EUSIPCO 2016 has officially recognized the validity of progressive research completed by partners of Behavox from the audio research group of Tampere University of Technology in Finland.
Their paper on the detection of whispering and other vocal fluctuation on voice calls was accepted by EUSIPCO 2016; it was given favorable comment in terms of its unique scientific approach, as well as the high level of detection accuracy that the algorithms can deliver.
The paper proposes a detection algorithm for whispered speech over telephone line based on deep neural networks. The researchers presented a solution that is resistant to significant background noise and also boasts accuracy rates for detection of 91.8%. The reviewers of the paper added that given the extremely low signal-to- distortion ratio, the detection results were impressive.
Detection of whispered speech in the presence of high levels of background noise has applications in fraudulent behavior recognition. For instance, it can serve as an indicator of possible insider trading. The research team continues to apply itself to new advances in detection of keywords and emotion to improve accuracy and detection rates. New improvements will be announced later this year.
Behavox has used these voice detection algorithms successfully in its deployment with financial services clients that need state-of- the-art monitoring systems to protect their institutions against forms of market abuse and generic insider threat.
The paper, first published in the Proceedings of the 24th European Signal Processing Conference (EUSIPCO-2016) in 2016 by EURASIP, can be read in full here: http://www.cs.tut.fi/~diment/papers/Diment16_WHI.pdf