Getting started

[mikesndbs]

Hi

Have been playing with Milllennium and Forensics live 6 demo plus xxxxxxxxx on a particular sound clip received from a friend.
Its a right mess but there is a male voice within.
I have got so far with it but need advice on what to try. This link is for the sound clip I am trying to clean up.
So far all I can hear is the word 'everybody'
All I have done to this clip is to normalise as the original was very low volume.
www.box.net/public/43g7h3jl04

Is there a filter or feature to firstly boost voice audio?

What filers and what order should I run?

Hope someone can help.

Cheers

Mike

[mikesndbs]

List dead?

IS this list still active????

[DougMac]

It's alive and well.

Mike,

We're out here. I went to your link and found a zip file. I'm hesitant to open a zip file from an unknown source, so I didn't check it out. Others here may have the same reservation.

Doug

[mikesndbs]

OH!

Its quite safe I assure you, just a wav file zipped up to save some file size.
Hang on I think I have a mp3 you can go to....


Yes here you go:
http://www.savefile.com/files/31170

This is my best attempt to clean the file up and in fact is taken from 29 seconds in.

The other file is the full clip and has much more in it.

Cheers

Mike

[cromag]

Mike --

Not ignoring you, just wasn't sure I could help. From your description, your project is the exact opposite of what I usually try to do, so there isn't much useful I can tell you.

I'm somewhat familiar with DCart32 (which I think predates Millenium) and DC-6, so I'm not sure what commonalities my software has with yours, but do you have low-pass, high-pass, or band-pass filters? I would assume they would help isolate voice/speech frequencies.

Finally, I suspect most folks on the board who try to do this kind of work are using the Forensics software -- specially aimed at this kind of problem, but more expensive.

[Craig Maier]

As was stated, there is a special program for extracting speech from noise wherein the speech level is near the noise floor.

But,

You could try using the bandpass filter. There is a preset in there called Speech Filter, as I recall. This should help the situation.

Then, you could take a noise fingerprint between words and use that as the "Sample Noise" for the Continuous Noise Filter. That will help you get rid of some more of the "In-Band" noise left behind after running the speech filter.

After that, consider running the Median Filter to improve the clarity of the sibilant and consonant sounds. Experiment with the number of samples used until an improvement is realized.

I hope that this helps some.

[mikesndbs]

Thanks

I have the demo of DC Live Forensics 6!
Even with this I have had little more success.
Its all the mp3 artifacts etc.

Re price! yes, I don't know how many they are going to sell at $1300!
For ones and zeros that is a lot of money and way over my price range

[Craig Maier]

If you have the Forensics Demo, try the AFDF (Adaptive Frequency Domain Filter) found under the Continuous Noise Filter. AFDF is one of the "CNF Modes" that you can choose. Just adjust the attenuation control for the optimal results.

As for the price, it is a lot of money for a bunch of ones and zeroes. But then again, the cost of a BMW is a lot of money to pay for a bucket of iron.

We have competitors in the Forensics Audio Field selling ones and zeroes (that do about the same thing our program does) for $25,000 bucks a copy. And those organizations that need these kinds of tools are happy to pay the price for what they can do for them.

And then, it takes a 3 day training course for folks to get on board with the Forensics product. It is not a cakewalk; people are not born knowing how to use it. Extracting signals at the noise floor from the noise floor is difficult and not usually tackled by the general public.

[mikesndbs]

HI

It was that filter that I used to get to where the clip is now.
Which is not a lot of good :-)

I will try some more.

OK, I can see I struck a nerve re the price so I better set my stall out properly.

$1300 for a audio programme to a individual hobbyist is extremely high.
I won't be making my living from it and will not even use it every day.
Ok I accept that its not aimed at the public, but from the demo it seems a bit more capable than Millennium and so I was tempted.
When you look at DC Millennium with its far more affordable price tag, DC Live6 Forensics Demo does not appear to do that much more of course this could be due to my lack of understanding.

Of course companies and organizations may be able to afford it and if they can then why shouldn't you charge what you can. My point was that for a individual it would just not be possible, unless of course I win that lottery Saturday

If your competitors charge even more then this simply puts their product in the realms of silly money while yours is just very expensive.
Again, this comparison may be a selling point to a large company, but not me who has to budget for the family out of a limited income.
Hope this explains my point better I was not having a pop at DC in any way.
Cheers

Mike

[DougMac]

Mp3

Thanks


Its all the mp3 artifacts etc.

(

Was the file you're trying to clean an MP3 file at some point? Since MP3 is a lossy format, a lot of information that the filters can use has been lost. I think this will have an effect on your results.

Craig, is this correct and could you give a little more technical explanation?

Doug

[Craig Maier]

Yes; DougMac is correct. Once the information is gone, that is it; almost impossible to interpolate it back into the file. Most of the time, the Forensics guys are working with bad files, but not ones using lossy compression like MP3.

Just quickly about Forensics. Mostly, that product is purchased by governments and government agencies and the military (FBI, CIA, DEA, NSA, and all the rest of those alphabet soup names). The priviate individuals who purchase it do Forensics Audio for a living, not as a casual endeavour, so the cost is part of their toolset for their chosen career field. So, it just seems to be expensive, but it is really very inexpensive.

ps mikesndbs- love your avatar!

[mikesndbs]

Hi

I think it probably was, I was sent in as wav but from the artifacts I would guess it was a compressed mp3.

Cheers

Mike

[Craig Maier]

I think that if you compress a file that is mostly noise and has a small signal on it, you end up with almost pure noise, with the signal being stripped away by the compression algorithm.

[mikesndbs]

Ok thanks everyone!

Looks like I have done the best I can with it already then.

Thanks re avatar nice pentacle

I am looking at some of my sound clips now and will try both progs on them.
These start off a 11025 hz wav files and I convert them to mp3 for distribution.
So I have the originals to play with

Cheers

Mike

p.s. ok I accept about the price, think I will buy Millennium a bit later, seems good.

[Craig Maier]

Quoting mikesndbs:

"Thanks re avatar nice pentacle"

-----------------------------------------

LOL - - - that is not a Pentacle. Pentacles have five sides and are sometimes associated with the practice of witchcraft. That symbol is a Mogan David (Hebrew for Shield of David) or Star of David, usually associated with Judaism. Geometrically speaking, it could be referred to as a hexogram but not a pentacle or pentagram.

-----------------------------------------

A bit more on lossy compression like MP3's. Signals below certain noise threshold values vs frequency are thrown out in the frequency domain in lossy compression schemes. These signals are assumed to be masked by the louder signal content and thus not completely necessary to keep unless very high quality audio is the order of the day. That is probably why you are having such a difficult time with the file that you are working on - - - the signal component probably has been thrown out by the lossy compression algorithm.