Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Pronunciation Assessment phoneme accuracy score are wrong. #1917

Closed
riiight opened this issue Apr 12, 2023 · 12 comments
Closed

Pronunciation Assessment phoneme accuracy score are wrong. #1917

riiight opened this issue Apr 12, 2023 · 12 comments
Assignees
Labels
accepted Issue moved to product team backlog. Will be closed when addressed. bug Something isn't working pronunciation assessment

Comments

@riiight
Copy link

riiight commented Apr 12, 2023

Describe the bug
If you do a pronunciation assessment on a word and make a clear mistake one of the phonemes or syllables, the scores for all phonemes decreases instead of just the one you made the mistake on.

To Reproduce
Steps to reproduce the behavior:

  1. Go to the Pronunciation Assessment tool in Speech Studio (https://speech.microsoft.com/portal/_____/pronunciationassessmenttool)
  2. Click on the "Enter your own script" tab
  3. Enter "automobile" in the text box to be the word you want to test.
  4. Click "Record audio with microphone" and say "automobile" correctly.
  5. In "Assessment Result", hover over the word "automobile" and see that its scores are nearly 100 across all phonemes.
  6. Try again but this time deliberately say "kertomobile".
  7. In "Assessment Result", hover over the word "automobile" and see that its all its scores are lower, not just the score for "ao" (the first phoneme that you deliberately said wrong).

Expected behavior
I believe that if only one phoneme is wrong, only its score should lower. It appears as though all phoneme scores lower together making them almost useless.

Please correct this behavior as it is vital for true pronunciation assessment. Thanks.

@pankopon
Copy link
Contributor

pankopon commented May 9, 2023

@yulin-li Can you please check.

@riiight
Copy link
Author

riiight commented May 10, 2023

image

Here's a screenshot from when I said, "kertomobile". This is actually not the worst result I've seen. Sometimes, I see the same score across all phonemes. I'd expect to see 0-20 for the first phoneme and then 95+ for the rest of them.

@yulin-li
Copy link
Contributor

@yinhew @wangkenpu could you please help to check?

@yulin-li yulin-li assigned yinhew and unassigned yulin-li May 29, 2023
@wangkenpu
Copy link
Contributor

Thanks for reporting the issue. We are working on it and will let you know once it's fixed.

@pankopon pankopon added bug Something isn't working accepted Issue moved to product team backlog. Will be closed when addressed. and removed in-review In review labels May 30, 2023
@chrisrollins65
Copy link

chrisrollins65 commented May 31, 2023

I came here to report this same thing but it looks like you've already reported it. I posted this issue in the Azure community Q&A but didn't get much of a response there. Here's a video I took of the problem: https://youtu.be/RUwA_ZhWpVU

This is happening both in the Speech Studio as well as in my own application where I integrated the API. For the time being I've had to switch to another pronunciation service since this one isn't working correctly.

@bobir01
Copy link

bobir01 commented Jul 12, 2023

switch to another pronunciation service since this one isn't working corre

hi @chrisrollins65 i am this issue too and some other bugs with STT services can tell which pronunciation APIs you found help and reliable Thanks, you can drop it mardonovbobir9@gmail.com

@chrisrollins65
Copy link

@bobir01 I'm using an API by Language Confidence: https://rapidapi.com/language-confidence-language-confidence-default/api/pronunciation-assessment1

It's not perfect, it's less robust and it's more expensive, but at least it differentiates between phonemes, unlike Azure's solution. That was the main feature I needed for my purposes.

@bobir01
Copy link

bobir01 commented Jul 14, 2023

@bobir01 I'm using an API by Language Confidence: https://rapidapi.com/language-confidence-language-confidence-default/api/pronunciation-assessment1

It's not perfect, it's less robust and it's more expensive, but at least it differentiates between phonemes, unlike Azure's solution. That was the main feature I needed for my purposes.

@chrisrollins65 Thanks

@wangkenpu
Copy link
Contributor

wangkenpu commented Aug 1, 2023

The issue has been fixed. @riiight @chrisrollins65 @bobir01
image

We can close this issue. @pankopon

@bobir01
Copy link

bobir01 commented Aug 1, 2023

The issue has been fixed. @riiight @chrisrollins65 @bobir01 image

We can close this issue. @pankopon

wow thanks,
how about this one Azure/azure-sdk-for-python#31102 (comment)

@wangkenpu
Copy link
Contributor

The issue has been fixed. @riiight @chrisrollins65 @bobir01 image
We can close this issue. @pankopon

wow thanks, how about this one Azure/azure-sdk-for-python#31102 (comment)

Hello @bobir01. Please discuss this issue in the corresponding thread. It is beyond our team's scope.

@glecaros
Copy link
Contributor

glecaros commented Aug 8, 2023

closing the issue as resolved

@glecaros glecaros closed this as completed Aug 8, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
accepted Issue moved to product team backlog. Will be closed when addressed. bug Something isn't working pronunciation assessment
Projects
None yet
Development

No branches or pull requests

8 participants