Several North Korean state universities and institutes have recently developed facial and vocal recognition programs based on artificial intelligence (AI) technology, DPRK media outlet Dawn reported last week.

The website said “dozens of types of artificial intelligence” were displayed at the 28th National Exhibition of IT Achievements, held between November 7 and 13.

According to reports, the mathematics institute of the State Academy of Sciences has developed facial recognition and fingerprint identification systems which can be used for tasks including “management, immigration control, and electronic payment.”

“The fingerprint identification algorithm which is used to design recognition system received an excellent evaluation at the international fingerprint verification competition,” the outlet reported last Friday.

On November 14, state-run outlet DPRK Today reported that a “Facial recognition access control system” had been on display at an exhibition held at Pyongyang’s Sci-Tech Complex.

Fingerprint sensors first appeared in North Korean media in September, when DPRK Today reported that the Arirang Information Technology Company had released a smartphone equipped with fingerprint authentication technology.

The college of information science at Kim Il Sung University has also, according to last week’s report, developed the speech recognition program Ryongnamsan (룡남산) 5.1. – named after a nearby mountain.

There is reportedly “no limitation in the number of the words that can be recognized,” and the program is said to “fully recognize” text related to the social and natural sciences, according to researcher Mun Sung Il.

“The [accurate] rate of speech recognition has improved based on the Subspace Gaussian Mixture Model,” Dawn reported. “The accuracy of speech recognition is 98 percent and the recognition speed is 6 characters per second.”

Reports suggest that Kim Il Sung University has expended significant resources in studying voice recognition technology.

Ri Jong Chol of the college of information and science said the speech recognition rate of the Ryongnamsan software “has been largely improved” based on the Subspace Gaussian Mixture Model – a common acoustic modeling approach – in a post published in late September.

The researcher said the technology was “successfully applied to small-scale tasks in the professional fields, including the recording of meetings.”

In a journal published this year, Kim Il Sung University’s Ri Hyong Sun and Mun Song Il also proposed means to enhance vocal recognition technologies.

“We improved the training speed of language model significantly by using a recurrent neural network with divided output layer and enabled to use it to training of large scale of language model for Korean continuous speech recognition,” an English-language abstract for their paper read.

The Pyongyang Information Technology Bureau has also developed a program for recognizing spoken Korean named “Chunji (천지),” the DPRK Today reported in September.

“This program has continuous speech recognition function and real-time recognition is guaranteed.”

The program can reportedly adapt the voice of the speaker into text, as well as learn texts to “increase recognition rate.”

In addition to its work on voice recognition technology, optical character recognition (OCR) software has also been developed by Kim Chaek University of Technology.

The program Sindong (신동), meaning child prodigy, can read documents written in Korean and foreign languages including English and Chinese, but its voice recognition engine is limited to the Korean language.

The software can also fix distorted images and can automatically identify and extract letters and pictures from scanned documents with a reported recognition rate of 99.7 percent.

An advertisement for the Sindong program suggests that the scanned documents can be converted into a Microsoft Word, Powerpoint, or Excel file, or into a PDF, and claims it can read 400 characters per second.

Kim Il Sung University has also developed test recognition software which can extract information written in Korean, according to the poster.

The program can also recognize low-resolution documents photographed by a digital camera, but the recognition rate reportedly differs depending on the quality of documents.

Edited by Oliver Hotham 

Featured Image: Dawn