Behavox research on continuous learning presented in the International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Prestigious Industry Recognition for Behavox Voice Platform

MONTREAL, May 16, 2022–(BUSINESS WIRE)–Behavox, which provides a suite of security products that help compliance, HR and security teams protect their business and colleagues from malicious actors, announced today that his academic research paper on artificial intelligence “Continuous Learning Using Free-MMI Networks for Speech Recognition” has been accepted by the International Conference on Acoustics, Speech and Signal Processing (ICASSP) . ICASSP is the world’s largest and most comprehensive technical conference focused on signal processing and its applications. Behavox presented this paper in Singapore for IEEE ICASSP 2022 on May 9, 2022.

“Sometimes scientists do research for the sake of science and society is unlikely to benefit from it in the near future. Instead, we have pushed science to a new frontier to find a solution to a current problem” , said Behavox CTO Joseph Benjamin. “We are honored to have received recognition from such a prestigious conference and are very proud of our team and the work they have done.”

The article, written by the machine learning team at Behavox, addresses the challenges of continuously improving speech recognition systems in privacy-preserving environments. This research is directly related to Behavox’s R&D approach to providing clients with the highest quality transcription in the financial market. In FinTech, the language and data that Behavox systems must reliably process is changing rapidly over time, the rise of cryptocurrencies and video communications being just two examples.

Therefore, machine learning models need to be frequently updated on relevant data. To make these updates possible without compromising data privacy, the model must be able to improve incrementally in secure customer environments without access to historical data and data from other customers. This is a difficult problem in machine learning due to catastrophic forgetfulness, where after improving a specific type of data, the model begins to perform poorly on previously learned data.

The Behavox ICASSP article contributes to the field of continuous learning for speech recognition. The authors propose a new algorithm that extends and refines the commonly used technique of learning without forgetting (LWF), which relies on a regularization term and forces neural networks to reduce forgetting. Specifically, the authors designed a novel sequence-level neural network learning loss that can be used in place of conventional point-in-time LWF. They demonstrated a significant reduction in forgetting when the neural network is sequentially tuned across various accents and speaking styles.

Behavox found that the use of voice platforms has increased by 94% year over year and that, on average, almost a third of all workplace communication is done by telephone or videoconference .

This technique has already been successfully applied to its Danish system, where the collaborative work of Behavox linguists, the machine learning team and a customer has enabled Behavox to build its first Danish transcription system that works from reliably on bank data.

About Behavox

Behavox offers a suite of security products that help compliance, HR, and security teams protect their business and colleagues from malicious actors.

Through AI-based analysis of all enterprise communication data, including email, instant messaging, voice and video conferencing platforms, Behavox helps organizations identify illegal behavior, immoral and malicious in the workplace.

Founded in 2014, Behavox is headquartered in Montreal with offices in New York, London, Seattle, Singapore and Tokyo.

See the source version on


Jackie loose,
[email protected], (347) 774 4108

Previous ENDRA Life Sciences will present at the HC Wainwright Global Investment Conference
Next Giovanni van Bronckhorst questions choice of Europa League venue and calls Rangers fan numbers