The Data Exchange with Ben Lorica Why Voice Security Is Your Next Big Problem
Jul 10, 2025
Yishay Carmiel and Roy Zanbel, co-founders of Apollo Defend, dive into the rapidly evolving landscape of voice AI security. They discuss the alarming implications of voice cloning technology, emphasizing its potential misuse and the urgent need for protective measures. The conversation highlights advancements in human-like speech generation and the complexities of defending against deepfake audio attacks. With voice agents proliferating in customer service, they stress the necessity of robust security measures to safeguard personal authenticity and data privacy.
AI Snips
Chapters
Transcript
Episode notes
Evolution of Voice AI Models
- Voice AI is evolving from cascading models to full speech-to-speech models integrating audio LLMs.
- This shift will enable more direct and sophisticated voice applications beyond what text-based systems can do.
Advances in Realistic Speech Synthesis
- Realistic AI-generated speech has unlocked conversational applications and increased language support.
- Text-to-speech technology is much improved but achieving fully human-like voice still depends on use cases.
Ease of Voice Cloning
- Voice cloning requires as little as 5 to 10 seconds of clean, articulate speech.
- Even non-native speakers' accents and tone subtleties are increasingly captured as models improve.