- PasswordGroup - associates the speakers that use the same password inside the same organization identified with specific API Key. Organizations may have got more than one PasswordGroup corresponding to the same password.
- PasswordGroupId - is an unique name of a PasswordGroup that will be used for speaker assignment. PasswordGroupId must be exclusive only inside one organization.
- VoicePrint - is a representation of the speaker inside VoicePIN system. It must correspond to only one passwordGroup.
- VoicePrintId - UUID of the VoicePrint. It is used for API calls associated to speaker enrollment and verification, as well as for resetting and removing the VoicePrint.
- Verification Score - Is an output of the verification method. It's the representation of likelihood that the verification attempt matches user statistical model (voiceprint). Higher values mean more reliable verification attempt.
- Verification Doubt Score - Is the one of the results of the verification method. The score is the maximum of similarity measures between the verification recording and previously used ones. Higher values mean more possible fraud attempt.
- Verification Decision - Is an enumerated value returned by the verification method, that should be used to decide whether verifying person should be authenticated or not. Possible values are:
- MATCH - when the speaker is recognized
- MISMATCH - when the speaker is not recognized
- FRAUD - when the playback is detected
- WRONG_PHRASE_SPOKEN - when the content of the utterance in provided audio sample does not match the chosen password
In order to use our biometric solution one has to register voiceprints first. Voiceprint is a set of biometric features that are representing the speaker based on his repeated utterances.
To register a voiceprint using our API, one has to call a POST method on resource: /passwordGroups/{passwordGroupId}. Returned id now will be pointing on a Voiceprint in our database.
By choosing the PasswordGroup one basically decides which password will be used for upcoming verifications.
It is not forbidden to add more than one representation of speaker to the same PasswordGroup but it is undesirable. If it is still necessary from whatever reasons (for example if we are aware that the same speaker will use different devices for biometric verification) then it is highly advised to make use of externalId while adding new voiceprints. It will increase the verification system usability for that speaker.
Additionally, all samples that are stored for the speaker identifying with the same externalId will be used to detect possible fraud attempt, what is described more particularly in section Playback Detection.
Next step is to perform the enrollment for specified voiceprint. To do so it will be necessary to call a POST method on resource /voiceprints/{voiceprintId}/enrollments at least three times, delivering correctly spelled samples of the utterance spoken by the same speaker and with the same content. Content of the utterance depends on the chosen PasswordGroup.
Person that participate in the registration, should speak naturally, using possibly the same tone of voice during the whole process.
Now, when the voiceprint is trained, it is possible to perform the verifications. It can be done by calling a POST method on resource /voiceprint/{voiceprintId}/verifications. Retrieved result consists of Verification Score, Verification Doubt Score and Verification Decision, described briefly in glossary section. Those parameters will help to decide whether user should be authenticated or not.
Provided samples are tested for possibility of fraud attempt during the verification process. Thus, submitting the same audio sample more than once will cause playback detection and FRAUD decision will be returned as a verification operation result along with Verification Doubt Score, described in glossary section.
To reset Voiceprint and allow speaker to perform enrollment once again, a DELETE method on resource /voiceprints/{voiceprintId}/enrollments should be used.
Unlike the Remove operation, resetting the Voiceprint does not erase the data used for Playback Detection. That makes this operation safe.
To completely remove the information about specified Voiceprint from database, a DELETE method on resource /voiceprints/{voiceprintId} should be called.
Unlike the Reset operation, removing the Voiceprint also removes the data used for Playback Detection. That makes this operation unsafe.
Wave Sampling frequency: 8000 Hz
Bit depth: 16 bit
Codec: PCM
Channels: 1 (mono)
Length: ~2-5s