training with mixed language in custom-stt(English & Korean)

VPA 21

Hi, I am working on training korean custom-stt, but in the training data , there are a few english words mixed in it. Some of them are well processed and accepted as train data but others get rejected such as winder, insulator, gripper, rewinding. below is the warning I get when I uploaded the mixed data.

message: Error: Generation of pronunciation of word "rewinding" failed. Please try an alternate spelling form.

I know there may be a limitation for processing two language at once, but since quite many english words have accepted as train data I am wondering if there is a waythrough to add english word that has been rejected

Thanks in advance

Ramr-msft 17,651 Reputation points

2022-10-15T11:34:16.677+00:00

@VPA Thanks for the question. Are you using the audio or text for training data? . Here is the document that could help to prepare the structured text markdown prompt.

https://video2.skills-academy.com/en-us/azure/cognitive-services/speech-service/how-to-custom-speech-test-and-train#structured-text-data-for-training
VPA 21 Reputation points

2022-10-15T12:11:45.34+00:00

I am using audio + human labeled transcript data(90% korean + 10% English) and some of the english words ( winder, insulator, gripper, rewinding) in text caused the problem mentioned above. would structured-text data help solving this kind of problem?
VPA 21 Reputation points

2022-10-25T10:41:12.637+00:00

The structured-text data didn't help solving my problem. It still shows me error when I submit rejected english words
Elias Salazar Zeledon (Manpower Costa Rica S A) 0 Reputation points Microsoft Vendor

2024-05-28T23:25:13.9066667+00:00

were you able to solve it?

Share via

training with mixed language in custom-stt(English & Korean)