I have seen quite a lot of definitions for SNR for speech + Noise signals. Segmented SNR, noise-reduction ratio etc. Is there a way of automating this or do you have to manually work this out with an audio editor. eg If I have a sentence lasting say 5 secs with additive noise then I would have to look between the words to get the noise. Anotehr thing is that in a real environment (as opoosed to a simulation) I can only measure Signal + Noise and Noise alone but the Noise won't always be the same noise since it will change continuoulsy for the non- stationary case. K.
SNR of Speech
Started by ●June 16, 2008
Reply by ●June 16, 20082008-06-16
kronecker@yahoo.co.uk wrote:> I have seen quite a lot of definitions for SNR for speech + Noise > signals. Segmented SNR, noise-reduction ratio etc.It depends on what use are you trying to make of SNR.> Is there a way of > automating this or do you have to manually work this out with an audio > editor.Write 20 lines in C.> eg If I have a sentence lasting say 5 secs with additive noise > then I would have to look between the words to get the noise. Anotehr > thing is that in a real environment (as opoosed to a simulation) I can > only measure Signal + Noise and Noise alone but the Noise won't always > be the same noise since it will change continuoulsy for the non- > stationary case.Give the full picture of the problem. What exactly are you trying to accomplish? Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●June 16, 20082008-06-16
On Jun 17, 1:22 am, Vladimir Vassilevsky <antispam_bo...@hotmail.com> wrote:> kronec...@yahoo.co.uk wrote: > > I have seen quite a lot of definitions for SNR for speech + Noise > > signals. Segmented SNR, noise-reduction ratio etc. > > It depends on what use are you trying to make of SNR. > > > Is there a way of > > automating this or do you have to manually work this out with an audio > > editor. > > Write 20 lines in C. > > > eg If I have a sentence lasting say 5 secs with additive noise > > then I would have to look between the words to get the noise. Anotehr > > thing is that in a real environment (as opoosed to a simulation) I can > > only measure Signal + Noise and Noise alone but the Noise won't always > > be the same noise since it will change continuoulsy for the non- > > stationary case. > > Give the full picture of the problem. What exactly are you trying to > accomplish? > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultanthttp://www.abvolt.comOk the SIgnal is speech...sentences such as "Turn on the light" Turn off the light" etc and the noise eminates from a radio so can be anything from speech to music (normally speech). K.
Reply by ●June 16, 20082008-06-16
kronecker@yahoo.co.uk wrote:> ... the noise eminates from a radio so can be > anything from speech to music (normally speech).That's interference, not noise. Very different and often harder to deal with. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●June 17, 20082008-06-17
On Jun 17, 7:34 am, Jerry Avins <j...@ieee.org> wrote:> kronec...@yahoo.co.uk wrote: > > ... the noise eminates from a radio so can be > > anything from speech to music (normally speech). > > That's interference, not noise. Very different and often harder to deal > with. > > Jerry > -- > Engineering is the art of making what you want from things you can get. > �����������������������������������������������������������������������Ok call it what you like - to me it's non-stationary noise. How to compute the SNR though? regards K.
Reply by ●June 17, 20082008-06-17
On 17 Jun, 05:02, kronec...@yahoo.co.uk wrote:> On Jun 17, 7:34 am, Jerry Avins <j...@ieee.org> wrote: > > > kronec...@yahoo.co.uk wrote: > > > � ... the noise eminates from a radio so can be > > > anything from speech to music (normally speech). > > > That's interference, not noise. Very different and often harder to deal > > with. > > > Jerry > > -- > > Engineering is the art of making what you want from things you can get. > > ����������������������������������������������������������������������� > > Ok call it what you like - to me it's non-stationary noise.It might be to you, but to whatever algorithm you intend to apply to the data, this is interference. The difference is that 'noise' has statistichal properties which are 'significantly' different from the desired signal, whereas 'interference' has statistichal properties which are similar to the desired signal. This means that the interference will interact with a signal processoing algorithm in similar ways as the desired signal, and thus will be harder to separate away than noise.> How to > compute the SNR though?There are two ways: get a separate noise recording, or to isolate the noise from one recording, much like you seem to have already done. Rune
Reply by ●June 17, 20082008-06-17
kronecker@yahoo.co.uk wrote: You have been told many times that how to compute the SNR depends on what kind of use you are trying to make of this parameter. Answer the question for yourself: The SNR is .... what?> On Jun 17, 7:34 am, Jerry Avins <j...@ieee.org> wrote: > >>kronec...@yahoo.co.uk wrote: >> >>> ... the noise eminates from a radio so can be >>>anything from speech to music (normally speech). >> >>That's interference, not noise. Very different and often harder to deal >>with. >> > > Ok call it what you like - to me it's non-stationary noise. How to > compute the SNR though?The SNR = 12345. Is this OK? VLV
Reply by ●June 17, 20082008-06-17
kronecker@yahoo.co.uk wrote:> On Jun 17, 7:34 am, Jerry Avins <j...@ieee.org> wrote: >> kronec...@yahoo.co.uk wrote: >>> ... the noise eminates from a radio so can be >>> anything from speech to music (normally speech). >> That's interference, not noise. Very different and often harder to deal >> with. >> >> Jerry >> -- >> Engineering is the art of making what you want from things you can get. >> ����������������������������������������������������������������������� > > Ok call it what you like - to me it's non-stationary noise. How to > compute the SNR though?SIR maybe? What you find depends on the nature of the interference. Radio or person, if it's a more-or-less human voice, then you want to separate one speaker from another. I don't know how to put a number on it, but unless the amplitudes are very different, that's nothing like a noise problem. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●June 18, 20082008-06-18
On Jun 17, 10:08 pm, Rune Allnor <all...@tele.ntnu.no> wrote:> On 17 Jun, 05:02, kronec...@yahoo.co.uk wrote: > > > On Jun 17, 7:34 am, Jerry Avins <j...@ieee.org> wrote: > > > > kronec...@yahoo.co.uk wrote: > > > > ... the noise eminates from a radio so can be > > > > anything from speech to music (normally speech). > > > > That's interference, not noise. Very different and often harder to deal > > > with. > > > > Jerry > > > -- > > > Engineering is the art of making what you want from things you can get. > > > ����������������������������������������������������������������������� > > > Ok call it what you like - to me it's non-stationary noise. > > It might be to you, but to whatever algorithm you intend > to apply to the data, this is interference. The difference > is that 'noise' has statistichal properties which are > 'significantly' different from the desired signal, > whereas 'interference' has statistichal properties > which are similar to the desired signal. > > This means that the interference will interact with > a signal processoing algorithm in similar ways as > the desired signal, and thus will be harder to > separate away than noise. > > > How to > > compute the SNR though? > > There are two ways: get a separate noise recording, or > to isolate the noise from one recording, much like you > seem to have already done. > > RuneActually what you say does make a lot of sense but in the literature it is refered to as non-stationary noise. I do believe your definition is better however - as borrowed from coms of course. K.