Azure Cognitive Services Speaker Identification problem ResultReason.Canceled

Bálint Tahi 6

Hi all,

I am trying out Azure's Speaker Identification (and also Verification).
I was able to Enroll with all VoiceProfileTypes speakers, so lets say, that one is working. I am also able to use the Verification (with a given profile ID + type + same passphrase), then it returns with an EnrolledSpeaker. So this should be fine as well.

However, I can not make the TextIndependentIdentification work.
If I have two different profiles and I am using this code:

var model = SpeakerIdentificationModel.FromProfiles(profiles);
Console.WriteLine("Speak some text to identify who it is from your list of enrolled speakers."); 
var result = await speakerRecognizer.RecognizeOnceAsync(model);

The ResultReason is Canceled. Allways. It doesn't matter who is speaking or for how long or what. The details are:
Bad request (400), Profile is not enrolled. But it is, for sure.

Is there anyone who had the same problem before?

I am using 1.20 API and C# .NET 6.0 WPF.

Kind regards,
Balint

romungi-MSFT 43,696 Reputation points Microsoft Employee

2022-03-03T08:11:16.807+00:00

@Bálint Tahi Have you enrolled the speakers again using voice profile type as Text Independent before passing profiles to SpeakerIdentificationModel.FromProfiles()?
Bálint Tahi 6 Reputation points

2022-03-03T08:26:46.013+00:00
Thanks for your replay.

Sure, I have two enrolled speakers with VoiceProfileType.TextIndependentIdentification, which I get with:

var profiles = await client.GetAllProfilesAsync(VoiceProfileType.TextIndependentIdentification);

When I had only one profile, it was working. But with more than 1 profile it is not.
And the other strange thing is the Verification (please note here, I have enrolled profiles for both types ofc). When I use the SAME passphrase what I used during enrollment it is working, but if I use something else from the list of

var phraseResult = await client.GetActivationPhrasesAsync((VoiceProfileType)speaker.Type, "en-us");

It is not working. Same applies if I use TextIndependentVerification and TextDependentVerification.
And currently basically I have only one profile for both (mine) and +1 for TextIndependentIdentification only for testing. So all together 4 profiles.... and it is not working.

I am not sure what I am doing wrong.
romungi-MSFT 43,696 Reputation points Microsoft Employee

2022-03-04T10:51:06.353+00:00

For a text dependent profile the same passphrase needs to be used for enrollment and verification phases. So, in this case it seems to behave as expected. As per the document.

Text-dependent verification means that speakers need to choose the same passphrase to use during both enrollment and verification phases

The other scenario though with speaker identification model is valid and you are getting the right profile to pass to the method SpeakerIdentificationModel.FromProfiles(profiles)

Would request to report the issue on the SDK github repo to check for any issues with the SDK.
Bálint Tahi 6 Reputation points

2022-03-04T10:54:23.257+00:00

I know. But where could be the problem then?
romungi-MSFT 43,696 Reputation points Microsoft Employee

2022-03-04T12:48:00.543+00:00

I think the profile did not enroll correctly or the voice signature is not properly captured for the profile. Have you enrolled with audio more than 20 seconds?

Share via

Azure Cognitive Services Speaker Identification problem ResultReason.Canceled