I can't make new custom voice model.

golden 21

I have prepared about 5000 voice files to create a new custom voice model.
These audio files follow the guidelines specified in the azure documentation.
And I proceeded as described in the documentation, but I was unable to create a new model.
This is because the Next button is disabled as shown below.
As I couldn't find any messages why it was disabled, I couldn't proceed any further.
So I ask for help in this forum.

46013-custom-voice-chrome-2020-12-08-%E1%84%8B%E1%85%A9%E1%84%8C%E1%85%A5%E1%86%AB-10-35-50.png

46071-custom-voice-chrome-2020-12-08-%E1%84%8B%E1%85%A9%E1%84%8C%E1%85%A5%E1%86%AB-10-35-54.png

46052-custom-voice-chrome-2020-12-08-%E1%84%8B%E1%85%A9%E1%84%8C%E1%85%A5%E1%86%AB-10-35-59.png

46081-custom-voice-chrome-2020-12-08-%E1%84%8B%E1%85%A9%E1%84%8C%E1%85%A5%E1%86%AB-10-36-06.png

46022-custom-voice-chrome-2020-12-08-%E1%84%8B%E1%85%A9%E1%84%8C%E1%85%A5%E1%86%AB-10-36-12.png

romungi-MSFT 44,166 Reputation points Microsoft Employee

2020-12-08T06:57:56.75+00:00

@golden Do you have any duplicate wav files across the datasets? I think this could be the reason, according to the requirements for all other languages other than en-US and zh-CN there is a limit of atleast 2000+ utterances.
golden 21 Reputation points

2020-12-08T12:25:24.183+00:00

The language I am trying to do is ko-KR. So I prepared about 5000 wav files. (1000 for each zip files)
There are no duplicate files here.

2000+ utterances doesn't mean 2000+ wav?
golden 21 Reputation points

2020-12-08T12:29:06.97+00:00

I downloaded the material here.
https://www.kaggle.com/bryanpark/korean-single-speaker-speech-dataset

And I modified the transcript.v.1.4.txt file to fit the format required by azure.

There were no errors in this part after uploading.
romungi-MSFT 44,166 Reputation points Microsoft Employee

2020-12-09T11:16:25.953+00:00

@golden I have tried to use the same dataset and uploaded about 2200 utterances and found the upload processing was successful but when I wanted to create the model the same behavior was seen. I later found that a report of the dataset contained some issues with the utterances which may be causing this, could you also check if the similar report contains issues in processing?
golden 21 Reputation points

2020-12-10T04:14:28.81+00:00

I downloaded the report file and checked it.
It also fixed errors on the list of issues and most warring.
Here are the issues I haven't solved:

-Dataset,,Warning,Script,"Exclamation script lines are less than 1/6 of total script lines.

-Dataset,,Warning,Script,"Question script lines are less than 1/6 of total script lines.

-Warning,Mismatch,"Start silence is shorter than 100 ms.",AutoFixed,"The start silence is extended to 100 ms. Please downloaded normalized dataset to review."

-Warning,Mismatch,"End silence is shorter than 100 ms.",AutoFixed,"The end silence is extended to 100 ms. Please downloaded normalized dataset to review."

Shouldn't there be any issue list?
Then I can no longer test the prototype...

Accepted answer

romungi-MSFT 44,166 Reputation points Microsoft Employee

2020-12-10T11:34:07.597+00:00

@golden Having reviewed the setup of speech studio with our team it looks like the NEXT button is not enabled because custom voice for korean is a gated technology where our team allows access to use this only when the customer fills out the form mentioned here.

Once your request is accepted you can continue to use the custom voice feature from speech studio and proceed further.
Please sign in to rate this answer.
golden 21 Reputation points

2020-12-10T12:20:30.41+00:00

Thank you.
I wasn't aware of these limitations.
Can you please tell me which document explains this?
Help them develop the power to solve alone in other similar situations.
Thanks again.

romungi-MSFT 44,166 Reputation points Microsoft Employee

2020-12-10T14:09:46.957+00:00

@golden The custom voice documentation page mentions this but it is not apparent if you try to use the portal directly. Since Korean voice is neural this request is required to enable the NEXT button and proceed further.
Sign in to comment

Use comments to ask for clarification, additional information, or improvements to the question.

Share via

I can't make new custom voice model.

0 additional answers

Your answer