I am looking for a C/C++ source code to remove "background noise" from
continuous dynamic speech input from a microphone connected to a
computer's sound card running Linux OS.

This "silence removal" process is also known as:
1. adaptive digital noise filter
2. broadband noise reduction
3. stationary noise elimination

As I understand it the background noise is removed dynamically from
the incomming speech stream making the speech audio much cleaner to
then input to a speech recogonition program called Sphinx.

The silence is collected and then used adaptively against the speech
input to remove the background noise.

I am currently doing this using the CoolEdit 2000 software package I
bought. CoolEdit 2000 does this under their Transform/Noise
Reduction/Noise Reduction dialog box. First you select some speech
free background sound, do "get Profile from Selection", Ok. Then you
select the speech segment and appply the Noise Reduction to the
speech. Away goes the background noise and out pops the cleaned up and
clear speech sounds and waves.

The code/algorithm should be specificaly for speech processing as far
as I know. If other stuff works that would be good to. I need to
integrate this into my own work/program and can not use a standalone
program.

If someone can supply binaries but is willing to support them for very
cheap/free they will be credited in all my software which is GPL and
OpenSource lecense. The code will need to be adapted to give the best
speech recognition results using various codecs, grammers and models
but will be kept OpenSource and GPL available.

---

Request for Question Clarification by jbf777-ga on 12 May 2003 09:03 PDT

Hello -

Have you had a chance to look at any of those options?

jbf777-ga
GA Researcher

---

Clarification of Question by vanderdecken-ga on 12 May 2003 12:22 PDT

I have looked at some of them. It takes a little while to evaluate
them as this is complicated.

What I need is a continous real-time adaptive "silence removal" (see
above) for Linux/Unix in C/C++ (Windows would work so long as it does
not use windows OS calls and does not require Windows OS to run. )I
have seen a couple of these systems before and they do static
reduction but we need dynamic in that as the speech come in the noise
is removed and then we can pass it on to the Sphinx recognizer.

If Sphinx is considered a poor recognizer then we are willing to go
with another if it has a good noise filter.

I would like a royalty free code base, GPL/OpenSource. This
recognition is for home use and a large, more then a couple of
dollars, is too great a royalty on a per household basis. Lots of
companies do this but they want $100 a house/seat license and my
product only costs $500 with the equipment costs.

I am going to contribute my work back to the GPL/OpenSource community
if I can get this to work, or not work.

Race V. 
Research Buyer

---

Request for Question Clarification by jbf777-ga on 21 May 2003 20:13 PDT

Hi -

Anything new?

jbf777-ga
GA Researcher

---

Clarification of Question by vanderdecken-ga on 22 May 2003 06:26 PDT

I am sorry to report that I have not had time to think about this yet.
I have been swamped with other work.

I will try to get to it this weekend, Memorial Day weekend.

Race

---

Request for Question Clarification by jbf777-ga on 22 May 2003 07:23 PDT

That's no problem...  just to let you know, though, your question
expires on June 8.

Check out this page as well:

http://www.bright.net/~dlphilp/linux_csound.html

"dnoise" is included with this; don't know if it will do it, however.

Hello -

I've posted this as a comment, because I'm not sure if these will do
it for you.  Please let me know if any of them will.

- Gnome Wave Cleaner from Redhawk.org
http://gwc.sourceforge.net/gwc_main.html
"The goals are simple -- denoise, dehiss and amplify audio files. With
the use of libsndfile, you can now do this on a multitude of audio
formats, wav, au, aiff, ... "

However, it requires Gnome.


- Denoiser or DNR
http://www.sci.fi/~mjkoskin/
Both may contain source in the tar files, but I don't know for sure. 
I sent an email out to the author.


Another worth looking into, but don't know if it's C/C++ or portable
to Linux:

- SpeechPro
www.speechpro.com
Clear Voice algorithms are available as SDK, DSP, ActiveX or PC
software. We can grant the right to manufacture, use and sell products
incorporating Clear Voice algorithms.  Clear Voice may be licensed for
use in software, hardware or integrated circuits (ICs) for
communication applications. The form of the transfer may include
software for PC or DSP, DLL, source code, hardware, devices, etc.

With regard to speechpro, check out this .PDF.  It says they license
to other OS's upon request:

ftp://speechpro.com/pub/leaflets/clearvoice.pdf