Can someone help me improve the sound in this WAV file using ffmpeg?

[guy24s]

I digitized an old VHS tape and the sound is awful. There's a constant noise and the sound is very low. I would like to improve it by removing or filter out as much of that noise as possible and make the sound overall more clear.

I previously received help on this forum to improve a similar sounding file using ffmpeg. I basically posted a snippet of the sound and someone gave me a command line with presets that improved the sound a lot. I am not very good at using ffmpeg so I tried reusing the presets that were created for the other file, in this new file, but it didn't work well.
I was wondering if someone would be willing to download this 3 minute snippet and see if you can improve the sound in ffmpeg and share the command line code I should use on the full file? Here is a link to the sample: https://easyupload.io/d1eocn
The reason I am not uploading the full file is because it's almost a Gb in size.
I would like the output to be in a 24bit WAV file.

The file may just be too distorted to fix, but maybe someone can give it a go?
I hope someone can help!

[pandy]

Firstly try to increase audio level - now it is ridiculously low - highest part is approx -30dBFS - capturing audio with higher level will improve signal to noise ratio - then you can start to think about proper denoising - this is foundation of signal processing - fully use dynamic of your analog to digital converter - avoid clipping so keep signal with some 6dB headroom but 30dB headroom is simply too much.

With sample you provided this can be a good start:

@set filename=%1

@ffmpeg -y -hide_banner -v 32 -stats -vn -sn -dn -i %filename% -filter:a "firequalizer=gain='if(gte(f,55),0,-INF)+if(lte(f,14500),0,-INF)',volume=volume=30dB,afftdn=nr=30:nf=-30:gs=7:tn=0,dynaudnorm=p=1/sqrt(2):m=100:s=12" -c:a pcm_s24le "%~n1_afftdn.wav"

Btw your sample is 16 bit 48ksps and 2 channel but i have impression that it is mono source so you can try to investigate if this true or not.

[pcspeak]

Hi.
I went in another direction and used Audacity.
This is the result. https://easyupload.io/rxt752

The process takes about 2 minutes, for me.
Yes, I did noise reduction 3 times. It's as quick for me as fiddling with the parameters.
In Audacity:
Highlight the whole track.
Effect/Amplify (OK the defaults that show)
(now, highlight a few seconds of the background hum)
Effect/Noise Reduction/Get Noise Profile
Highlight the whole track
Effect/Noise Reduction/OK

Noise reduction again
(again, highlight a few seconds of the background hum)
Effect/Noise Reduction/Get Noise Profile
Highlight the whole track
Effect/Noise Reduction/OK

Noise reduction again
(again, highlight a few seconds of the background hum)
Effect/Noise Reduction/Get Noise Profile
Highlight the whole track
Effect/Noise Reduction/OK

File//Export/Export as WAV.
Cheers.

[pandy]

@pcspeak - yes, using modern audio editor is way better and easier - normally i use Audition 3.0 where this is relatively straight forward - you just need to find place with pure noise without any signal harmonics - mark relatively short period of noise to be sure no spectral leak from signal, i usually use moderate FFT lenght - 1024 - 2048 as longer FFT may increase risk of spectral leaks. Then select whole clip and apply denoise - probably noise reduction higher than 40dB is not required. To verify that proper settings was used it is better to select in denoise filter option that only noise should be left - if noise doesn't have tones related to signal then everything is set optimally - there should be remaining steady, unmodulated noise without audible chirping modulated by signal itself. This is key for proper denoising.
In ffmpeg this is also possible but require lot of manual tweaking, also capturing noise profile is possible but you need capture your signal with this intention (so provide safe zone without signal) - denoise can be combined to remove noise of capture card (setup) from noise of source.
I also removed frequencies bellow 55Hz (so primary power line frequency - usually VCR don't provide useful signal so low anyway so no loss but this can be of course replaced by group of notch filters to remove only power line hum and harmonics), also anything above 14500Hz is removed as commonly VCR using static head don't provide anything useful above 9000Hz and at 15625Hz there is strong peak of horizontal frequency - such band pass allow to focus on useful signal without being affected by variability of interference's.
In my script example some aspects can be tuned better to get higher quality but as i've wrote - you need many tries to find proper noise threshold. This is should be quite close to provided threshold (tune range within 1..2dB max).

[guy24s]

With sample you provided this can be a good start:

I went in another direction and used Audacity.
This is the result.

Thanks to the both of you!