2
$\begingroup$

Hi signal processing community,

currently I am searching a topic for my master thesis in computer science and since I am a very good singer I want to do something with music. I thought about a topic with pitch detection algorithms. But I am very new in this context and I want to inform me first, so I hope you can help me and answer me some of my questions.

I know that I have to work with FFT (Fast Fourier Transformation) and algorithms like AMDF or Yin. Can you give me a hint of a starting point (a paper or website).

I start with a signal which I got from my microphone, but what are the next steps.

I hope you can help me, thank you all.

$\endgroup$
  • $\begingroup$ i am not at all impressed with either frequency-domain methods (which make use of the FFT) nor Yin, which i consider to be just another variation of an minimize AMDF or ASDF (and minimizing ASDF is pretty much the same as maximizing autocorrelation). questions around pitch detection have been asked before here. i had a few different answers. $\endgroup$ – robert bristow-johnson Mar 23 '18 at 23:48
1
$\begingroup$

1) Get some practical experience on what you are studying. By this I mean is take your recordings of your voice and look at them close up on you computer. You can use a program like Audacity.

2) Yes, you will have to learn about the Discrete Fourier Transform (DFT, the FFT is a computationally efficient version). I write blog about it explaining it in my own way. You can find the link on my profile page.

3) Pick a different topic that is worthy of a Master's Thesis. There are many techniques for pitch determination, which is a pyscho-acoustic phenomenon vs frequency estimation which is basically a physics problem. Since you are a singer, you might be interested in how sound waves are attenuated as they travel through air. You simply have to set up some microphones at different distances from the source and look at their waveforms to see what I am talking about.

4) Depending on your math chops, you may consider coming up with data visualization techniques for signal data. There is lots of room for creativity here.

Hope this helps.

Ced

$\endgroup$
  • $\begingroup$ i wouldn't recommend any pitch detection method that uses the FFT (unless maybe one uses the FFT to perform "fast autocorrelation"). but the OP should learn about the DFT and the theorems and simple applications anyway. if the OP wants to do decent pitch detection, i would recommend looking into the correlation methods to estimate the fundamental frequency which, if sub-harmonics are properly understood and de-emphasized, is closely related to the psycho-acoustic notion of pitch. $\endgroup$ – robert bristow-johnson Jun 23 '18 at 8:20
1
$\begingroup$

I start with a signal which I got from my microphone, but what are the next steps.

I think you have already answered that question yourself.

But I am very new in this context and I want to inform me first,

That is one of the core activities of a scientific thesis. You do good old fashioned literature research by plowing through papers, books, blogs, online resources, etc. Find the good ones and discard the bad ones. Starting from the good ones, figure out what currently works well in the field and what doesn't. Select an aspect that you think can be approved on. Make this the topic of your thesis.

Pitch detection is very well researched and developed. You can download a free app for your phone that will easily tell you the note that you are singing. In order to make this a viable thesis, you probably need to have some idea how to make it better or different.

$\endgroup$

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.