0
$\begingroup$

I want to develop a feature vector from an audio input.

Up to now, I have identified fundamental frequency, max phonation time, timbre to be among the key features to be identified.

Can someone please confirm whether it will be possible to extract these features from the audio?

$\endgroup$

closed as unclear what you're asking by Marcus Müller, Matt L., MBaz, A_A, Peter K. Jan 9 '17 at 17:14

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ It's possible. Was that really your question? $\endgroup$ – Marcus Müller Jan 8 '17 at 8:32
  • $\begingroup$ @MarcusMüller well, I want to know whether it is possible to obtain an exact value for each audio sample. $\endgroup$ – mgw2016 Jan 8 '17 at 15:25
  • $\begingroup$ no. A single sample is just a number. It can't have something like a frequency: What is the frequency of $0.4$? $\endgroup$ – Marcus Müller Jan 8 '17 at 15:35
  • $\begingroup$ Things like timbre and fundamental frequency only make sense when considered for a sequence of samples – but I'm not telling you anything you don't know, I guess; I really just try to find out what your precise question is. $\endgroup$ – Marcus Müller Jan 8 '17 at 15:38
  • $\begingroup$ @MarcusMüller I meant an audio sample of duration like 20 - 40 ms :) Not a single sample! Of course it's a sequence of samples..... Sorry for the bad definition! $\endgroup$ – mgw2016 Jan 9 '17 at 7:42
2
$\begingroup$

Have a good read of various features that can be extracted from popular audio analysis libraries like librosa: https://librosa.github.io/librosa/feature.html

  • Fundamental frequency (F0) can definitely be extracted. This itself a whole subfield, so google and previous SO questions are your best starting point after reading through what a pitch tracker looks like.

  • For max phonation time, MFCCs are typically the features computed in the front-end of a speech recognition system, which can give you "phone" sequences and their timing. But this would be a lot of processing as the concept of a "phone" in a speaker-independent sense is quite far from raw audio features one might compute directly on the audio signal (like RMS, ZCR, spectral features, etc)

  • For timbre, have a read through Bello's notes: http://www.nyu.edu/classes/bello/MIR_files/timbre.pdf

Good luck!

$\endgroup$
  • 1
    $\begingroup$ Ah! Now that makes better sense! Thanks for the guidance! $\endgroup$ – mgw2016 Jan 9 '17 at 7:39

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