Error handling¶

Every exception the client raises inherits from VeriClientError, so a single except catches anything the library can throw:

from vericlient.exceptions import VeriClientError

try:
    client.generate_credential(model_input)
except VeriClientError as error:
    print(f"The call did not go through: {error}")

Most of the time you want something narrower. The tree is shallow on purpose.

Shared¶

Exception	Raised when
`VeriClientError`	Base class. Never raised directly.
`AuthorizationError`	The API key is missing or rejected.
`ServerError`	The service failed, or answered something the client cannot interpret — including an error body that is not JSON.
`UnsupportedMediaTypeError`	The service does not accept the media type that was sent.
`InvalidCredentialError`	A credential is malformed or not accepted.

das-Peak¶

All of these inherit from DaspeakError. The ones about the audio itself inherit from AudioInputError, so you can catch every "this recording will not do" case at once:

from vericlient.daspeak.exceptions import AudioInputError

Exception	Raised when
`TooManyAudioChannelsError`	The audio has more than two channels.
`UnsupportedSampleRateError`	The sample rate is neither 8 kHz nor 16 kHz.
`AudioDurationTooLongError`	The recording is longer than 30 seconds.
`UnsupportedAudioCodecError`	The codec is not `PCM_16`, `ULAW` or `ALAW`.
`SignalNoiseRatioError`	The recording is too noisy.
`NetSpeechDurationIsNotEnoughError`	Less than 3 seconds of actual speech. Carries the duration that was detected.
`InsufficientQualityError`	The quality is too low, or more than one speaker is present.
`InvalidSpecifiedChannelError`	The requested channel is neither 1 nor 2.
`CalibrationNotAvailableError`	The model does not support that calibration. Use `get_model_calibrations()` to see which ones it does.
`ModelNotAvailableError`	No model exists with that hash.

Net speech, not duration

NetSpeechDurationIsNotEnoughError is about speech, not file length. A 10 second recording with 2 seconds of talking in it will still be rejected. The exception message carries the duration the service actually detected.

VCSP¶

All of these inherit from VcspError.

Exception	Raised when
`EmptyFileError`	The sample carries no content.
`InvalidClaimsError`	The claims do not satisfy the credential configuration.
`InvalidAssuranceError`	The assurance does not satisfy the assurance method.
`InvalidTagsError`	One of the tags is not valid.
`InvalidCredentialConfigurationUrnError`	No credential configuration with that URN.
`InvalidAssuranceMethodUrnError`	No assurance method with that URN.
`CredentialConfigurationUrnAlreadyAssignedError`	The subject already holds a credential with that configuration.
`InvalidAudioFormatError`	The audio format is not accepted.
`InvalidSnrError`	The audio is too noisy.
`VoiceDurationIsNotEnoughError`	Not enough speech in the sample.
`InsufficientQualityError`	The sample quality is too low.
`FaceNotFoundError`	No face in the image.
`MoreThanOneFaceError`	More than one face in the image.
`FaceTooSmallError`	The face is too small to be usable.
`FaceAlignmentError`	The face is not aligned well enough.
`AssuranceValidationError`	The sample did not meet the assurance thresholds.
`AssuranceMethodNotFoundError`	No assurance method with that URN.
`AccountNotFoundError`	No account with that subject id.
`CredentialNotFoundError`	No credential with that id.
`RequestValidationError`	The request is malformed.

A note on media types¶

VCSP answers 500 when the media type declared in the request does not match the content it receives, so a client-side mistake can look like a server fault. The client guards against this by declaring the media type per multipart part, inferring it from the file extension for a path and from the magic bytes for a bytes object. Where the guess would be wrong, set it explicitly:

from vericlient.vcsp.models import EnrollmentInput

EnrollmentInput(sample=raw_bytes, applicant=applicant, content_type="audio/wav")