Jump to content

Talk:Spectrogram

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Request soundfiles to sonogram examples

[edit]

It would be nice if the exemplary sonograms were accompagnied by the corresponding sound files (.ogg) so that we could hear what we see :-) --Dipoar (talk) 01:06, 3 October 2008 (UTC)[reply]

Aphex Twin used Coagula

[edit]

I removed the parenthetical claim that AT used Coagula because this site that's in the links section says he used metasynth. If anyone has a really solid reference, we can put it back up. --Hurtstotouchfire 02:45, 15 October 2007 (UTC)[reply]

Aphex Twin mp3

[edit]

the picture is still visible in the mp3, just not as clear. i just checked. Omegatron 06:35, Feb 1, 2004 (UTC)

Depends on the MP3. Differently coded MP3s will lose different information. TMC1221 07:04, Feb 1, 2004 (UTC)
true, true. Omegatron
Not really. It's still the same acoustic signal, so the spectrogram will be the same, just with more noise in it. High MP3 compression would have a similar effect as high JPEG compression of the spectrogram. --129.94.166.204 (talk) 04:32, 23 October 2012 (UTC)[reply]

Another example

[edit]

If this image reappears, put it back in:

Sonogram example

A sonogram of a male voice saying [tatata], as displayed by Praat.

Mother tongue of the "tatata" speaker

[edit]

The label for the "tatata" spectrogram should state the mother tongue of the speaker.

Possibly. But this sample is doubly artificial, since it not only is in the constructed language Lojban, but also comes from the corpus of a diphone speech synthesis system in that language, and therefore is a nonsense word.
In any case, we have since lost contact with the Belgian speaker of that corpus, so we cannot determine what his native language is. As the original uploader of that illustration I have no problem with seeing it go, since other spectrograms have now appeared. arj 20:10, 18 Apr 2005 (UTC)
That's a good example of the type of spectrogram traditionally used in voice though. I can make more cool edit spectrogram images of whatever, if you want. - Omegatron 20:30, Apr 18, 2005 (UTC)

Tubular Bells

[edit]

Could we have a reference for the Tubular Bells, Rugby Transmitter spectrogram?

VLF section

[edit]

I think it should read 24Hz instead of 24Khz, could someone more knowledgeable than me check this ? Zeno 04:02, 9 January 2007 (UTC)[reply]

Nevermind, i'm an audio guy, not a radio guy ! 24k is very low radio freq ! Zeno 04:08, 9 January 2007 (UTC)[reply]

Merge Waterfall_plot

[edit]

I am suggesting that Waterfall plot be merged into the Spectrogram page. They are essentially the same thing and the terms are interchangable. A waterfall plot can be 2D with color being the third dimension. A spectrogram can also be a 3D display. The only issue is that the Waterfall Plot page has a focus on room acoustics while Spectrogram is more of DSP term. (Spectrogram 05:51, 27 January 2007 (UTC))[reply]

While I agree that they carry the same information, I'm not sure they're the same. From my limited knowledge, a waterfall plot is merely a type of 3D plot that is often used to display freq. response vs. time. Check http://www.caspur.it/risorse/softappl/doc/matlab_help/techdoc/ref/waterfall.html for MATLAB's description which is a completely different description than what they have for spectrogram. Weston.pace 19:29, 5 July 2007 (UTC)[reply]

I have clarified the definition of waterfall plot and believe it is now contains enough unique content to no longer require a merge. Weston.pace 21:30, 5 July 2007 (UTC)[reply]

Format

[edit]

I added a image of a FM spectrogram. The 'edit' link below it was moved over and I was unsure how to move it back.stemperm 20:35, 13 February 2007 (UTC)[reply]

Reading Spectrograms of Speech

[edit]

How do people feel about a section regarding how to read spectrograms for speech analysis? Subsection? Separate article? I'd be willing to contribute. --Coyne025 17:17, 22 February 2007 (UTC)[reply]

I think that would be a great idea but if it is heavy on details then it might be better if it was over at a place like Wikipedia:WikiProject_Phonetics. The problem with spectrogram analysis is that it is used in so many different fields, speech happens to be just one of them. Don't want to give any particular usage too much coverage. For example: there might be too much information about images being embedded in popular music (see the recent NIN entries). While I think that those edits are great, they might be a bit too much. (Spectrogram 22:30, 22 February 2007 (UTC))[reply]

Three-dimensional???

[edit]

The second line of this article is: "It is a three-dimensional plot of the energy of the frequency content of a signal as it changes over time." I don't know anything about a spectogram but I can only see two dimensions: 1) the energe of the frequency content of a signal and 2) time. Isn't two-dimensional more correct? Lova Falk (talk) 16:10, 17 January 2008 (UTC)[reply]

The third dimension is the intensity--62.147.133.191 (talk) 17:23, 26 January 2008 (UTC)[reply]
I would go so far as to say that this article should be generalized. While a spectrogram can often be used to plot frequency information this is not its only use. More generally a spectrogram is just a way of plotting three dimensional data on a two dimensional surface using color spectra to indicate the third dimension. For example, in the work I do we often use spectrograms of time vs particle energy vs particle flux or time vs particle angle vs particle flux. If people don't object in the next few weeks, I'm going to clean up this article so it doesn't always define spectrograms as plots of time vs frequency vs power. —Preceding unsigned comment added by 128.97.68.15 (talk) 00:21, 14 May 2008 (UTC)[reply]
Allow me to strongly disagree. What you're talking about is not a spectrogram, just an unrelated plot. A spectrogram is time and frequency, period. And it's wrong to call it 3D, because it can only have one value in that "dimension" you call power. Also you might wanna look up the definition [2] --89.127.175.78 (talk) 14:51, 15 May 2008 (UTC)[reply]
1. On the issue of the definition of a spectrogram.
From http://www.merriam-webster.com/dictionary/spectrogram
"Spectrogram - a photograph, image, or diagram of a spectrum"
So what is a spectrum?
From http://www.merriam-webster.com/dictionary/spectrum
"Spectrum - 1 a: a continuum of color formed when a beam of white light is dispersed (as by passage through a prism) so that its component wavelengths are arranged in order b: any of various continua that resemble a color spectrum in consisting of an ordered arrangement by a particular characteristic (as frequency or energy): as (1): electromagnetic spectrum (2): radio spectrum (3): the range of frequencies of sound waves (4): mass spectrum c: the representation (as a plot) of a spectrum. 2 a: a continuous sequence or range <a wide spectrum of interests> <opposite ends of the political spectrum> b: kinds of organisms associated with a particular situation (as an environment) c: a range of effectiveness against pathogenic organisms <an antibiotic with a broad spectrum>"
I don't see merriam-webster specifying any particular choice of axes. In fact a spectrum can be any "ordered arrangement by a particular characteristic." They list a large variety of potential spectra which could all be plotted as spectral diagrams. So it seems to me that, as I initially suggested, the choice of axes is commonly(canonically) those of a power spectrum but that does not mean it is exclusively a diagram of a power spectrum. Why do you think a spectrogram must always be time v. frequency v. intensity?
2. On the issue of the dimension of the data. It is a system of three variables; two independent, one dependent. To specify any point of data for a spectrogram, it requires three numbers. So when I say dimension I mean strictly in the technical sense, as in "dimension of a vector space" most specifically over the standard basis. You do have a point that the data is constrained. It might be more accurate to think of a spectrogram data set as a surface spanning two dimensions of a three dimensional space. We could also say that the third parameter in each data point in the space bears a functional relationship to the first and second parameters.
One way to verify that we are, in fact, working in three dimensions is to note that I have to specify the units and scaling on the x-axis(horizontal). the y axis(vertical), and the z axis(color) to be able to properly read the plot.
That said, I totally agree that the article should clearly specify the way in which spectrogram data are constrained. 76.91.90.112 (talk) 09:08, 17 May 2008 (UTC)[reply]


From user Spectrum1234: I say it is 3-D. Is very thin paper 2-D or 3-D (even when viewed on your 2-D monitor) ?

Look at these spectrum plots (90' and 60') from the Goldwave program of an extremely quiet noise burst (thus it is about as 2-D (flat) as it could be, yet) it still 'looks' like it is 3-D (to the extent possible on a 2-D monitor):

High quality image produced with "The_Voice" displayed using "Goldwave" from this file (converted, with some loss, from a .wav file) [[1]]
Every point on a spectrogram is defined by three coordinates: time (x-axis), frequency (y-axis), intensity (color or z-coordinate). three coordinates -> three dimensions -> 3D.80.128.76.51 (talk) 14:06, 27 May 2008 (UTC)[reply]
So it is definitely 3-dimensional. What do people think about putting a note that spectrograms don't always have to be time v frequency v power graphs? I could find an excellent example image from the physical sciences. —Preceding unsigned comment added by 76.91.90.112 (talk) 08:01, 12 July 2008 (UTC)[reply]
It's about as 3-dimensional as a picture. Intensity isn't a dimension, because it can only have one value. i.e. at a precise x,y coordinates, you can only have 1 intensity value. Saying that spectrograms are 3 dimensional would be like saying that the space you perceive is 6 dimensional because you have 3 space dimensions and 3 colour dimensions. A value isn't a dimension, that's the point. —Preceding unsigned comment added by 89.124.240.31 (talk) 15:34, 18 December 2008 (UTC)[reply]
I hope that the absurdity of this position is apparent to anyone who reads this. By this logic, a hemisphere placed on a table would be two-dimensional because each point of the hemisphere has only one value of height above the table. 99.236.53.71 (talk) 06:47, 11 December 2012 (UTC)[reply]


Someone should link this page to the italian corresponding article in italian, "Spettrogramma" —Preceding unsigned comment added by 151.67.178.125 (talk) 11:42, 13 June 2009 (UTC)[reply]

sonograph vs. sonography

[edit]

Where sonograph redirects to this page, sonography redirects to medical ultrasonography. This does not seem consistent, should we change this? --130.89.139.73 (talk) 13:26, 23 July 2009 (UTC)[reply]

Awful voice spectrograms

[edit]

These two voice spectrograms are awful, they use such a low frequency resolution that you cannot distinguish any of their sonorants, they appear all fused together, which is misleading, and doesn't show how the pitch of voice varies (doesn't show the pitch at all actually). --89.124.241.234 (talk) 14:19, 13 November 2009 (UTC)[reply]

Please supply some better ones :) Wrapped in Grey (talk) 07:48, 28 June 2010 (UTC)[reply]

spectrograms are positive?

[edit]

This edit added the following to the lead section: "In the field of time–frequency signal processing, it is one of the most popular quadratic Time-Frequency Distribution that represents a signal in a joint time-frequency domain and that has the property of being positive." Does that mean something to the regular contributors? Maybe there's just a simple typo throwing me off, but as a first-time visitor here, even with some experience with signal processing, I can't understand a word of it. For example, the terms "quadratic", "positive" and "distribution" are not defined, nor do they play much of a role in the article. Spiel496 (talk) 17:44, 27 April 2011 (UTC)[reply]

Removed. 93.145.142.238 (talk) 18:46, 9 August 2011 (UTC)[reply]
[edit]

They do not bear any apparent relation to the subject (spectrogram) or content of the article. Did some previous editor get confused between cryptograms and spectrograms? — Preceding unsigned comment added by 167.216.14.162 (talk) 22:56, 15 September 2011 (UTC)[reply]

I removed the {{Hidden messages}} navbox, as I could see no obvious relation between it and this article. –Sparkgap (talk) 18:48, 16 September 2011 (UTC)[reply]

Why does this article claim that spectrograms are just for sound?

[edit]

According to [3] a spectrogram is a "A graphic or photographic representation of a spectrum" - which is what I always understood the general term spectrogram means. For example you could have a spectrogram of the light observed from a star. This article gives the impression that a spectrogram is always related to sound waves. You can also find scientific papers ([4]) citing the term spectrogram in relation to other spectra than just sound waves. 202.180.125.97 (talk) 18:54, 2 October 2013 (UTC)[reply]

Well I certainly use them for vibration, and I have seen them for radio spectra as well. I'll have ago at removing the bias. The reason that many of these spectral analysis articles is stufffed is that the medical/speech boys think they are the only ones using them, and 'correct' any other usage (eg cepstrum). Greglocock (talk) 22:24, 2 October 2013 (UTC)[reply]

See an online spectrogram of speech or other sounds captured by your computer's microphone.

[edit]

When you click this link under "External Links", it links to a page that doesn't exist. Can someone fix this maybe? 2601:8B:4400:C180:2048:CDF2:690E:727 (talk) 01:33, 30 December 2015 (UTC)[reply]

Spectrogram and spectograph

[edit]

Some recent edits have raised the issue again about how this Spectrogram article is restricted to sound while the Spectograph article seems more restricted to EM radiation. The sentence Kvng reverted is generally correct, albeit added without source. A spectogram is the output of a spectrograph, or someimtes the result of digital transformation of the input signal in a manner that is generally analogous to what a spectrograph does. Who wants to step up to resolve this? It requires an editor who can pull information from multiple disciplines. I'm happy to help, but don't think I can do the entire thing myself.  —jmcgnh(talk) (contribs) 23:16, 22 August 2016 (UTC)[reply]

I can do some work on this. I'm not intimately familiar with these terms but I do know how to research this. Can you provide any links to previous talk page discussion on this mess? It looks like Spectrometer and Optical spectrometer should also be put into the mix here. ~Kvng (talk) 23:45, 22 August 2016 (UTC)[reply]
Kvng: Agreed. (I didn't see this until today).
This feels like a bigger change than I'm comfortable with at my current stage of becoming an editor. Fortunately there's no deadline, right?  —jmcgnh(talk) (contribs) 01:34, 11 September 2016 (UTC)[reply]
I have expanded the scope of this article to clearly include non-audio signals. I have proposed a merge for Spectrograph. ~Kvng (talk) 16:30, 11 February 2018 (UTC)[reply]

Resolution of a spectrogram from band-pass filters

[edit]

How would the resolution of a spectrogram obtained with a "series of band-pass filters" compare against that created conventionally via an FFT? Particularly in the low-mid audio frequency range. An FFT or a "digital" equalizer is quite poor at resolving narrow tones without compromising the time-resolution across the spectrum. I guess there must be a catch, because this type of filter isn't used anywhere. I would like to hear an explanation with minimal maths.  J7n (talk) 17:20, 11 February 2017 (UTC)[reply]

OK. Minimum maths. The equation of interest is dF>=1/T. dF is the frequency resolution, T is the characteristic time of the analysis method, in the case of DSP it is the length of one frame of data or the number of samples divided by the sampling frequency
In the case of an analogue filter the characteristic time is governed by the same equation. That is the settling time of the filter will be inversely proportional to the bandwidth of the filter.
The inequality in the equation is because in practice you don't quite get as much resolution as you'd like, due to a whole bunch of real and theoretical effects.
Apparently that equation is analogous to, or even a consequence of, Heisenberg's uncertainty principle. Cool eh?
real time filtering systems can probably be most easily implemented by using cascades of digital filters in the time domain, each with a high bandwidth and with short settling times, plus feed forward. At that point I am out of my depth.Greglocock (talk) 03:55, 12 February 2017 (UTC)[reply]

Some other variable

[edit]

I am not aware of spectrograms that don't use time as one of their axis. Greglocock asserts that RPM can replace time and provides this citation. These plots have frequency, amplitude and RPM axis but there is no indication they're called or considered spectrograms. I have marked this with {{fv}}. I think Greglocock would prefer if I just marked "or some other variable" in the lead with {{cn}} but, like I said, I'm not aware of any spectrograms that don't use time and it is difficult to prove a negative. ~Kvng (talk) 15:34, 9 November 2017 (UTC)[reply]

http://www.machinedesign.com/fea-and-simulation/noise-prediction-moving-mechanisms-using-coupled-mbd-acoustics-analysis for a free article using that terminology. What a waste of effort. Greglocock (talk) 03:20, 10 November 2017 (UTC)[reply]
My interpetation is that the spectrograms mention in this article are time based. The time axis is labeled RPM in Figure 6 because, for the measurement, the machine is sped up linearly as a function of time and recorded with a microphone so there is a proportional relationship between time and RPM. Here's the description of the measurement from the article, "Once simulations are completed, the acoustic pressure at microphones surrounding the virtual prototypes is outputted. The outputs time domain series can be easily converted in to audio files to listen to the response of the prototype to a certain excitation. The time domain data can be further post-processed in the frequency domain by means of spectrograms which show multiple signal spectra allowing the display of noise levels evolution with respect to the working regimes."
Whether or not you accept my interpretation of this source, the source does not provide a definition or description of spectrogram; it is merely demonstrating an in-context use of the term so should not be used to override our more direct sources which clearly state that a spectrogram is a time-frequency-intensity plot. ~Kvng (talk) 15:14, 12 November 2017 (UTC)[reply]
I think "some other variable" just feels off, and think, given the very dominant use of time, we could use - "as they vary, generally, with time." Or some similar wording.MusicPoliticsFlowers (talk) 05:29, 19 January 2018 (UTC)[reply]
I've removed "or some other variable" for now because there isn't a consensus to include it. We can continue discussion here and try to reach consensus. ~Kvng (talk) 15:29, 21 January 2018 (UTC)[reply]