I programmed a sample accurate spectrogram some time ago and i still have never seen a picture of another one. It's interesting maths, i was told it's trivial, so why has no one else posted research on sample accurate sgrams on the internet? Can you find one. Here is a pic of my version:  Zoom in of 1ms of sound in mid frequencies

this is the same program without the zoom and in 2d Mode:

enter image description here

  • $\begingroup$ Oh Thank you Antoine! in Wigner-Ville results i found this kind of spectrogram, which is of the kind of detail resolution i am researching, this is what a sonogram should resemble, and i find these views of sound fascinating: freeware-download.com/screenshots/2/12482-a.jpg $\endgroup$ May 1 '15 at 10:37
  • $\begingroup$ Indeed, I'm in another area now but this is the kind of problem that got me into signal processing. Have fun :) $\endgroup$ May 1 '15 at 10:40
  • $\begingroup$ What do you mean by "sample accurate spectrogram"? $\endgroup$
    – endolith
    May 1 '15 at 13:52
  • $\begingroup$ I meant a spectrogram of the the same sample resolution as the audio file. input a 1 second sound with 22k samples, output a spectrogram with 22k values on y axis. $\endgroup$ May 1 '15 at 14:11
  • $\begingroup$ @ufomorace Hum then I may have misled you. With a signal with 22k values, you may perfectly have a "22k values" spectrogram. It is not related with accuracy. For instance, you could use a 1-sample step and a 5k-samples length window, your spectrogram would have 22k values on y axis but a very poor temporal resolution (yet an exelent frequency resolution). Do you know about uncertainty principle and time-frequency tradeoff ? $\endgroup$ May 1 '15 at 15:46

Producing accurate time-frequency analysis is an area of research. Some pointers are : Wigner-Ville transformation, re-allocation methods, wavelets.

  • $\begingroup$ i found an implementation program here which uses Wigner Ville transformation called sonogram 4 here. christoph-lauer.de/sonogram $\endgroup$ May 1 '15 at 10:55
  • $\begingroup$ Thanks a lot i have to test FFT implementation with single sample increments because Wigner Ville seems to be not very clear and rather slow. I did a detailed comparison study of a birdcall and here is the result: s29.postimg.org/th7pjz66v/sonogram_study.jpg $\endgroup$ May 2 '15 at 11:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.