What is Voice Fingerprinting?

Voice Fingerprinting is a backup speaker identification method used in Practice by Numbers' Call AI system. It analyzes an employee’s unique voice characteristics and builds a voice profile (or "fingerprint") to help the AI identify them across future calls.

Key Benefits:

Increases overall speaker identification accuracy
Automatically detects which employee handled a call, even if they don’t say their name
Helps train the AI to learn employee voices over time

Important: Always confirm employee consent before creating a voice fingerprint.

Consistency Reminder: If you enroll one employee for voice fingerprinting, you must enroll all employees who handle calls. Incomplete enrollment can lead to inconsistent attribution, misidentification, and skewed analytics.

Where to Access Voice Fingerprinting

There are two access points, each with a distinct purpose:

1. Voice Settings → Voice Fingerprints

Use this page to create, manage, retrain, and upload voice samples for all staff.

2. Call AI Dashboard → Call Details

Use this for call-level tagging, where you assign a speaker to a fingerprint for a specific call.

1. Voice Settings → Voice Fingerprints

Use this centralized settings page to manage all speaker voiceprints, improve accuracy through retraining, and add new voice samples.

How to Access:

Navigate to Settings > Voice > Voice Fingerprints

What You Can Do Here:

Retrain the Model

Click the green brain icon to retrain the voiceprint model.

Tooltip: Retrain model to potentially improve accuracy

Retraining may improve model accuracy with existing samples.

This is useful when you’ve already collected voice data but want the model to process it again for improved matching performance.

Add Voice Samples

Click the microphone icon to add more samples and improve fingerprint accuracy.

Tooltip: Add voice samples to improve fingerprint accuracy

This opens the Upload Voice Sample modal.

How to Upload a Voice Sample

You’ll have two options:

Record Audio directly
Upload Files manually

Steps to Record:

Choose a category like:
- English – Verification Paragraphs
- English – Natural Speech
- Español – Verificación De Voz
Select at least 2 sample texts for training.
Follow the prompts to record and upload your voice clips.

These samples help the system better identify the speaker across future calls.

Use Case Summary: The Voice Fingerprints Settings page is best for administrative management of voice data. It allows you to update, retrain, and refine fingerprint profiles for better speaker recognition platform-wide.

2. Call AI Dashboard → Call Details

Use this path when you want to identify a speaker within a specific call.

How to Access:

Go to the Call AI module.

In the left-hand menu, click Call Details.
Locate the specific call you want to tag.
Click the “Add Voice” button.

This will open the voice fingerprinting screen, where you can assign an existing voiceprint to the speaker on that call.

Use Case Summary:
This method is ideal for one-off tagging, such as identifying a new speaker during post-call reviews.

How to Fingerprint a Call

When you click "Add Voice", the system walks you through two steps:

Step 1: Listen and Select Channel

Listen to the left and right audio channels.
Determine which one is the employee (Image 1).

🔁 Inbound Calls:

Left Channel = Patient
Right Channel = Agent

🔁 Outbound Calls:

Left Channel = Agent
Right Channel = Patient

Step 2: Select Employee

Use the dropdown to choose the employee whose voice is being fingerprinted.

Once submitted, this audio sample is added to their fingerprint profile.

How It Works

1. Name Detection (Primary Method)

The primary method for identifying who handled a call is based on spoken introductions.

AI transcribes the call and looks for phrases like:
“Hi, this is John from the office…”
It matches the detected name to staff records.
This method achieves 90%+ accuracy when introductions are clear.

2. Voice Fingerprint (Backup Method)

If the primary method fails (e.g., no intro given), the system falls back to the voice fingerprint.

The system:

Processes Audio – Removes noise and silence.
Analyzes Voice – Examines voice pitch, tone, cadence, etc.
Matches Voice – Compares to stored fingerprints using similarity scores.

Best Practices for High Accuracy

Clear Introductions: Staff should start every call by introducing themselves.
Complete Enrollment: All phone agents must have fingerprints for accurate AI identification.
Quality Samples: Submit 3–5 clear samples per employee.
Check Quality Metrics:
- Good signal-to-noise ratio
- Clear voice (no distortion)
- The sample duration is sufficient

When to Use

Use the Voice Fingerprints Page When:

You're onboarding a new employee who answers or places calls.
You want to improve system-wide speaker recognition.
You need to manage or delete existing voiceprints.
You’re trying to increase the Confidence Score for a specific person.

Use “Add Voice” in Call Details When:

You are reviewing a call where the speaker wasn’t recognized.
You want to match a caller to an existing fingerprint manually.

Best Practice: Start by enrolling staff in Voice Fingerprints, then use Call Details > Add Voice to fine-tune call-level tagging and speaker attribution.