Neural Defend ranked #1 in the world for audio deepfake detection

Neural Defend, a New Delhi company, says its Arcas 290M and Arcas 90M models took first and second place among 28 audio deepfake detection systems on a benchmark run by the voice-AI evaluation firm Podonos. Each scored 99.93% on 4,524 clips, with three errors apiece.

The release says most vendors publish accuracy on test sets they control, so figures cannot be compared. Podonos, backed by NAVER, keeps its answer keys private and scores submitted predictions itself, on the same clips and with the same method.

Of the 28 systems, 18 are commercial products from 16 vendors. Pindrop placed 13th at 95.05%, Hive 17th at 83.53% and Reality Defender 18th at 71.27%. The account comes from a syndicated company press release, with quotes from Neural Defend executives.

Podonos called the result "the closest any system has come to a clean sweep of this set." The leaderboard is public at podonos.com, and Podonos founder Soohyun Bae congratulated Neural Defend on LinkedIn, the release says.

Neural Defend CTO Sivashankar Selvarajan said the company sent predictions without knowing the answers. CDO Sumit Singh said the 90M model matches much larger models from unicorn companies at a fraction of the cost.

CEO Piyush Verma said AI voice fraud is growing fast and global, hitting banks, contact centres, insurers and telecoms. Neural Defend works with clients in India, the Middle East, Southeast Asia and the United States.