Interpol working on voice database
26 June 2018 12:00 GMT

Interpol will soon be launching a new software platform that aims to identify violators of the law by using voiceprints . 

The largest police network in the world has entered the era of voice technologies. He experimentally implemented a set of software that would match voice samples over the phone or voice messages from social networks, to voice recordings of criminals stored in a very large database shared by government agencies. 'law application.

Dubbed the SIIP (Speaker Identification Integrated Project), in 2014, at the initiative of the European Commission, with the collaboration of Interpol, the voice software platform, as described by its developers, uses algorithms of analysis of speech to filter vocal samples by gender, age, language and accent.

It is the programming of an algorithmic approach that will allow the identification of suspects by a process of linking a speaker recognition solution, which compares a voice recorded during wiretapping. recordings on crime scenes or from sound files from social networks, with those in the database.

The developers successfully completed the field-based SIIP performance tests in March 2017 in Lisbon, Portugal. Researchers, forensic scientists, judicial police investigators and experts in the field of speaker recognition were able to witness successful cross-referencing demonstrations of voice recordings from social networks using key identifiers , such as gender, age, language and accent. The next step will be the presentation of the project next June in Brussels.

The SIIP platform will be managed by Interpol at its headquarters in Lyon, France, with the aim of increasing the accuracy of voice data and improving its judicial reliability and admissibility. While the SIIP voice recognition system can handle any "legally intercepted" sound, including ambient conversations, its intended use would be to match the collected voices to a database representing a vocal "blacklist".

A kind of criminal fingerprint file of criminals and terrorists archived by the security services. Samples can come from recordings on mobile, landline or mobile networks, and through Voice over Internet Protocol (VoIP) applications, such as Viber, Skype, or Whatsapp, or from audio data collected from recruiting videos. or propaganda posted on social networks. This recorded data essentially becomes an activated widget on a production line.

This sound file may already include some descriptive metadata added by law enforcement officials who initially secured it. The software will then tag this file by adding new information about the speaker's age or accent, for example.

To facilitate this task, SIIP creates a template for recording a phone call made by the police, indicating the acoustic parameters that characterize the voice signals of the file. These characteristics are called identity vectors, which are used to attempt a search for matches in the "voicemail list".

To realize the platform SIIP, the developers realized programs on the basis of specific algorithms intended to treat queries related to the mathematical processing of the sounds of the natural speech.

Thus, newly recorded voice samples can be sorted to make them more intelligible by using background noise filtering techniques, improving voice clarity, and isolating useful sounds.

The aim of these highly technological treatments is to achieve, under the best conditions, the mathematical correspondence of the acoustic vectors of the recorded sound file with those contained in the database of sound samples of the "blacklist" which can contain up to to one million voice files.

Access to these "voiceprints" is managed by Interpol in collaboration with law enforcement agencies from 192 countries. According to the results of the real tests carried out in the field by the experts of Interpol, the performances of the platform SIIP are exceptional.

It can match voice samples from social media platforms, including Twitter, Google+, LinkedIn, YouTube, and Facebook. By combining multimedia content according to search criteria such as geolocation of language, the system will label and process this material, and find similar files in the low data.