plug-in or software to which I can boost this 16/44.1 to 16/88.2, or even 24/88.2 without introducing bad data. Some audio is 48khz, looking for a way to extrapolate 96 from it too.

Re: Upsampling 44.1->88.2, 48->96

Reply #1
There is no point to this.

 

Re: Upsampling 44.1->88.2, 48->96

Reply #2
Indeed, it will not improve the sound, and in fact will add a small amount of noise. Plus it will make your files much larger.

Re: Upsampling 44.1->88.2, 48->96

Reply #3
Adobe Audition

Re: Upsampling 44.1->88.2, 48->96

Reply #4
Again, no amount of upsampling will improve the audio quality, unless the hardware is incredibly terrible at playing lower sample rates, which is highly unlikely.

There is no grand extrapolator which can intelligently reproduce information that never existed in the original signal, or was lost due to recording at a lower sample rate or downsampling. A neural network could be trained to try to replicate lost information, using original and downsampled audio, but this sort of replication is not totally reliable, and has only really been trained and demonstrated on visual data, not really on audio. And considering how slow it is on just images, it would be insane to expect it to be usable on massive quantities of samples like you have with audio. Well, not entirely. Figure 12 million samples in one dimension instead of the multiple of two dimensions. Still, don't expect miracles like this to actually produce anything you can hear, if your original signal is already at least Redbook quality.

Re: Upsampling 44.1->88.2, 48->96

Reply #5
A neural network could be trained to try to replicate lost information, using original and downsampled audio, but this sort of replication is not totally reliable, and has only really been trained and demonstrated on visual data, not really on audio.
Yeah, attempting to marry two waifus can cause you a lot of trouble, especially waifus of an audiophile. :P

Re: Upsampling 44.1->88.2, 48->96

Reply #6
A neural network could be trained to try to replicate lost information, using original and downsampled audio, but this sort of replication is not totally reliable, and has only really been trained and demonstrated on visual data, not really on audio.
Yeah, attempting to marry two waifus can cause you a lot of trouble, especially waifus of an audiophile. :P


Why, yes, you did get the hint that I was referring to waifu2x, a neural network that was trained against anime and manga style artwork, both original and downscaled versions of the original, to give it a broad set of knowledge about what some combinations of pixels would look like when downscaled, so it may attempt to replicate lost information and upscale images. It actually works quite well, even on some photographs.

Audio upscaling, I'm not sure if it would work better to train and work a neural network against time domain data, or frequency domain. Someone will probably attempt it some day, and possibly also implement it for time and pitch shifting, but I don't expect it to be miraculous. Considering how much processing power goes into image scaling, and the sample rates of audio, it may be possible to use GPU or other heavy OpenCL hardware to process audio in real time, assuming you're not already starting from Ludicrous Sample Rates.