Boost Security and Accuracy with a Multi‑Engine Voiceprint Fusion Service

Background

In high‑security scenarios such as finance, security, and intelligent customer service, voiceprint recognition is an important biometric authentication method. However, single‑vendor solutions face challenges: unstable enrollment, varying real‑time and accuracy requirements, and difficulty unifying commercial services with self‑developed models.

Product Overview

We built a voiceprint fusion service that integrates Tencent Cloud, iFlytek, and a self‑developed engine, using flexible scheduling to provide higher availability, precision, and flexibility.

The overall solution consists of three core modules:

1️⃣ Engine Integration

Tencent Cloud Voiceprint : high‑availability, standardized API for large‑scale concurrent calls, meeting finance‑grade compliance.

iFlytek Voiceprint : excellent performance on Mandarin and multiple dialects, optimized for domestic language diversity.

Self‑Developed Voiceprint : based on a proprietary deep‑learning framework, supports basic voiceprint capability and horizontal scaling.

2️⃣ Intelligent Scheduling Strategy

Multi‑engine concurrent verification : the same audio can be submitted to multiple engines simultaneously, reducing single‑engine risk.

Unified result output : aggregate and compare results from all engines into a consistent API response.

Scalable foundation : supports elastic node expansion and load balancing.

3️⃣ Flexible Access

Cloud service support : deployed on high‑availability cloud infrastructure, on‑demand scaling.

Unified call entry : a single API endpoint abstracts vendor‑specific integration.

Technical Highlights

Multi‑engine fusion algorithm : submit audio to several engines, aggregate results for higher availability and accuracy.

Custom self‑developed model : noise robustness and short‑phrase compatibility.

High‑concurrency low‑latency architecture : micro‑service design, asynchronous task scheduling, distributed queues, auto‑failover and retry.

Unified API and management : consistent API, multi‑tenant isolation.

Typical Application Scenarios

Phone banking & remote account opening : multi‑channel verification, automatic model switching under poor network conditions.

Smart government identity verification : voiceprint + facial multi‑factor authentication, deployable in on‑premise data centers.

Enterprise attendance & access control : precise employee identification, integration with attendance systems.

Usage Specification

1️⃣ Interfaces

Create voiceprint repository

Add audio feature

Update audio feature

Delete specific audio feature

Delete repository

Feature compare 1:1 (audio vs repository)

Feature compare 1:1 v2 (audio vs audio)

Feature compare 1:N (audio vs repository)

2️⃣ Call Flow

Direct audio‑to‑audio comparison → use 1:1 v2.

Audio vs repository → create repository → add feature → compare 1:1.

Audio vs repository (multiple) → create repository → add feature → compare 1:N.

3️⃣ Example

Create repository

curl -X POST "http://localhost:8080/voiceprintRecognition/createRepository" \
    -H "Content-Type: application/json" \
    -H "x-zcs-version: v1" \
    -H "Authorization:Bearer xxx" \
    -d '{ }'

Add audio feature

base64_output=$(base64 -w 0 test_audio.mp3)
curl -X POST http://localhost:8080/voiceprintRecognition/addAudioFeature \
    -H "Content-Type: application/json" \
    -H "x-zcs-version: v1" \
    -H "Authorization: Bearer xxx" \
    -d '{
        "VoiceFormat": 2,
        "SampleRate": 16000,
        "Data": "'$base64_output'",
        "GroupId": "OtvMnkXnQGGmcSCmHbDycbQaeZtsgkCV"
    }'

Feature compare 1:N

base64_data=$(base64 -w 0 test_audio.mp3)
curl -X POST http://localhost:8080/voiceprintRecognition/compareAudioFeatureN \
    -H "Content-Type: application/json" \
    -H "x-zcs-version: v1" \
    -H "Authorization: Bearer xxx" \
    -d '{
        "VoiceFormat": 2,
        "SampleRate": 16000,
        "Data": "'$base64_data'",
        "TopN": 10,
        "GroupId": "OtvMnkXnQGGmcSCmHbDycbQaeZtsgkCV"
    }'

Cooperation and Access

The core functions are complete and the service is open for pilot integration across industries. Partners can access via the standard API to co‑develop and expand voiceprint applications.