# How to upsample voice properly?

I have a low quality 24 kHz mp3 voice and I'd like to upsample it to 48 kHz for farther mixing with another 48 kHz track.

What is the right steepness curve in this case? I'm going to use an ideal low-pass filter with:

Filter steepness=2000
Cutoff shift=1
Pre-ringing=1


but I'm not sure if that would create noise/artifacts or degrade the signal?

Upsample with steep curve:

Reduced steepness:

Low steepness creates artefacts.

## 1 Answer

It barely matters – speech has no significant content close to 12 kHz, so using a relaxed filter works. Reduce the filter steepness a lot; as long as things up to 6 kHz pass through, you won't hear a difference.

• Thank you. Could you explain why an ideal low-pass filter (with maximum steepness) is bad in this case? I'm trying to save as much information as possible. What Pre-ringing is better? Aug 31 at 18:19
• there's no such thing as an ideal low-pass filter – just very steep filters that necessarily become very long filters. And these come with problems such as strong phase distortion and numerical instability (at least in the IIR case), they are computationally undesirable (read: much CPU) and have no positive effect if there's nothing close to the Nyquist frequency to suppress as in your case. Aug 31 at 18:21
• They call it ideal low-pass filter in the docs: s3.amazonaws.com/izotopedownloads/docs/rx8/en/resample/… I updated the post. Aug 31 at 18:51
• they lie. An ideal low-pass filter cannot exist, mathematically. Aug 31 at 18:58
• @dereks phrasing... you're the one asking here, so please be patient with the people trying to help you! And nothing about what Ryan nor I wrote was speculation. you're the one that believes there's implementations of ideal filters... Aug 31 at 20:57