Forums

removing Background Noise

Started by Ronen Yacov June 10, 2007
Hi Everybody,
I am working on removing Background Noise from speech  Since I don't
have access to the reference noise and the only thing  at hand is
speech+noise record, that's why it is a bit tricky. Can anyone help me
in this regard.
Many Thanks


Ronen Yacov wrote:

> Hi Everybody, > I am working on removing Background Noise from speech Since I don't > have access to the reference noise and the only thing at hand is > speech+noise record, that's why it is a bit tricky. Can anyone help me > in this regard.
You can build an adaptive filter based on the estimation of the parameters of the speech. This can give a substantial improvement if the input SNR is good enough. However if the SNR is too low, not much can be done. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
On Jun 11, 5:22 am, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> Ronen Yacov wrote: > > Hi Everybody, > > I am working on removing Background Noise from speech Since I don't > > have access to the reference noise and the only thing at hand is > > speech+noise record, that's why it is a bit tricky. Can anyone help me > > in this regard. > > You can build an adaptive filter based on the estimation of the > parameters of the speech. This can give a substantial improvement if the > input SNR is good enough. However if the SNR is too low, not much can be > done. > > Vladimir Vassilevsky > > DSP and Mixed Signal Design Consultant > > http://www.abvolt.com
There is a lot that can be done - a hugh literature in fact. Much of it relies on Voice Activity detectors and the estimation of 'noise- alone' statistics or acoustic transfer functions. There is adaptive acoustic beamforming and so on. It is a big area.
gyansorova@gmail.com wrote:
> On Jun 11, 5:22 am, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: >> Ronen Yacov wrote: >>> Hi Everybody, >>> I am working on removing Background Noise from speech Since I don't >>> have access to the reference noise and the only thing at hand is >>> speech+noise record, that's why it is a bit tricky. Can anyone help me >>> in this regard. >> You can build an adaptive filter based on the estimation of the >> parameters of the speech. This can give a substantial improvement if the >> input SNR is good enough. However if the SNR is too low, not much can be >> done. >> >> Vladimir Vassilevsky >> >> DSP and Mixed Signal Design Consultant >> >> http://www.abvolt.com > > > There is a lot that can be done - a hugh literature in fact. Much of > it relies on Voice Activity detectors and the estimation of 'noise- > alone' statistics or acoustic transfer functions. There is adaptive > acoustic beamforming and so on. It is a big area.
How does beam forming apply to a signal that's already been acquired? Jerry -- Engineering is the art of making what you want from things you can get. &macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;
gyansorova@gmail.com wrote:
> On Jun 11, 5:22 am, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: >> Ronen Yacov wrote: >>> Hi Everybody, >>> I am working on removing Background Noise from speech Since I don't >>> have access to the reference noise and the only thing at hand is >>> speech+noise record, that's why it is a bit tricky. Can anyone help me >>> in this regard. >> You can build an adaptive filter based on the estimation of the >> parameters of the speech. This can give a substantial improvement if the >> input SNR is good enough. However if the SNR is too low, not much can be >> done. >> >> Vladimir Vassilevsky >> >> DSP and Mixed Signal Design Consultant >> >> http://www.abvolt.com > > > There is a lot that can be done - a hugh literature in fact. Much of > it relies on Voice Activity detectors and the estimation of 'noise- > alone' statistics or acoustic transfer functions. There is adaptive > acoustic beamforming and so on. It is a big area.
Well, you are both right really. There is a lot that can be done, but very little of it is useful. There are many implementations of grand schemes for serious noise reduction. However, I've never heard one where the final result was an actual improvement. Sure there is less background noise, but the side effects are awful. Steve
On Jun 11, 12:04 pm, Jerry Avins <j...@ieee.org> wrote:
> gyansor...@gmail.com wrote: > > On Jun 11, 5:22 am, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > > wrote: > >> Ronen Yacov wrote: > >>> Hi Everybody, > >>> I am working on removing Background Noise from speech Since I don't > >>> have access to the reference noise and the only thing at hand is > >>> speech+noise record, that's why it is a bit tricky. Can anyone help me > >>> in this regard. > >> You can build an adaptive filter based on the estimation of the > >> parameters of the speech. This can give a substantial improvement if t=
he
> >> input SNR is good enough. However if the SNR is too low, not much can =
be
> >> done. > > >> Vladimir Vassilevsky > > >> DSP and Mixed Signal Design Consultant > > >>http://www.abvolt.com > > > There is a lot that can be done - a hugh literature in fact. Much of > > it relies on Voice Activity detectors and the estimation of 'noise- > > alone' statistics or acoustic transfer functions. There is adaptive > > acoustic beamforming and so on. It is a big area. > > How does beam forming apply to a signal that's already been acquired? > > Jerry > -- > Engineering is the art of making what you want from things you can get. > =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=
=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF Ha - I didn't read the question properly! Maybe spectral subtraction in that case. Try Cool-Edit Pro for instance. If it is severe non- stationary noise then you've had it. Wang
Check out Compandnet's NoiseMuter, it's an excellent professional 
solution for adaptive noise cancellation:
http://www.compandent.com/products_anc.htm


Ronen Yacov wrote:
> Hi Everybody, > I am working on removing Background Noise from speech Since I don't > have access to the reference noise and the only thing at hand is > speech+noise record, that's why it is a bit tricky. Can anyone help me > in this regard. > Many Thanks >
>Hi Everybody, >I am working on removing Background Noise from speech Since I don't >have access to the reference noise and the only thing at hand is >speech+noise record, that's why it is a bit tricky. Can anyone help me >in this regard. >Many Thanks > >
Hi, a short audio sample of this recording would be helpful. You could just try to use certain type EQ for this. I normally use AiXcoustic Creations Electri-Q (posihfopit edition) - http://www.aixcoustic.com/index.php/Electri-Q-posihfopit/30/0/ as a VST plug-in in some software which allows to re-process the audio data w/ EQ or just live (i.e. route input through EQ). Electri-Q has all basic filter types (Peak, HighShelf, LowShelf, Highpass, Lowpass, Bandpass, Notch) and 30+ special filters to choose from (the Commercial (Full) version of Electri-Q has all basic filter types (Peak, HighShelf, LowShelf, Highpass, Lowpass, Bandpass, Notch) plus 80+ special filters). But as no audio exaple available, I can't comment if this would be suitable method for your problem. jtp
>>Since I don't >>have access to the reference noise
Forgot to mention that there are plenty of analyzer software/plugins available you can use to find those main frequencies for noise (if there are these). As for couple examples, Standalone: Har-Bal VST plug-in: Voxengo SPAN jtp

gyansorova@gmail.com wrote:

>>>I am working on removing Background Noise from speech Since I don't >>>have access to the reference noise and the only thing at hand is >>>speech+noise record, that's why it is a bit tricky. Can anyone help me >>>in this regard. >> >>You can build an adaptive filter based on the estimation of the >>parameters of the speech. This can give a substantial improvement if the >>input SNR is good enough. However if the SNR is too low, not much can be >>done. >> > > There is a lot that can be done - a hugh literature in fact.
:)))))) If there are many medicines for a disease, then the disease is uncurable.
> Much of > it relies on Voice Activity detectors and the estimation of 'noise- > alone' statistics or acoustic transfer functions.
This works only if the SNR is good enough already.
> There is adaptive > acoustic beamforming and so on.
This requires the multiple channels. See OP's post.
> It is a big area.
It all depends on the input SNR. The better is the SNR, the better it works. However if the input signal is severely garbled by noise, so it it difficult to recognize the speech, then not much can be done. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com