How to remove the generated 'Sous-titres réalisés para la communauté d'Amara.org'

Question

How to remove the generated 'Sous-titres réalisés para la communauté d'Amara.org'

Amper, Charwin (Contractor) 85

Hi, when using Speech Whisper Model via Batch Transcription. I noticed that there is an extra line of texts added in the JSON output. User's image

The texts line in red are not even spoke or heard in the source audio. The image above is using the French locale but maybe it could happen on other locales also. What could be the reason why it is there and how to remove it?

navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T06:14:44.73+00:00
@Amper, Charwin (Contractor) Welcome to Microsoft Q&A Forum, Thank you for posting your query here!

Could you please let me know which region (location) is your service deployed to ?

Could you please share the sample audio which you are using ?

Also please let me know the steps to reproduce this issue at my end ?
.

Awaiting your reply.
Amper, Charwin (Contractor) 85 Reputation points

2024-01-15T10:21:21.3166667+00:00
Hi,

Thank you for your immediate response.

Could you please let me know which region (location) is your service deployed to ?

WestEurope

Could you please share the sample audio which you are using ?

https://www.transfernow.net/dl/20240115WWgtI1kk

Also please let me know the steps to reproduce this issue at my end ?

I am using the ff.

model - https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/models/base/8cd2b1e5-985c-4c77-bfb0-e6ab1798826b

Whisper Model via Batch Transcription
Amper, Charwin (Contractor) 85 Reputation points

2024-01-15T10:29:32.74+00:00
Hi, here are the info.

Could you please let me know which region (location) is your service deployed to ?

westeurope

Could you please share the sample audio which you are using ?

https://www.transfernow.net/dl/20240115WWgtI1kk

Also please let me know the steps to reproduce this issue at my end ?

Model - https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/models/base/8cd2b1e5-985c-4c77-bfb0-e6ab1798826b

Speech Whisper Model Via Batch Transcription
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T10:30:16.1366667+00:00

@Amper, Charwin (Contractor) Thanks for your reply. I will look into this and get back to you with an update.
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T12:48:39.4833333+00:00
@Amper, Charwin (Contractor) I am unable to reproduce this issue at my end. I used the same french.mp3 file with my speech resource in West Europe region and it worked fine for me.
I am attaching the transcription here for your reference: contenturl_0.txt .
I am sending the curl command for your reference. May I know how did you input the mp3 file in the contentUrls ? Are there any proxy in your network ? Could you try uploading the mp3 into storage and try using this curl command ?

curl -X POST "https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/transcriptions" \ -H "Ocp-Apim-Subscription-Key: 027cXXXXXXXXXde65e" \ -H "Content-Type: application/json" \ -d '{ "contentUrls": ["https://demo123storage.blob.core.windows.net/training/french.mp3"], "properties": {}, "locale": "en-US", "displayName": "Transcription using default model for en-US", "model": { "self": "https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/models/base/8cd2b1e5-985c-4c77-bfb0-e6ab1798826b" } }'

Awaiting your reply.
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T15:32:10.63+00:00

@Amper, Charwin (Contractor) Adding to the above, I also tried using the Batch speech to text using the SpeechStudio and even that worked fine. Please test the same at your end too.
Amper, Charwin (Contractor) 85 Reputation points

2024-01-16T02:52:50.8333333+00:00

Hi, could you please try using the 'fr-Fr' as the locale. Also attached is the result of my Transcription result.txt
Amper, Charwin (Contractor) 85 Reputation points

2024-01-16T06:58:14.8166667+00:00

Hi, Thank you for your assistance.
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-16T07:03:21.59+00:00

@Amper, Charwin (Contractor) Thanks for getting back. Please do not forget to "Accept the answer” and “up-vote” if the below information provided helps you, this can be beneficial to other community members.

Accepted answer

0 additional answers

Your answer

navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T06:14:44.73+00:00

@Amper, Charwin (Contractor) Welcome to Microsoft Q&A Forum, Thank you for posting your query here!

Could you please let me know which region (location) is your service deployed to ?

Could you please share the sample audio which you are using ?

Also please let me know the steps to reproduce this issue at my end ?
.

Awaiting your reply.
Amper, Charwin (Contractor) 85 Reputation points

2024-01-15T10:21:21.3166667+00:00

Hi,

Thank you for your immediate response.

Could you please let me know which region (location) is your service deployed to ?

WestEurope

Could you please share the sample audio which you are using ?

https://www.transfernow.net/dl/20240115WWgtI1kk

Also please let me know the steps to reproduce this issue at my end ?

I am using the ff.

model - https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/models/base/8cd2b1e5-985c-4c77-bfb0-e6ab1798826b

Whisper Model via Batch Transcription
Amper, Charwin (Contractor) 85 Reputation points

2024-01-15T10:29:32.74+00:00

Hi, here are the info.

Could you please let me know which region (location) is your service deployed to ?

westeurope

Could you please share the sample audio which you are using ?

https://www.transfernow.net/dl/20240115WWgtI1kk

Also please let me know the steps to reproduce this issue at my end ?

Model - https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/models/base/8cd2b1e5-985c-4c77-bfb0-e6ab1798826b

Speech Whisper Model Via Batch Transcription
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T10:30:16.1366667+00:00

@Amper, Charwin (Contractor) Thanks for your reply. I will look into this and get back to you with an update.
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T12:48:39.4833333+00:00

@Amper, Charwin (Contractor) I am unable to reproduce this issue at my end. I used the same french.mp3 file with my speech resource in West Europe region and it worked fine for me.
I am attaching the transcription here for your reference: contenturl_0.txt .
I am sending the curl command for your reference. May I know how did you input the mp3 file in the contentUrls ? Are there any proxy in your network ? Could you try uploading the mp3 into storage and try using this curl command ?

curl -X POST "https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/transcriptions" \ -H "Ocp-Apim-Subscription-Key: 027cXXXXXXXXXde65e" \ -H "Content-Type: application/json" \ -d '{ "contentUrls": ["https://demo123storage.blob.core.windows.net/training/french.mp3"], "properties": {}, "locale": "en-US", "displayName": "Transcription using default model for en-US", "model": { "self": "https://westeurope.api.cognitive.microsoft.com/speechtotext/v3.2-preview.1/models/base/8cd2b1e5-985c-4c77-bfb0-e6ab1798826b" } }'

Awaiting your reply.
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-15T15:32:10.63+00:00

@Amper, Charwin (Contractor) Adding to the above, I also tried using the Batch speech to text using the SpeechStudio and even that worked fine. Please test the same at your end too.
Amper, Charwin (Contractor) 85 Reputation points

2024-01-16T02:52:50.8333333+00:00

Hi, could you please try using the 'fr-Fr' as the locale. Also attached is the result of my Transcription result.txt
Amper, Charwin (Contractor) 85 Reputation points

2024-01-16T06:58:14.8166667+00:00

Hi, Thank you for your assistance.
navba-MSFT 27,550 Reputation points Microsoft Employee Moderator

2024-01-16T07:03:21.59+00:00

@Amper, Charwin (Contractor) Thanks for getting back. Please do not forget to "Accept the answer” and “up-vote” if the below information provided helps you, this can be beneficial to other community members.

Answer 1

@Amper, Charwin (Contractor) Thanks for your reply. While I used the fr-Fr locale, I am able to reproduce this issue at my end. I had reached out to the Speech Service Product Group team and got and update that, "This is the hallucination in Whisper model itself because of its training data. translate Sous-titres réalisés para la communauté d'Amara.org to English is Subtitles created by the Amara.org community which is the common pattern observed in Whisper's training data source. So this is an expected behavior.
.
Currently, We are not making changes to Whisper's behavior. Also, there is no workaround on Whisper model in Batch. You could choose other default SR model in batch service to see that meets your quality requirement. Hope this answers.

Share via

How to remove the generated 'Sous-titres réalisés para la communauté d'Amara.org'

0 additional answers

Your answer