Are there any open-sourced library / Deep Learning models that convert an audio clip of a word to its IPA representation? In this case, the audio is from a non-native speaker and the goal is to generate the corresponding IPA representation of the word as uttered by the speaker. This representation is expected to be much different than the 'correct' or 'true' IPA representation. Basically I want to understand the deep learning model behind https://elsaspeak.com/en/
Add a comment
|
1 Answer
There is no program that converts samples speech to IPA. Some program might appear to do that, by converting an utterance of language Z into its orthographic form then following conventional text-to-IPA rules to yield a psuedo-trancription, which that would not detect variations (among native speakers) between [i] and [ɛ] as the first vowel of "economic". There is an insufficient corpus of expert IPA productions to make such a program possible (example: there are no expert samples of coda consonant; there are no expert samples of [t] between high vowels or round vowels). For what you are describing, you need actual expert reference values, not well-meaninged attempts at producing reference values (such as the Smalley tapes).
