HFSoundRestore: restore high frequenzies in your audio files

[chackl]

Hello!

I'll release now the first alpha of my new tool HFSoundRestore.

EXE & FSM Schema Download:
https://www.dropbox.com/s/jdqknfqh0l5g7 ... 1.zip?dl=0
You need to extract all files ino a directory where you can read and write!

sample-screenshot.res.png (274.05 KiB) Viewed 12868 times

For details how the process works see here:
viewtopic.php?f=2&t=36444

What maybe would be some examples:

• Using CMD-lets / CLI tools for processing: SOX / Rubberband
• Loading long wave-files to view the wave by resampling them to samplerate 1000Hz
• Doing a FFT-specrtrum in a full song with 10-1000 hann-windows
• Storing the hann windows and restore this file to get restored FFT
• my own ruby class "Frequenzgang" that is able to do some operations (Durty)
• Ruby Functions freuquenzies Hz to lin and a Ruby FFT window
• Autodetection for fall-off in FFT and calculationg the db-fall of harmonics.

If you download...
... you agree that you know that this project uses SOX and rubberband as audio libs internal
... you agree that you do not sell restored audio files

And i am sorry that the schematic is that durty at the moment. I'll clean that in future.

Regards, C.Hackl

[wlangfor@uoguelph.ca]

Nice project . You have a lot of skill

[TheAudiophileDutchman]

A testdrive as promised. Here's TAD's first impression of HFSoundRestore - alpha 1

Disclaimer: your mileage may vary, this is just a snapshot and my view will probably change in due course as the app is still in development.

A fascinating idea as presented and the elaborate implementation made me quite curious. So, let’s start with what my ears are telling me: I do hear extra space and definition in many classical recordings, sometimes emphasizing unwanted details of the recording, like paper sizzling of sheet music and other non-musical noises of the orchestra.

By the way HFSoundRestore seems to normalize soundfiles, as I noted that loudness of some classical pieces got louder while pop songs typically got softer. I don’t value perceived sound quality increases by increased loudness, but a blind-test might. Basically, this may hinder the natural track to track volume flow of CD mastering (who is still listening to complete albums instead of streaming single tracks these days?).

It is stated that the goal of HFSoundRestore is just to get a better sound (hence not to keep or restore the original master recording). Does that work for the two scenarios mentioned?

* Better than low quality MP3?
I.m.h.o. yes, a significant improvement in many cases.

* Better if you do an up-sampling from 44,1kHz to 96kHz?
It depends. I found that, especially in recordings which do not contain a lot of reverb, the characteristic sound of e.g. an acoustic grand or drum sound is altered slightly… In solo piano recordings I found examples in which the perceived timbre of the instrument changed, like listening to a different piano. I would not say that the sound quality itself is worse, just that it can drift further away from the original.

Some additional background explanation is given in the introduction of HFSoundRestore: "..music Instruments operate most from 20Hz to 4200 Hz. Every frequency generated after 4200 Hz are harmonics and follow the scheme of the falling Fourier series."
This is probably a ‘close enough for jazz’ approximation related to harmonics, but what about high-frequency non-harmonic transients, especially in the attack-phase of musical instruments? I must admit that, so far, I have not yet performed many manual tweaks to the settings, because the conversion time is quite long for doing iterative testing and it would also defeat the main purpose (holy grail) for my personal use: automatic batch conversion of 44.1kHz CD-material / 48kHz FLAC files to Hi-Res.

For hi-res I dislike the upsampling by way of using non-integer ratios. I prefer upsampling 44.1kHz to 88.2 or 176.4kHz and 48kHz to 96 or 192kHz.

Note it is great that FLAC read/write functionality is included.

Keep up the good work!!!

[chackl]

Hi Thank you for your review!

At the moment i am traveling to several places with good sound-systems and trying to compare.
I did notice some points and i am researching now by splitzting the full track into FFT optimice it and do an iFFT operation. It realy sound more interesting cause the interpolation is sinusial correct and there are no frequenzies added throu the interpolation. The HFSoundRestorer misses this now at the moment.

I noted some points i wanted you to write:

I do hear extra space and definition in many classical recordings, sometimes emphasizing unwanted details of the recording, like paper sizzling of sheet music and other non-musical noises of the orchestra.

This was one of my fist thought when i head jazz files with background noise. Yes it seems to get back a lot of details.

By the way HFSoundRestore seems to normalize soundfiles

Yes it does a normalization to -1db gain peak. This is that you have space for further harmonics and the track does not clip. It also will garantie that you have no quality losse - as you may think that half loudness is 1 bit you wuld loose. I have it on my todo-list make something better - But for CD Collections in a continuous mix it is maybe better to restore the whole mix handled with a cue-file.

In solo piano recordings I found examples in which the perceived timbre of the instrument changed, like listening to a different piano

could you send me some samples? Maybe the automatic parameter detection failed that bad. In normal case this should not happen - but if you have a piano without reverb in most cases is nothing / few to restore so you would notice no different - so i fink the parameters failed here.

This is probably a ‘close enough for jazz’ approximation related to harmonics, but what about high-frequency non-harmonic transients, especially in the attack-phase of musical instruments?

Well it seems that this is working for any genre - i would say it is a physical law like newtons gravity. If you look closer to Fourier calculations you see always the same result in high frequenzies - a linear fall down of the spectrum (Mostly 3-12db/oct in my tests) IF they have no equelizer mastering. (that destrois the sound in my opinion)
If you force tho have no linear falldown it sound more like a quite soft cutoff filter.

automatic batch conversion of 44.1kHz CD-material / 48kHz FLAC files to Hi-Res.

Yes i am working on that - but it will be maybe more different - i am eperimenting with FFTs now - and FFT spectrums can be calculated automaticaly.

For hi-res I dislike the upsampling by way of using non-integer ratios. I prefer upsampling 44.1kHz to 88.2 or 176.4kHz and 48kHz to 96 or 192kHz.

Yes you are correct - this was one of the fist things i saw when i started FFT processing.
FFT only works with integer upsampling - DFT can do more but takes to long (arround 2 Days for Queens Bohemian Rhapsody) - But i am working on a solution that maybe works. I have no samples at the moment because doing a full track in FFT and iFFT with multithreading in Flowstone now is quite impossible.

Note it is great that FLAC read/write functionality is included.

I could program any format that sox (http://sox.sourceforge.net/) or ffmpg is able to encode.
I also want to add some output things:
- Samplerate 44100*2, *4, *8 and 48000*2, *4
- 32bit float (It is not limmited to 0db like 24bit or 16bit) so i do not have to think about clipping the signal.
- 64bit float to store High-Res Masterings (As in the new version i whant to place the FFT interpolation)

Thanks for taking you time - i have a lot of intut now that i can grab
C.Hackl

[tulamide]

Normalizing shouldn't affect the sound very much, as the dynamics keep the same. Therefore it also doesn't change loudness, just the amplitude is oriented at a max value. And at some point you simply have to alter the amplitude to prevent clipping/distortion when additional amplitudes are added. I don't see a way around it.