Hello,
An interesting lecture from 2005 -that most probably most of you yet know- about the use of continuous wavelets and also the improvment brought by the use of some preconditionning filtering.
http://fulcrum-acoustic.com/wordpre...ponse-with-digital-signal-processing-2005.pdf
Someone used to critized the use by me of the name "wavelets spectrogram" preferring the term " scalogram". You can see that this paper use the "name spectrogram".
Best regards from Paris, France
Jean-Michel Le Cléac'h
An interesting lecture from 2005 -that most probably most of you yet know- about the use of continuous wavelets and also the improvment brought by the use of some preconditionning filtering.
http://fulcrum-acoustic.com/wordpre...ponse-with-digital-signal-processing-2005.pdf
Someone used to critized the use by me of the name "wavelets spectrogram" preferring the term " scalogram". You can see that this paper use the "name spectrogram".
Best regards from Paris, France
Jean-Michel Le Cléac'h
Hello,
I would know your opinion about some settings you think are better for the analysis of impulse reponses of loudspeakers using wavelets transform.
On the attached GIF file I gathered 6 wavelets spectrograms of the same impulse response.
Each spectrogram is labelled with a value of Q ranging from 0.5 to 1.0.
This Q parameter is the fractionnal bandwith (ratio between the bandwith of the gaussian wavelet and its center frequency). I used as a limit of the envelop Bwr = -50dB.
Values of Q toward 0.5 correspond to a smaller bandwith so every spots on the spectrogram tends to have a resonant beahaviour and this explains why they prolongate in time.
Values of Q toward 1.0 correspond to a larger bandwith so every spots on the spectrogram tends to prolongate in frequency .
Which value of Q do you think give a better view of the beahaviour of the loudspeaker?
Do you think we may vary Q with frequency (as in the Bark method).
Best regards from Paris,
Jean-Michel Le Cléac'h
I would know your opinion about some settings you think are better for the analysis of impulse reponses of loudspeakers using wavelets transform.
On the attached GIF file I gathered 6 wavelets spectrograms of the same impulse response.
Each spectrogram is labelled with a value of Q ranging from 0.5 to 1.0.
This Q parameter is the fractionnal bandwith (ratio between the bandwith of the gaussian wavelet and its center frequency). I used as a limit of the envelop Bwr = -50dB.
Values of Q toward 0.5 correspond to a smaller bandwith so every spots on the spectrogram tends to have a resonant beahaviour and this explains why they prolongate in time.
Values of Q toward 1.0 correspond to a larger bandwith so every spots on the spectrogram tends to prolongate in frequency .
Which value of Q do you think give a better view of the beahaviour of the loudspeaker?
Do you think we may vary Q with frequency (as in the Bark method).
Best regards from Paris,
Jean-Michel Le Cléac'h
Attachments
Hello,
Here the link to an interesting paper.
Give a look at pages 28- 31 to the wavelets spectrogram of the IR of various diaphragms and specially page 31 (fig21) to the wavelets spectrogram of a large beryllium dome.
http://www.almainternational.org/as...arge compression drivers-alma ws 2011.doc.pdf
Best regards from Paris, France
Jean-Michel Le Cléac'h
Here the link to an interesting paper.
Give a look at pages 28- 31 to the wavelets spectrogram of the IR of various diaphragms and specially page 31 (fig21) to the wavelets spectrogram of a large beryllium dome.
http://www.almainternational.org/as...arge compression drivers-alma ws 2011.doc.pdf
Best regards from Paris, France
Jean-Michel Le Cléac'h
Personally, the different Q pics tell me the same thing because the general trend is the same. I would prefer bark bands and also viewed in such scale mainly to understand what I will hear in audible terms. The "noodles" presented previously allows for easier comparison.
The diaphragm wavelet spectrograms look interesting. I did not see what size the diaphragms were, but for a 1 inch titanium diaphragm of my design, I could not see any breakup mode even close to manifesting at 20KHz when Klippel scanned. It looked like it might possibly be close to 40KHz, but since Klippel could not scan so high frequency, I could not confirm this.
The diaphragm wavelet spectrograms look interesting. I did not see what size the diaphragms were, but for a 1 inch titanium diaphragm of my design, I could not see any breakup mode even close to manifesting at 20KHz when Klippel scanned. It looked like it might possibly be close to 40KHz, but since Klippel could not scan so high frequency, I could not confirm this.
Possibly we can not have it all (at the same time) - but as for me, I'd rather keep measurement and interpretation separated rather than mix it into one plot by means of adding a "psychoacoustic shaping".
So I'd love to see either overlayed plots that reflect both comb filtering effect (along the frequency axis) as well as "stepped decay" (along the time axis) - which - in the end results in grid structures like this:
An externally hosted image should be here but it was not working when we last tested it.
as Elias has shown here:
http://www.diyaudio.com/forums/mult...udio-measurements-what-how-3.html#post2151544
--------
second best attempt - for me - is to show two plots
- one with frequency resolution emphasis and another one with time resolution emphasis
--------
Nice plots visualizing an aspect of the sonic pattern of materials in this paper
Thanks for linking !
Best
Michael
Hello Pano,
Thanks for you input concerning the optimal value of Q (a mean 0.75 value).
About the paper, you are right. Comparing the wavelets spectrograms of the beryllium and aluminum diaphragm, we can ask ourself : is a beryllium diaphragm worth the price?
and you surely notice the sentences inthe paper:
"At least, it appears that the aluminum and the beryllium diaphragms have similar decay characteristics, and both are better than titanium in that regard."
Best regards from Paris, France
Jean-Michel Le Cléac'h
Thanks for you input concerning the optimal value of Q (a mean 0.75 value).
About the paper, you are right. Comparing the wavelets spectrograms of the beryllium and aluminum diaphragm, we can ask ourself : is a beryllium diaphragm worth the price?
and you surely notice the sentences inthe paper:
"At least, it appears that the aluminum and the beryllium diaphragms have similar decay characteristics, and both are better than titanium in that regard."
Best regards from Paris, France
Jean-Michel Le Cléac'h
possibly with the plots you show, Q 0.5 for me seems to be the best compromise - but - if I'm after detecting of reflections / loops :
the higher Q the better IMO, simply as time resolution gets even better.
I think you might be right that
"multiply wavelets graph may be somewhat confusing for many people."
But actually this is it - I mean - we are at the boarder of detecting spectral distribution along a time and frequency axis - so its kind of a "dualism issue"
To constantly "change hat" is confusing for everyone, no?
Michael
the higher Q the better IMO, simply as time resolution gets even better.
I think you might be right that
"multiply wavelets graph may be somewhat confusing for many people."
But actually this is it - I mean - we are at the boarder of detecting spectral distribution along a time and frequency axis - so its kind of a "dualism issue"
To constantly "change hat" is confusing for everyone, no?
Michael
Hello,
I find your definition of Q a bit extraordinary since you define Q as proportional to bandwidth, but I would define Q as proportional to inverse of bandwidth, since in my thinking high Q leads to narrow bandwidth and thus lower temporal resolution and low Q leads to wide bandwidth and greater temporal resolution. But never mind it's just a definition
I was also pondering what 'Q' should be shown in a wavelet plot. Positive thing is that the variability of the Q can be used to emphasize the phenomena under study, that means if you like to investigate temporal effect you can choose a 'narrow' wavelet for best visualisation. However, at the end it still kind of bothered me, since never cannot be sure what Q is 'correct'. That lead me to the Bark and the predefined bandwidths of human sound processing. At least Bark helps me to understand what is relevant and what is less relevant. If I see something on a Bark plot I propably should worry about it and if I don't see much then I can concentrate to more relevant matters.
Of course an engineer like to do engineering propably because of the engineering itself and he will propably like the Q approach better over Bark. Unfortunately human perception does not enter the loop as much as it should, I think.
- Elias
I find your definition of Q a bit extraordinary since you define Q as proportional to bandwidth, but I would define Q as proportional to inverse of bandwidth, since in my thinking high Q leads to narrow bandwidth and thus lower temporal resolution and low Q leads to wide bandwidth and greater temporal resolution. But never mind it's just a definition
I was also pondering what 'Q' should be shown in a wavelet plot. Positive thing is that the variability of the Q can be used to emphasize the phenomena under study, that means if you like to investigate temporal effect you can choose a 'narrow' wavelet for best visualisation. However, at the end it still kind of bothered me, since never cannot be sure what Q is 'correct'. That lead me to the Bark and the predefined bandwidths of human sound processing. At least Bark helps me to understand what is relevant and what is less relevant. If I see something on a Bark plot I propably should worry about it and if I don't see much then I can concentrate to more relevant matters.
Of course an engineer like to do engineering propably because of the engineering itself and he will propably like the Q approach better over Bark. Unfortunately human perception does not enter the loop as much as it should, I think.
- Elias
Mel, the Barking ERB
There are several common ways of plotting our perception of pitch
MEL, ERB and Bark.
This paper gives insight into the ERB method. Equivalent Rectangular Bandwidth
I don't have much experience with these plots, but they do seem a good choice for "seeing" what is important in the horn plots.
There are several common ways of plotting our perception of pitch
MEL, ERB and Bark.
This paper gives insight into the ERB method. Equivalent Rectangular Bandwidth
I don't have much experience with these plots, but they do seem a good choice for "seeing" what is important in the horn plots.
Hi,
BTW, I use ERB while generating the gammatone wavelets.
I think Bark may be known by more people due its longer history.
- Elias
BTW, I use ERB while generating the gammatone wavelets.
I think Bark may be known by more people due its longer history.
- Elias
Hello Elias,
I never wrote that the parameter Q in my wavelets routine is a "quality factor" as used for resonators. As the author of the routine don't I have any right to choose any name I want for the parameters I am using as long I define it as did it :
"Each spectrogram is labelled with a value of Q ranging from 0.5 to 1.0.
This Q parameter is the fractionnal bandwith (ratio between the bandwith of the gaussian wavelet and its center frequency)."
I could have taken bw as it is what the Matlab function Gauspuls refer to for the fractional bandwith.... It is the same approach to say that the wavelets at each frequency has fixed number of periods width (which is commonly done).
About the Bark approach. I have my own "Bark based Wavelet Tranform routine" and I use it essentially to analyse impulse responses having very low frequency (<600Hz). To analyse compression drivers and horns as it is the purpose of this thread the Bark approach is useless IMHO.
Best regards from Paris, France
Jean-Michel Le Cléac'h
I never wrote that the parameter Q in my wavelets routine is a "quality factor" as used for resonators. As the author of the routine don't I have any right to choose any name I want for the parameters I am using as long I define it as did it :
"Each spectrogram is labelled with a value of Q ranging from 0.5 to 1.0.
This Q parameter is the fractionnal bandwith (ratio between the bandwith of the gaussian wavelet and its center frequency)."
I could have taken bw as it is what the Matlab function Gauspuls refer to for the fractional bandwith.... It is the same approach to say that the wavelets at each frequency has fixed number of periods width (which is commonly done).
About the Bark approach. I have my own "Bark based Wavelet Tranform routine" and I use it essentially to analyse impulse responses having very low frequency (<600Hz). To analyse compression drivers and horns as it is the purpose of this thread the Bark approach is useless IMHO.
Best regards from Paris, France
Jean-Michel Le Cléac'h
Hello,
I find your definition of Q a bit extraordinary since you define Q as proportional to bandwidth, but I would define Q as proportional to inverse of bandwidth, since in my thinking high Q leads to narrow bandwidth and thus lower temporal resolution and low Q leads to wide bandwidth and greater temporal resolution. But never mind it's just a definition
I was also pondering what 'Q' should be shown in a wavelet plot.
Hello Panomaniac,
Let's consider Elias wavelets tranform graphs of an ideal pulse (Dirac):
1) based on a Bark window
http://i952.photobucket.com/albums/ae6/_E_P_/diyaudio/Ideal_impulse_50-20kHz_Barkwavelet.png
2) based on a constant Q window
http://i952.photobucket.com/albums/ae6/_E_P_/diyaudio/Ideal_impulse_50-20kHz_constantQwavelet3.png
Despite the different horizontal scales, we can see that there is nearly no difference of shape ( between the 2 graphs above 900Hz. (which is the range of frequency that interest us when we use compression drivers and horns (most of...).
For a fractional bandwith around .7 there will be no difference at all.
Best regards from Paris, France,
Jean-Michel Le Cléac'h
Let's consider Elias wavelets tranform graphs of an ideal pulse (Dirac):
1) based on a Bark window
http://i952.photobucket.com/albums/ae6/_E_P_/diyaudio/Ideal_impulse_50-20kHz_Barkwavelet.png
2) based on a constant Q window
http://i952.photobucket.com/albums/ae6/_E_P_/diyaudio/Ideal_impulse_50-20kHz_constantQwavelet3.png
Despite the different horizontal scales, we can see that there is nearly no difference of shape ( between the 2 graphs above 900Hz. (which is the range of frequency that interest us when we use compression drivers and horns (most of...).
For a fractional bandwith around .7 there will be no difference at all.
Best regards from Paris, France,
Jean-Michel Le Cléac'h
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.