Voice cloning gets worryingly easy
Researchers and commentators showed that open‑source TTS like Quen3TTS can clone a voice from just ~3 seconds of audio, enabling phone‑call scams and deepfake video interceptions. (youtube.com) More reputable providers like 11 Labs now require live, randomized verification scripts, but the tech is proliferating globally (reportedly with few restrictions in places like China) and tools like Truecaller are being pitched to detect and remotely hang up scam calls. (youtube.com)
Qwen3‑TTS is an open‑source TTS family released by the Qwen team at Alibaba Cloud under an Apache‑2.0 license and published on GitHub with a live demo on Hugging Face. (github.com) Public documentation and community guides list Qwen3‑TTS model variants roughly between 0.6B and 1.7B parameters, advertise multi‑language support (about 10 languages), and report first‑packet streaming latency down to ~97ms. (dev.to) ElevenLabs’ Professional Voice Cloning (PVC) workflow requires users to submit verification audio and to complete liveness checks by reading system‑generated lines, with failed attempts requiring a 24‑hour wait or manual review via support. (elevenlabs.io) Truecaller introduced an “AI Call Scanner” on May 29, 2024 that it says can warn users in real time when a caller’s voice appears AI‑generated, and the rollout has targeted Android users with some features gated behind its Premium subscription. (medianama.com) Industry threat reports show rapid growth in deepfake call fraud: Hiya’s Q4 2024 report put the average U.S. fraud loss at $539 per victim, and other industry summaries documented victims reporting losses exceeding $6,000 in individual deepfake‑enabled incidents. (hiya.com) U.S. regulators and local authorities are already responding — the FTC has run a Voice Cloning Challenge and signaled applying the Telemarketing Sales Rule to AI impersonation, while a 2026 police bulletin from the Royal Borough of Greenwich tied a “lifestyle survey” scam ring to the use of cloned voices. (consumer.ftc.gov)
