Advance Processing

In specific scenarios like encountering poor audio quality or speaker distractions, the ASR might face difficulties accurately transcribing certain segments of a conversation. In such instances, incorrect data labeling can happen, posing a risk of potential data leakage depending on the configuration in use.

Example

In this instance, our expectation is for the customer to furnish the agent with their credit card number. Unfortunately, the ASR could only partially transcribe the numbers due to poor audio quality. Nevertheless, upon manually listening to the audio, all 16 digits can be discerned.

AGENT: May I have your number?

CLIENT: It's 553 444 121.

NLP Result

Due to the quality of the audio, the numbers get labeled as a Phone Number.

CLIENT: It's [PHONE_NUMBER].

Configuration

Upon activation of the Advanced Processing feature, the insight will utilize the entities identified by the NLP Engine to effectively redact sensitive information.

{
    "sid": "NER006",
    "name": "NER_Processing",
    "active": true,
    "agent_location": "left",
    "inbound": true,
    "scrubaudio": true,
    "scrubtext": true,
    "insights":[
        {
            "name": "Credit_Card",
            "label": "[CREDIT-CARD]",
            "active": true,
            "description": "insight for credit card validation",
            "leg": "agent",
            "validation": true,
            "real_time": true,
            "trigger":{
                "expression": "(\"can i\" | \"may i\" | digit)  ~> number",
                "redaction": {
                    "type": "all"
                }
            },
            "response": {
                "expression": "\\b\\d{3,16}\\b",
                "before": 2,
                "after": 2,
                "advanced": true,
                "redaction": {
                    "type": "all",
                    "audio": true,
                    "text": true
                },
                "rules":[
                    "PhoneNumber",
                    "Id"
                ]
            }
        }
    ]
}

Updated Result

CLIENT: It's [CREDIT-CARD].