NeXt-TDNN Speaker Verification

🤖 Model Selection

Choose Model:

Higher channel models provide better accuracy but take longer to process

🎤 Audio Inputs

Audio 1

📁 Click to upload audio file

Audio 2

📁 Click to upload audio file

Verification Result

Embedding Details

Audio 1 Embedding

Dimensions:

Audio 2 Embedding

Dimensions:

Note: Embeddings are normalized unit vectors that represent the speaker's voice characteristics.

About Speaker Verification

This demo uses the NeXt-TDNN neural network to compare voice characteristics between two audio samples. The similarity score ranges from 0% to 100%, where:

70%+ similarity: Very likely the same speaker
50-70% similarity: Possibly the same speaker
Below 50% similarity: Likely different speakers

Quick Start

npm install onnxruntime-web @jaehyun-ko/speaker-verification

// Simple usage
const verifier = new SpeakerVerification();
await verifier.initialize('standard-256');
const result = await verifier.compareAudio(audio1, audio2);
console.log('Similarity:', result.similarity);
            