How do I train speech bot to learn non-English words pronounciation?

Question

Hello, I'm currently trying to train a custom speech-to-text bot to recognize some slangs and jargons in my niche. The language is Korean.

I understand I need to upload a txt file with a word and pronounciation, such as:

segosairus seg is aurus

agkistrodon acast road on

My question is, how do I do that with non-English language? Do I put pronounciation in English alphabet?

Answer 1

@Andrew Kim I think in the case of Korean under custom speech support pronunciation is not supported as per the language support page.

For the other forms that are supported you need to add the text in Korean. For example, for audio + human-labeled transcript the audio and text need to be mentioned in the transcript file referring to the audio file in supported naming format as seen below.

Here, my audio file is in folder 1 with file name 1_0003.wav and the transcript of the text is tab separated written in Korean. Additional entries can be mentioned in separate lines.

1/1_0003.wav 그는 아내를 많이 아낀다.

I hope this helps!!

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.