Batch enrollment in Amazon Connect Voice ID using audio data from prior calls - Amazon Connect

Batch enrollment in Amazon Connect Voice ID using audio data from prior calls

You can get a jump start on using biometrics by batch enrolling customers who have already consented for biometrics. Using stored audio recordings in your S3 bucket, and a JSON input file that provides the speaker identifier and a link to the audio recordings, you can invoke the Voice ID batch APIs.

To enroll customers programmatically, pass the following data to the API:

  1. The domain ID to specify the domain to associate recordings to.

  2. The location for the output file.

  3. An input file containing a list of speakers. See Input and output file schema for the Speaker Enrollment Job in Amazon Connect Voice ID.

    For each speaker the file must include:

    • A link to a call audio recording in a .wav file with 8KHz sample rate and PCM-16 encoding.

    • The corresponding CustomerSpeakerId for the customer.

    • A channel for the caller in the audio recording. If the audio has multiple channels, you can select only one.

  4. A KMS key to use when writing the output.

  5. A role that Voice ID can assume. It must have access to the S3 bucket where the audio files are stored. This role must have access to any KMS key used to encrypt the files. It must also be able to write to the specified output location and use the KMS key requested for writing the output. Specifically, it must have the following permissions:

    • s3:GetObject on the input bucket.

    • s3:PutObject on the output bucket.

    • kms:Decrypt on the KMS key used for input bucket’s default encryption.

    • kms:Decrypt and kms:GenerateDataKey on the KMS key provided in the input which will be used for writing output file to the output bucket.

    You must have iam:PassRole permissions when making the call and providing the dataAccessRole. To enable confused deputy protection for the dataAccessRole, see Amazon Connect Voice ID cross-service confused deputy prevention.

  6. Optionally, a fraud check skip flag in case you want to skip checks for fraud and voice spoofing on the enrollment audio.

  7. Optionally, specify a fraudster watchlist that you want to perform fraud checks against. If no watchlist is specified, Voice ID uses the default fraudster watchlist for the domain.

  8. Optionally, the fraud threshold in case you want to raise or lower the risk.

  9. Optionally, a flag to re-enroll enrolled customers. This is useful if you want to refresh the audio recording, since the default is to ignore previously enrolled customers.

The batch enrollment returns the CustomerSpeakerId, GeneratedSpeakerId, and associated status for each entry. It stores this data in a JSON file at the output path you specify in the API.

Note

You are charged for enrolling speakers. For more information, see Amazon Connect Voice ID Pricing.