Flow block: Set Voice ID

Description

• Enables audio streaming and sets thresholds for voice authentication and detection of fraudsters in a watchlist. For more information about this feature, see Voice ID.

• Sends audio to Amazon Connect Voice ID to verify the caller's identity and match against fraudsters in watch list, as soon as the call is connected to a flow.

• Use a Play prompt block before Set Voice ID to stream audio properly. You can edit it to include a simple message such as "Welcome."

• Use a Set contact attributes block after Set Voice ID to set the customer ID for the caller.

  The CustomerId may be a customer number from your CRM, for example. You can create a Lambda function to pull the unique customer ID of the caller from your CRM system. Voice ID uses this attribute as the CustomerSpeakerId for the caller.

  CustomerId can be an alphanumeric value. It supports only _ and - (underscore and hyphen) special characters. It does not need to be UUID. For more information, see CustomerSpeakerId in the Speaker data type.

• Use a Check Voice ID block after Set Voice ID to branch based on the results of the enrollment check, authentication, or fraud detection.

• For information about how to use Set Voice ID in a flow, along with Check Voice ID and Set contact attributes, see Step 2: Create a new Voice ID domain and encryption key in Enable Voice ID.

Supported channels

The following table lists how this block routes a contact who is using the specified channel.

Channel	Supported?
Voice	Yes
Chat	No - Error branch
Task	No - Error branch

Flow types

You can use this block in the following flow types:

• Inbound flow

• Customer queue flow

• Customer whisper flow

• Outbound whisper flow

• Agent whisper flow

• Transfer to Agent flow

• Transfer to Queue flow

Properties

The following image shows the Properties page of the Set Voice ID block. It shows the Voice authentication section. In this example, the Authentication Threshold is set to 90. This is the recommended threshold.

Start streaming audio for Voice ID

When this option is selected, Amazon Connect begins streaming audio from the customer's channel to Voice ID.

You can add this block several places in a flow, but after Start streaming audio is selected, it cannot be disabled, even if later in the flow there are other Set Voice ID blocks that do not have it enabled.

Voice authentication

Authentication threshold: When Voice ID compares the voiceprint of the caller to the enrolled voiceprint of the claimed identity, it generates an authentication score between 0-100. This score indicates the confidence of a match. You can configure a threshold for the score which indicates whether the caller is authenticated. The default threshold of 90 provides high security for most cases.

• If the authentication score is below the configured threshold, Voice ID treats the call as not authenticated.

• If the authentication score is above the configured threshold, Voice ID treats the call as authenticated.

For example, if the person is sick and calling from a mobile device in their car, the authentication score is going to be slightly lower than when the person is well and calling from a quiet room. If an imposter is calling, the authentication score is much lower.

Authentication response time

You can set the authentication response time between 5 and 10 seconds, which determines how quickly you want Voice ID authentication analysis to complete. Lowering it makes the response time faster at the tradeoff of lower accuracy. When you're using self-service IVR options where callers do not talk a lot, you may want to reduce this time. You can then increase the time if the call needs to be transferred to an agent.

The following image shows the Authentication Response time section of the block. The response time is set manually to 10 seconds.

Choose Set dynamically to set the authentication threshold based on certain criteria. For example, you may want to raise the threshold based on the membership level of the customer, or the type of transaction or information they are calling about.

Fraud detection

The threshold you set for fraud detection is used to measure risk. Scores higher than the threshold are reported as higher risk. Scores lower than the threshold are reported as lower risk. Raising the threshold lowers false positive rates (makes result more certain), but raises false negative rates

Choose Set dynamically to set the fraud threshold based on certain criteria. For example, you may want to lower the threshold for high wealth customers, or the type of transaction or information they are calling about.

The watch list you select is used when evaluating the voice session. Choose Use default watch list to use your domain's default watch list. For Set manually, the watch list ID must be 22 alphanumeric characters.

Similarly for the watch list, choose Set dynamically to set the watch list based on criteria given. For example, you may want to use a stricter watch list given the type of transaction or information they are calling about.

Configuration tips

• For the Authentication threshold, we recommend that you start with the default of 90 and adjust until you find a good balance for your business.

  Every time you increase the value of the Authentication threshold beyond the default of 90, there's a tradeoff:

  • The higher the threshold, the greater the false reject rate (FRR), that is, the likelihood that an agent will need to verify the customer's identity.

    For example, if you set it too high, such as greater than 95, agents will need to verify every customer's identity.

  • The lower the threshold, the greater the false acceptance rate (FAR), that is, the likelihood that Voice ID will incorrectly accept an access attempt by an unauthorized caller.

• When Voice ID verifies that the voice belongs to the enrolled customer, it returns a status of Authenticated. Add a Check Voice ID block to you flow branch based on the returned status.

• For the Fraud threshold, we recommend that you start with the default of 50 and adjust until you find a good balance for your business.

  If the caller's score is above the threshold, it indicates there's a higher risk for fraud in that call.

• For the Fraud watch list, the format is validated when the flow is published.

  • If a watch list is dynamically set and the format is not valid, the contact is routed down the Error branch of the Set Voice ID block.

  • If a watch list ID is set manually or dynamically with a valid format but the watch list is not available in the Voice ID domain of the instance, the contact is routed down the Error branch of Check Voice ID block when the Check Voice ID block is used later in the flow.

Configured block

The following image shows an example of what this block looks like when it is configured. It has the following branches: Success and Error.

More information

See the following topic for more information about this block:

• Use real-time caller authentication with Voice ID

• Flow block: Check Voice ID

• Use Voice ID