Amazing explanation. Brilliant idea to connect this topic to analog telephones, so it is easier to understand. Very well done, could get me a better grade at my exam ;)
Should the power in dB not be from: 20*log_(10)(abs(X[m;k]))? Since power is equal to the magnitude^2? i.e. if you wanted to use 10log10, you'd need to include the power of 2 inside the logarithm: abs(X[m;k])^2
hi, how do you get the time and you say that the y axis represents magnitude of DFT coefficients, but its just the frequency axis. magnitude is represented by hue variation of the colour. could you please help to get away with confusion ?
The y axis represents the frequency axis. The hue represents the magnitude of the corresponding frequency. So to be more clear, if you find a lighter or as the guy in the video says, a "brilliant" band somewhere on the spectrogram, the corresponding y axis value represents how high the frequency of the signal is and the amount of "brilliance" represents how loud the signal of the corresponding frequency is
Hey, this is an awesome video to understand Spectrogram. I was so excited with your youtube video and so I created blog post that reproduce many of the plots in your presentation in python fairyonice.github.io/implement-the-spectrogram-from-scratch-in-python.html. Thank you very much for this great tutorial. Just one comment: Your slide at 2:39 contains 6 peaks and you say that two of the each peaks corresponds to one dial sound. As you are using the digits 1, 2 and 3, I would expect to see 4 peaks at 697Hz, 1209Hz, 1336Hz and 1477Hz, rather than 6 peaks in the frequency domain plot. Maybe you are rather using digits 1,5,9? I am not sure. But anyway, the concept was very clearly explained so thank you very much!
THIS IS AWESOME.Very clear and very good examples, now it is so much easier to understand. Thank you ! I mean it.
My concepts are now clear as day! Can't thank you enough
This is the clearest explanation of STFT I have found so far
Amazing explanation. Brilliant idea to connect this topic to analog telephones, so it is easier to understand. Very well done, could get me a better grade at my exam ;)
Thank you for your great explanation. You are the best. Please continue.
This is the best explanation I’ve ever seen! Thanks.
Hi, you mention references to wavelets in the bibliography for the class. Where can these be found ?
best explanation out there!
Wow, what a nice lecture! Thanks a lot!
This is gold
Fantastic lecture! Thanks!
Very good explanations
a question, is stft() in Matlab is calculated as a short time fft
awesome, dude! btw, when u wrote down dt*df=2pi I must admit i kinda went "damn nigga, thats some real quantum-mechanics-behind-the-curtain"
Thanks, it was pretty useful
where can i find speech samples to analyze in? or the sppech corpus? I need it for my project.
Should the power in dB not be from: 20*log_(10)(abs(X[m;k]))? Since power is equal to the magnitude^2? i.e. if you wanted to use 10log10, you'd need to include the power of 2 inside the logarithm: abs(X[m;k])^2
excellent overview !!!
Solid. Thanks!
what is the window function in the video?
hi, how do you get the time and you say that the y axis represents magnitude of DFT coefficients, but its just the frequency axis. magnitude is represented by hue variation of the colour. could you please help to get away with confusion ?
The y axis represents the frequency axis. The hue represents the magnitude of the corresponding frequency. So to be more clear, if you find a lighter or as the guy in the video says, a "brilliant" band somewhere on the spectrogram, the corresponding y axis value represents how high the frequency of the signal is and the amount of "brilliance" represents how loud the signal of the corresponding frequency is
What is the code python of STFT?
Thank you ! I understand now :)
it's great !
Hey, this is an awesome video to understand Spectrogram. I was so excited with your youtube video and so I created blog post that reproduce many of the plots in your presentation in python fairyonice.github.io/implement-the-spectrogram-from-scratch-in-python.html.
Thank you very much for this great tutorial.
Just one comment: Your slide at 2:39 contains 6 peaks and you say that two of the each peaks corresponds to one dial sound. As you are using the digits 1, 2 and 3, I would expect to see 4 peaks at 697Hz, 1209Hz, 1336Hz and 1477Hz, rather than 6 peaks in the frequency domain plot. Maybe you are rather using digits 1,5,9? I am not sure. But anyway, the concept was very clearly explained so thank you very much!
how to convert spectrogram to audio signal?
you can't, because you discard frequency phase information during the spectrogram generation
thanks!!!!