Piano World Home Page Forums Our Most Popular Forums Piano Forum Spectrogram software - Windows?

Would you happen to know of any Windows software that can produce a 3D spectrogram like the one above?

Ideally with high resolution capability along the time axis to home in on the attack between 0.5 and 1 second.

Thanks

You've got some serious 50 or 60 Hz hum going on there.

If you don't fear using a programming language, have a look at python with numpy and scipy. There are lots of cut-and-paste templates waiting for you, ready to test and adapt. Start your research with this stackoverflow answer.

That will be a challenge for a rainy day. Not my hum, glad to say. Thank you.

3D spectrograms seem to be (literally) one for the birds (and birdsong). For reference I see Audacity has a few plug ins: SoundRider, Span, StarryDave.

Are you sure you want a 'flattened' 3-D representation? They look cool, but are not the easiest to extract information from. A 2-D grey-scale spectrogram may be more practical for measurement/visualization.

For the latter, I recommend Praat: https://www.fon.hum.uva.nl/praat/ . Open and free and very high-quality.

It is designed for speech, but works for any sound. It shows spectrogram + waveform at the same time, and you can easily set spectrogram parameters to your liking. Praat is an extensive program intended for speech scientists, but to do what you want, you just open your sound file from the menu, and choose the "View&Edit" button.

So Withindale, what kind of mischief are you wanting to get into with such software?

You can just hand it off to your tech and say "fix this".

No, I am not sure. For my own purposes, you are probably right. To make a point to others, 3D may be better.

Yes, but as it happened, our local techs were unable to "fix this" so I got into doing it myself. Example, they said the bass was not tuneable when the keybed needed levelling.

Well that was not the question, but mischief is the word. I want to persuade two European gentlemen they are missing something.

In the plot above you can just see an orange "fence" going northwest from 0.5 seconds, separating a strip of green "grass" from the partials. That plot may be sufficient to make the point, but it would be nice to show the fence more clearly.

So that begs the next question: what do you want them to see that they are missing? :-)

The effects of removing (some or all of) the "fence" and the "barrier" at the bass end.

You might be interested in the SPEAR program, which I describe and link to in this post: https://forum.pianoworld.com/ubbthre...ve-and-on-macs-evolving.html#Post3182149 . Allows one to see the partials changing amplitudes over time, and more.

I have been using Oscillometer for the past decade. It is free, and provides a vast variety of information for audio analysis. I have published many DFFT graphs here using it. It has a 3D graph function as well, and operates in real time or from just about any audio file. Here is a picture I took today of a tuning fork. The two low frequency bumps are my hitting the piano with it and then a second or so later getting it to resonate on some piano wood.



The left graphs are instantaneous amplitude, the middle are a Frequency Domain DFFT, and the right are 3D -Frequency on the X axis Amplitude on the Y axis, and Time on the Z axis.

It can be dowloaded from this site Oscillometer

I'm not quite sure why the "fence" is there. It looks like the duration is very short, and it's mostly white noise. How does it sound like, is it a click or pop?

About what could cause it:

Maybe it's in the samples? E.g. somebody cut them in in the middle of the attack rather than before the sound started, in an attempt to make the engine more "responsive". Or maybe there's clipping going on, difficult to say as there's no scale other than color.

Maybe it's the filters? Most engines have FIR and IIR processing for pre-effects like mixing of velocity layers, and post-effects like resonance, EQ, reverb, etc. If one of them isn't properly initialized and has garbage or previous unrelated data in their window, it might produce unexpected output until it stabilizes (or evicts the garbage from the window in case of FIR).

Maybe it's just an ordinary programming error? For example a slightly off sample memory address calculation, resulting in the actual sample data being preceeded with a bit of unrelated garbage initially.

Once you can articulate an educated guess, you get into a position where you can think up an experiment to support it or discard it. It's tedious and time consuming work, but you seem to be determined.

Ian, why don't you send me some .wav files of the note or notes that are at issue. If it is the attack in which are interested, the files need be only 2 seconds or so long. A 15 second file will yield better decay data. The .wav files can be 16 bit up to 24 bit, and 44.1kHz up to 96kHz. I can also do .mp3 if necessary.

Make sure the mic is not touching the piano when you record. That 'fence' is a broad spectrum noise source caused by the key bedding and/or hammer hitting the string. Maybe some loose change inside the piano.

PM me for an email address.

Thank you all for your posts. I will have to get back to this tomorrow. I took the original image from a thesis (not mine) merely to show what I was looking for. Meanwhile here, from Player13's gallery, is a spectrogram of C6 from a Yamaha S7X in a showroom. The fence looks more like a ridge and I think it represents the effect of the hammer hitting the string as Prout suggests.

That's a nice graph. It shows C6 struck twice with a typical fundamental amplitude variation over time as the three strings interact with each other. It shows a nice number of partials - 9 on the second strike, and a typical noise spectrum for the key hitting the bottom.

Yes, this is about that noise spectrum. One of its effects is to muddy the sound as described in Classical Recording by Decca engineers (see Google Books).




I have just discovered Askenfelt measured the string borne noise and the structure borne noise in 1993:



You may have have to read the article to make sense of the diagram but, in brief, the top plot is the frequency analysis of the bridge acceleration 1-2 ms after the hammer hits the strings, and the bottom plot is the same measurement when the strings were replaced with a dummy for the hammer to hit. The top one represents the combined string borne and structure borne noise while the bottom one is just the structure borne noise.

One point to note is the structure borne noise is the main component of the noise in the bass and the tenor. Another is the structure borne noise is driving the soundboard from the rim and thereby accelerating the bridge, not the other way round. It's possible the amount of structure borne noise emanating from the soundboard is a bit greater than the bottom plot indicates.

What would be interesting is to repeat the S7X C6 plot on another piano for C1, C2, C3 ... and see the variation from the bass up into the treble.

Thank you for kind offer, Prout, I am wondering if you would like to have a go at that exercise on your piano - I have acoustically isolated mine to reduce the structure borne noise. Please do not feel under any obligation. Plan B is to do it on and off myself when I have some time. Plan A is ask one of the European gentlemen about setting up a master's research project into structural vibration and acoustic isolation.

Hi Ian,

I have previously recorded all 88 notes of my M&H BB many, many, times and have all the files on hard drives. I'll pick a few and post the DFFTs with comments.

Keybed noise is a very important part of the piano experience, and is exhaustively studied by Digital Piano researchers to increase the fidelity of the DP aural experience relative to the AP.

That being said, the proper levelling of the keydesk in the piano case is critical to both noise reduction and to touch sense and may change with humidity and the seasons.

Ian, I have loaded a bunch of graphs of Keyboard Noise from all the Cs on the piano, as well as graphs of the Cs showing the noise and partial decays of each note. There is one short audio file of all the fully muted Cs that accompanies the noise graph.

The results are really cool. The upper Cs all produce a very consistent noise 'fence'. The lower Cs are each different and cover a larger range of frequencies.

I didn't bother to show them here as the resolution is too high to load, and, if compressed, too low to read.

Piano Noise Spectra