Third-party cyber evaluations involving OpenAI models
openai.com·14h ago
AI Research
arXiv CS.AI·Jun 29·5 min read
DG^VoiC: Speaker Clustering for Fraud Investigation under Real Call-Centre Conditions
TL;DR
Fraud detection in call centers is challenging due to overlapping speech and background noise. DG^VoiC is a novel speaker clustering technique designed to identify and separate speakers in real-time call center environments.
✦ Why It Matters
Engineers can implement DG^VoiC to enhance speaker identification in noisy environments, improving fraud detection systems.
Key Takeaways
- Utilize DG^VoiC for effective speaker verification in fraud cases.
- Leverage speaker identity as a signal in fraud detection workflows.
- Implement anonymization techniques to protect sensitive information.
- Evaluate clustering performance using metrics like AMI and ARI.
How It Works
DG^VoiC combines several techniques to cluster speakers from call center audio. It starts with anonymization to protect sensitive data, followed by preprocessing of the audio to enhance speech features.
The framework then extracts speaker embeddings using a sliding-window approach, which captures speaker characteristics over time. Finally, it applies cosine similarity to group similar speaker embeddings, effectively identifying repeated speakers across different calls.