0
$\begingroup$

Hello does any one know or at least can lead me to the right way?

What does not change pronunciation of words when someone is speaking. The idea is to create a app that checks if word/sentence is pronounced correctly. To do so I need something in pronunciation that does not change at all or changes in small range.

For example someone reads a text. His voice is recorded. Now someone else, who can not read so well reads it too. Then the recordings are compared to eacother. Then software displays on the computer screen how wrongly the second person pronounced.

Please do not suggest any software that does it!

$\endgroup$

closed as too broad by lennon310, jojek, Matt L., Phonon Oct 22 '14 at 0:38

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. 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.

  • $\begingroup$ Do not expect this to be simple. If you are working towards a program for use on a phone or tablet, I would express doubts as to whether or not there is enough processing power (to do the math) and storage (for the needed data.) You might do better to target a full PC first, and see how much storage and processing power you will need. You might find that it won't fit a phone or tablet, or you might find that it is just a little too much and that you can optimize things to fit - and you might just find that a phone can handle it easily, though I doubt that will be the case. $\endgroup$ – JRE Oct 15 '14 at 8:46
  • $\begingroup$ You can do the machine learning in the cloud. Siri and Google Now do harder things than this and they work on cell phones. $\endgroup$ – Emre Oct 15 '14 at 18:26
1
$\begingroup$

You need the right audio features, and the modern way to do this is with deep neural networks, which learn them automatically. I would use one built for speech recognition (it's basically the same problem), then model the users by the output of the top layers. Accents should manifest as clusters of the features, and languages as superclusters. I don't have much experience here but I'm making an educated guess.

Here is a relevant paper (Deep Neural Networks for Acoustic Modeling in Speech Recognition), and here is a relevant student project (Accent Recognition with Neural Network).

$\endgroup$

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