Your Korean speech recognition is grading all answers as correct

Question

Your Korean speech recognition is grading all answers as correct

Callan 0

The Korean speech recognition is grading all inputs as correct. We have native Korean speakers saying Korean words deliberately incorrectly and it's marking them as correct.

Something changed in your service a few weeks ago because previously it was working.

It also applies to some other languages. We are paying for a service that no longer works - please urgently investigate

Rohit Mungi 49,126 Reputation points Microsoft Employee Moderator

2024-08-20T11:57:00.0566667+00:00

@Callan Could you add details of text used, along with the region in which your resource is located along with the API used and its options used when the request is passed to the API. Since this is an issue specific to a language, these details will help product teams identify any potential issue with this feature. Also, since you mentioned the same worked previously, you could add details or response of previous requests which will help in the team's investigation.

If this is really an urgent issue, then please raise a support case through azure portal with the above details to get immediate support. Thanks!!
Callan 0 Reputation points

2024-08-20T13:28:08.2433333+00:00
Hi,

It can be can reproduced on the official Azure website, so we feel the region and nature of the API called is irrelevent.

It is recreateable with any Korean word as the target, and any other Korean word being spoken.

For example, this could be the word you're trying to say: 행복한 (haengbokhan), and saying 큰 (keun) is marked as 100% correct.

We can't give previously requests as we do not store them from our users.
Rohit Mungi 49,126 Reputation points Microsoft Employee Moderator

2024-08-22T08:48:32.38+00:00

@Callan I think you are referring to pronunciation assessment screen from speech studio. Did you only try the words as text for assessment or did you use them in a much longer sentence? AFAIK there is a sample on the same screen and if we use that and try to pronounce the assessment seems to be a realistic score. I am checking if just using the words gives better results from experts from this area. Thanks!!

1 answer

Your answer

Rohit Mungi 49,126 Reputation points Microsoft Employee Moderator

2024-08-20T11:57:00.0566667+00:00

@Callan Could you add details of text used, along with the region in which your resource is located along with the API used and its options used when the request is passed to the API. Since this is an issue specific to a language, these details will help product teams identify any potential issue with this feature. Also, since you mentioned the same worked previously, you could add details or response of previous requests which will help in the team's investigation.

If this is really an urgent issue, then please raise a support case through azure portal with the above details to get immediate support. Thanks!!
Callan 0 Reputation points

2024-08-20T13:28:08.2433333+00:00

Hi,

It can be can reproduced on the official Azure website, so we feel the region and nature of the API called is irrelevent.

It is recreateable with any Korean word as the target, and any other Korean word being spoken.

For example, this could be the word you're trying to say: 행복한 (haengbokhan), and saying 큰 (keun) is marked as 100% correct.

We can't give previously requests as we do not store them from our users.
Rohit Mungi 49,126 Reputation points Microsoft Employee Moderator

2024-08-22T08:48:32.38+00:00

@Callan I think you are referring to pronunciation assessment screen from speech studio. Did you only try the words as text for assessment or did you use them in a much longer sentence? AFAIK there is a sample on the same screen and if we use that and try to pronounce the assessment seems to be a realistic score. I am checking if just using the words gives better results from experts from this area. Thanks!!

Answer 1

Callan 0

@Rohit Mungi to be clear we're using voice recognition (speech) input, not text. We have tried it with individual words as well as sentences. Both of these were working in the past, now they do not work and pass everything as correct.
This is also the case for Greek and Ukrainian

Share via

Your Korean speech recognition is grading all answers as correct

1 answer

Your answer