Audio restoration
 

I'll get straight to the point. Couple of days ago, I bumped into some .mp3 files, extracted music from "Freelancer" to be precise, the thing is,
those files were badly compressed and the sound quality was very poor. Generally I would have just ignored the problem, but in this case,
I couldn't just leave these files in the way they are. The main reason for this is the music itself, it was quite nice.

And suddenly I was struck with this looney thought, maybe there is some way to increase the overall quality by some special decompressing techniques and/or restoration software. In Sound Forge 8.0, I've tried to edit the streams in 32-bit resolution, used some DirectX filters like Smooth/Enhance, but unfortunately, results of my experiments were not that consolatory.

I'm not much of a sound engineer so I don't know all the aspects of sound restoration, this is why I'm posting here. It would be very useful to know the basics. And to do something with those source files in my disposal :eagletear:

Here's a couple of samples to give an idea, with what I am dealing with. They are in "untouched" quality - 80 kbps bitrate, and 22 kHz frequency. Exactly in the way they were in the game.

I don't see how it's possible. You can't add in information that's not there. All you can do is to take yet more information away. MP3 uses lossy compression, so when you decompress the files, you don't automatically get back all the data that was discarded.

Here's an interesting link: http://www.team-gizka.org/2005-04-03.html
At the very bottom, there is an "Music Restoration" article, there is also an archive with samples and a text document.

And it has to be said, results of that restore is quite impressive.
This is proving, that it is possible to get good results even from a very limited source.

Impressive...?

What am I not hearing? I don't hear anything in the "restored" version that was not in the original version, personally. I can see that it has been re-encoded at a higher bitrate, and that it is in stereo. What I don't see is how that has improved either of the tracks at all.

After all... you can't just draw additional data from thin air. Quite nice music though. I might try to find a gamerip now, or even buy a copy of the game just for the music.

Hey man,

Interesting thread. This is close to my heart, because a lot of Western VGM from the mid-90's was WAV audio at extremely low quality- downsampled from the masters to 22.050 kHz, 8-bit, and mono (quarter of usual CD stereo quality).

I've recently discovered some tricks with Cool Edit Pro (now Adobe Audition) that allow me to 'remaster' (and I use this term very loosely) the audio a little, and make it semi-stereo instead of rubbish-sounding mono.

This is a very beta example I posted on my site a couple days ago, but check it out anyway:
http://smc.sq7.org/assorted/RESOURCE2288.wav - Original Mono source file (in-game audio)

http://smc.sq7.org/assorted/Lands%20Above.mp3 - 'Remastered' (or messed with :P ) file

Don't expect them to sound hugely different, but I call that an improvement, for the purposes of making a gamerip that's listenable to. Note that I haven't worked on an ending for it yet, so don't mention it ;)


On the KOTOR audio samplers- for fun, I processed it a little bit, tell me what you think:
http://smc.sq7.org/assorted/mus_bat_...20STEREO)2.mp3
(I don't like it so much, but I put all of 5 minutes into it- I really did it to show you can make something 'more stereo'.)

- Spike

@ Soluzar
That restored sample sounded better in my opinion, though not too significant... at least it's more listenable.
Restoration is all about removing imperfections like hiss, noise and sound anomalies. I understand, that it's not possible to replace the missing data.


@ Spikey
I have listened to those samples, and in comparison, that's definitely an improvement. Though there is some hiss in the remastered version.

Also, could you provide more details about the trick you mentioned?
Better yet, could you try your method on those attached files in the first post?
I'm just curious, what is possible to do.

You're right, you can't do that, but you can add transparent sound patterns that enhance the overall experience of what you hear. Sort of like tricking your brain. Think of all the available frequencies in a graphic equalizer for instance, and the endless combinations of settings. Much like VBR is a transparent encode setting, it is indistinguishable from 320CBR, yet it occupies less space.

I've worked with restoration before, it's tricky, but it's worth it. I'll see what I can do, maybe it'll turn out good.

EDIT: Here's an example; normalized, equalized a little bit with a high cut, minor hiss reduction and some pops and cracks eliminated.

Thanks for the effort Killy, I really appreciate it. As for the sample... well, let's just say it didn't turned out that well.

Most recently I've been examining more articles on this matter, with more theories than practical suggestions. If I manage to do some progress in this area, I'm sure to post here right away.

Dude, it won't get any better than that. You're already dealing with a lost case right there. The best thing you can do is just lower the clipping and equalize it - if you ask me that sample is already beyond saving. No sound engineer would work on it anyway, take it or leave it.

I agree with Killy mostly, but that said, I had a go and I think I did a better job (Killy's sounds flat to me). I posted three examples of how I could approach a track.

Usually there's two, the third is a kind of combination of the first two (although not exactly a combination mind you).

Check them out here:

http://smc.sq7.org/assorted/music_bar_li04a.mp3
http://smc.sq7.org/assorted/music_bar_li04b.mp3
http://smc.sq7.org/assorted/music_bar_li04c.mp3

I like 'C' myself. I think it's actually listenable to.

But, I'm much more interested in everyone else's thoughts.

Once again, please note stereo and improved sound was the only goal here, I haven't faded out or ended it or anything.

- Spike

No, all of those are clipping. You haven't really improved it. Panning and/or adding echo isn't going to restore it. :rolleyes: To reduce clipping - you have to flatten it, if you flatten it without keeping it even and re-equalize it to hell, you'll end up cutting and reducing the sample even more.

I created a sample before I posted the one above, although it does indeed sound less flat than what I posted - it has lost far too much quality, thus making the end results worthless to say the least. Going with simple clip reduction, hiss reduction and flattening isn't going to give you a perfect sounding sample - that's not even the point of audio restoration, as far as I'm concerned, but like I said, this sample won't get any better than what it is. The original is far too lossy.

EDIT: The sample is already stereo, Spikey.

Try this.

If you mean clipping as going over 100% volume, 0 dB, then no, it's not clipping. But if you mean parts are clipped, then I can't magically 'unclip it'.

All 3 samples I posted are normalised to 96%, which is what's generally accepted (and what CoolEdit recommends).

I never claimed to have some magic snake oil. I just feel it sounds better than crappy mono-sounding stereo.

I'm not sure what you're getting at. Did you listen to what I posted? Do you like it? Do you think it's a pointless exercise? Etc, etc.


As for it already being stereo, it's not very good stereo, which is the whole point of this thread, I thought so anyway. As opposed to generally-speaking audio restoration, which is beyond the scope of this thread.



Parjay: Sounds worse than the original. Maybe just a smidge more stereo. Did you normalise it? It's pretty tough to compare one volume level with another..

- Spike

I will admit that there's a bit more stereo in some of these samples. I'll also admit that the Freelancer music is fairly nice, and worth a listen. I just don't think the gains outweigh the losses though. When you're recompressing a lossy MP3 to another lossy MP3, you will lose quality.

Isn't something like this more likely to offer the better quality you're looking for? Are there tracks which exist in the game but are not on the OST?

http://music.msn.com/album/?album=39728626

I don't usually do that. I usually work with WAV game audio, so there's only one lot of compression. And with the work I'm doing on Torin's Passage (sample posted earlier, and I've improved on it), there's definitely a big improvement- it's mono audio originally. Hissy mono sucks compared to fake stereo, no matter what anyone says :)

- Spike

Hahahaa, oh wow.

Yes, let's add crappy panning and echo sounds to the original sample and still have it melt our eardrums! Yay! :sarcasm:

I usually work with mono to stereo, not making a stereo piece improved. I probably wouldn't touch that Freelancer piece if it was a soundtrack I'd listen to (which it isn't, not my kind of game music).

I notice you guys haven't commented on each other's samples, either.

And, no doubt you guys are more knowledgeable than me about audio. But I get the feeling I'm the only one who is actually:
a. trying to help the guy
b. trying to have a proper discussion about this

If you have a comment like, "I don't like you what you did because X", or "Person X's sample is better than yours", that's fine- being assholes because you want to or because you can't be assed justfiying your opinion is just weak.

Bottom line- I hate listening to mono game audio or poorly mastered stereo, and if I can do something to improve that (which I believe I'm doing with Torin's Passage), I sure will. If someone has comments or improvements on that, that's fine too. I'm certainly no expert and could use that.

- Spike

Come on man, enough bullshit already. Are you the only one trying to help him? Jesus, want a ribbon to go with that?

Also, I don't feel I have to justify my opinion by telling you a lie - I'm not going to say "Hey Spikey, you did a great job!" when in fact, you did something I would never even consider doing. Being dishonest to you will not help you improve, if that's truly what you want to, it'll only make matters worse.

If you take your time and read through my first response after you posted your three samples, you might even catch a hint of constructive criticism. :rolleyes:

To topic starter - what you are trying to do is mathematically impossible to do. All you can do is adding some filters to increase percieved quality, but you cannot restore data lost by lossy compression, in any way.

Not at all. It's really hard to reply to your posts without being unnecessarily rude though, I wouldn't mind a ribbon for that :)

Again, not at all what I said or implied.

To quote myself:
In other words, I'm hardly asking for false praise.

Add 'doesn't even have the courtesy to read the person he's flaming's posts' to the list of annoying shit you've contributed to this thread.

I did read it, and I appreciate it. A little jargonistic for my tastes, and not very useful, but still, better than what followed.


Dude, I have nothing against you, and am trying to be helpful while getting useful responses for my own benefit (I definitely value honest responses!). I honestly have no idea what's causing your ridiculous outbursts in this thread, and frankly, I hope they stop and we can back to the much more interesting on-topic discussion.

- Spike

Outbursts? Right, well I think that started right about where you actually thought you knew what you were doing and persisted on being the only one with a solution, when you were in fact wrong.

Furthermore, it's interesting that you think of my last post as me flaming you - indeed it's cynical and sarcastic to start with, but I'm not out to scorn you, you can lower your banner now.

This is getting off-topic, so I'm not going to waste any more breath on this guy, I don't visit these forums for mindless fights.

For the record, I class a flame as a post which has no benefit but to cause offence and/or to draw a negative response from someone.


Anyway. I'd like to apologise to the original poster for the direction the thread has taken. If you want to discuss it further, you're welcome to do so at my forums:
http://smc.sq7.org/forum

I mean, I know I'm no expert in digital audio editing, but at least you'll get some civil, quasi-helpful advice :)

- Spike

Just to update the thread on a much more positive note, I've posted some new samples, which can be found here:
http://www.smc.sq7.org/assorted/RESOURCE2288.wav (Original)

http://www.smc.sq7.org/assorted/RESOURCE2288New.mp3 (New edited WAV)

The balance isn't perfect, but I think it's a pretty damn good effort with such a source file.

But don't just take my word for it, take the word of the original composer, who now works for Microsoft (currently working on Flight Sim 2008 music):
"Nice work! Those sound pretty good."


Anyway, I'm eager to hear people's thoughts (yes, even Killy's), and I hope the original poster hasn't run a mile from this thread (although I think Freelancer is a bad case to try this stuff with).

- Spike

I haven't abandoned the thread, I've just decided to take a little break.

Also, here's my current progress with Freelancer restoration. Comments please. Personally I'm really satisfied with results. I will do a little more exploration on working with audio editor's, might even try to do a gamerip. Who knows.

I have this soundtrack (courtesy of bishop743). The point is, it contains almost all music that is written by Visual Music and James Hannigan, but it totally misses out these bar themes, written by Andrew Sega.

Additional Spam:
I'm not trying to return files to they're original state, I'm trying to make the music a bit more listenable. Just for the sake of musical experience (:

In that case I sympathise with your plight. I still say there's nothing significant you can do, but I do sympathise.

I'm sorry, but no, the sample's still clipping, you need to cut it alot more than that. Check Parjay's sample.

If I to cut it a bit more, there will be not much of a music left. Couple of clicks here and there are okay by me (:

Anyway, it's the way original file was and I really don't want to mess with that.

So what's the point of restoring then? Good luck not messing with it if you want to restore it.

I work with audio all the time on a daily basis as part of my work, and Im not new to doing remastering and restoring. Maybe I can help you out?

Can any of you guys briefly explain to me how you restore an audio file like this? I figure you guys use Adobe Audition, but what exactly is it you do you remove clipping and such?

While there are ways to reduce the static sound that results from a clipped waveform, there is no true way to recover the lost information. It's no different than an overexposed photograph: you can darken the while areas, but you will never recover the textural detail; the area can only become a gray mess.

Looking at the waveform, I am fairly disgusted. It reminds me of Californication. A frequency analysis reveals that all frequencies above 10kHz are nonexistent. (Of course, considering the sampling rate). I didn't check the RMS of the file; I assume it must be very high. Tsk tsk! There's not much that can be done here besides basic clip restoration.

music_bar_ku04.flac

I could apply some EQ, a little dehiss and rumble filters and -- if you wanted to -- generate some of the missing frequencies. But it takes many hours.

You can't truly create information that was never there in the first place. This is especially true for complex sounds such as transients and instrument overtones. In this file, the upper frequencies simply do not exist. What would you generate them from? (I'm actually curious here.)

Any way, I don't think the file needs any more filters applied to it than already have been. It's had enough, I think.

I have a piece of software that does it, but it takes forever and its expensive as hell lol.

But what I'm wondering is, how does this software work? How does it "know" what kind of information to add in? For example, let's say we have a 22050Hz waveform audio file of a violin playing. At this sampling rate, frequencies above 11025Hz cannot be reproduced because the sampling frequency was lower than the frequency of the sounds in question. The timbre of a violin is very complex, especially in the upper treble frequencies. Would this software be able to determine the nature of the sound in question and add in the proper harmonic content? How does it determine this? What is it working from? Wouldn't any high frequencies generated by this program only be extrapolations based on the available data?

I'm not trying to be argumentative here, really. I guess I'm just a skeptic at heart. I have no reason to believe that you're lying, but I just can't see how such a program would (or could) produce realistic results.

Ask the people who made the Creative Sound Bullshit X-Fi. It has the same "Restore your 128k mp3s to studio quality with our 24bit crystalizer!!!1!" bullshit.

It's impossible to restore audio this way, no more then you can make a 200x200 px image wallpaper sized. You can increase the percieved quality with filtering and whatnot, but thats only an approximate and may sound (or look) completely different then what it originally was.

I'm kinda wondering why this topic is still open.