Breaking01 / AI
FEND aims to identify who answers voice AI calls
The proposed framework distinguishes people, voicemail, automated phone menus and AI to help voice agents choose how to respond.

Researchers proposed FEND, a framework for identifying who or what answers outbound voice AI calls, in an arXiv paper published on 16 Nov. It is designed to distinguish people, voicemail, automated phone menus and other AI systems.
According to the paper, FEND analyses audio alongside a live transcript. It defaults to a human label if none of its detectors for automated responses activates. The design returns a final classification within 20 seconds.
The researchers reported analysing 1,000 randomly sampled calls from their own outbound deployment. AI answered 6.9% of calls in that sample. People answered 48.1%.
The researchers reported 100% voicemail detection accuracy on a public benchmark. In that evaluation, no live person was misclassified as voicemail. In a separate evaluation of operational calls, 2.9% of people were incorrectly labelled as automated systems.
The paper explains that standard answering machine detection (AMD) distinguishes only between humans and machines. That design was intended for call centres staffed by people. The researchers argued that voice AI needs different responses for leaving a message, navigating a phone menu and speaking with another AI.
The paper's abstract does not disclose the number of calls used in the public benchmark.