The U.S. government’s Intelligence Advanced Research Projects Activity (IARPA) has awarded a multi-million contract to Honeywell to lead the development of real-time speech anonymizing technology. Honeywell’s Real-time Anonymization & Speech Protection (RASP) technology can turn spoken words into anonymized speech to protect people from exploitation and help convert heavily accented dialogue into understandable words. This could allow an air traffic controller, for example, to understand radio transmissions from a non-native English speaker. 

Honeywell's RASP work is part of IARPA’s Anonymous Real-Time Speech (ARTS) program. The company's goal is to “safeguard users from unintended speech signals and support real-time communication in industries such as aviation.

“By anonymizing speech, RASP is designed to help address the growing risk of exploitation and privacy threats that may be associated with devices such as smartphones, digital assistants, televisions, and smart appliances that contain microphones that may be perpetually listening. It will aim to protect an individual’s identity, as well as attributes like dialect, gender, and age, along with dynamic traits such as fear, stress, and anger.”

The three-year ARTS program is focused on English first and will then examine Spanish and other spoken languages. Honeywell is leading a team from the University of Rochester, Texas A&M University, and the University of Texas at Dallas.

“With voice technology becoming commonplace in every home, office, and public space, our voices can reveal key information about our identity and emotions that may not be appropriate for others to know,” said Tor Finseth, principal investigator, RASP, Honeywell Aerospace Technologies. “Until now, no one has been able to anonymize speech in real-time while also making it easy to understand. Through our RASP system, Honeywell aims to meet U.S. government requirements to protect against speaker identification tools, human listeners, and machine-learning assessments. We also aim to improve anonymous speech latency, understandability, and naturalness.”