0
$\begingroup$

I want to create an AI that converts words to International Phonetic Alphabet (IPA), but I am not sure which architecture I am supposed to use.

It is not possible to translate the characters one by one since there are multiple characters in the source word corresponding to one IPA character. There are solutions for this kind of problem, for example using an Encoder that encodes the content of the input which the decoder then translates, but I am uncertain if this isn't too abstract for this problem.

Can anyone think of a suitable solution for this task?

$\endgroup$
2
  • $\begingroup$ You could try a finite state machine. $\endgroup$ Jul 20, 2021 at 15:45
  • 2
    $\begingroup$ I don't think what you're suggesting is too abstract -- this sounds suitable for seq2seq architectures, such as those used in neural machine translation. It might not be the simplest possible model for the job though. You might be able to adapt this tutorial. $\endgroup$ Jul 20, 2021 at 15:50

1 Answer 1

0
$\begingroup$

Look for sequence-to-sequence modelling, aka, seq2seq. https://blog.keras.io/a-ten-minute-introduction-to-sequence-to-sequence-learning-in-keras.html https://en.wikipedia.org/wiki/Seq2seq

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .