Artificial intelligence firm Verint has developed voice recognition software that wil help Interpol find criminals using only their voice.
The police network plans to employ the voice tech to try and catch more wrongdoers from its base in Lyon, despite it coming under fire from privacy campaigners for invading the privacy of those it’s recording. This is because it can not only detect ‘direct’ voices from phone conversations and social message videos, for example, but it can also identify people from ambient conversation – such as in a restaurant or event with more than one person talking.
Verint’s voice recognition platform uses speech analysis algorithms to filter voice recordings, identifying gender, age, language, and accent – significantly boosting what’s already possible and making it more reliable to find criminals.
It will then match what it hears and identifies against a database of criminals. For example, if it’s recording a conversation from a drug deal, it can pick up unique identifiers, which can be compared against criminal profiles to find those involved.
The system has been extensively tested for accuracy already (after all, it would’t want to identify a completely innocent person as being a drug baron), but it will expand upon last year’s trials with fresh pilots in Brussels, just to make sure it really does work.
It was developed using recorded voice samples. The algorithms filtered through the recordings, learning certain nuances. It worked in tandem with a “processing chain built on open-sourced architecture,” which was used to create the actual process of recognition.
It will apply other factors into finding the criminal too, such as geolocation and language relevance to try and find the true perpetrator.
If the sound is coming via a video, the clever algorithms can separate the sound from the visuals, split it into mono sound and then format it into 16 kilohertz WAV files so it can be searched and tagged.
Verint also worked with Airbus, Singular Logic, and Nuance, plus used keyword spotting components from Sail and Swiss research nonprofit IDIAP.
