How to improve low quality mp3 voice?

[qo4on]

I have a low quality mono mp3 voice 24 kHz, 48 kbps CBR, 32 bit float that I use as input for farther audio editing: first convert to pcm_f64le than upsample to 48 kHz, EQ, mixing with other pcm_f64le inputs. There is no noise there.

Is there anything I can do to make it better? I mean add dithering, audio super resolution or anything else?

[davexnet]

Post a sample here to the thread

[qo4on]

Post a sample here to the thread

See the attachment.

[davexnet]

I heard a little sibilance, what "improvement" do you think it needs?

[qo4on]

Ideally, apply a voice super resolution neural net that upsamples it to 48 kHz. Maybe add some dithering, I'm not sure. Reducing that little sibilance can only make it worse.

[davexnet]

Do you have any software in mind that does this? I've never heard of any software that has 'super resolution neural net" for
something as basic as increasing the sample rate. Usually the dither is done when the bit depth is downsized,
for reducing the sample rate, an anti-alias filter is usually applied

I have an old copy of Sound Forge, I haven't tried it, but I wouldn't expect the up-sample to be detrimental to the sound .

[johns0]

The voice sounds good enough for me,any adjustments wouldn't give it any difference in quality.

[qo4on]

Do you have any software in mind that does this?

https://paperswithcode.com/task/audio-super-resolution

any adjustments wouldn't give it any difference in quality.

You're probably right.

[davexnet]

I don't think there is any working software available yet, might have
been interesting to play around with.
Izotope RX are using neural networks in some of their modules,
such as music rebalancing

[qo4on]

Izotope RX

Thank you, that's what I need. Their spectral recovery does exactly what I wrote above.
It turned out to be quite a powerful program. Do you know if they have a tool for removing that "little sibilance"?

[Hoser Rob]

If the voiccve was recorded with a cheap mic, which is almost always the case, no software is going to make it sound like an audiophile recording.

[davexnet]

Izotope RX

Thank you, that's what I need. Their spectral recovery does exactly what I wrote above.
It turned out to be quite a powerful program. Do you know if they have a tool for removing that "little sibilance"?

Their de-ess module

[qo4on]

I used ffmpeg for upsample it to 48 kHz like `aresample=48000:resampler=soxrrecision=33`.
But Izotope has a curve instead. Do you know what settings should I use in this case?

[davexnet]

Izotope know what they're doing but I'd like an explanation on that also.
My understanding was that the anti-aliasing filter was only required when
when the sample rate was lowered, yet the filter in rx doesn't make it clear

[qo4on]

That's what I think too, but I'm not sure. I can't imagine any reason why I should cut off some of the upper frequencies when upsampling. Do you know anything about Pre-ringing?

[davexnet]

That's what I think too, but I'm not sure. I can't imagine any reason why I should cut off some of the upper frequencies when upsampling. Do you know anything about Pre-ringing?

Not really, I'm reaching my limit on this stuff. There are others in the forum who may respond,
might have to be patient, or start a new thread with a more relevant title

[qo4on]

Ok, thanks.