top of page

Deepfake Detection

Prepare for Advancing Deepfake Technology

The "Deepfake Detection Solution" is a solution that detects unnatural aspects that humans cannot distinguish from fictitious images, videos, and audio data generated by deepfake technology, and determines whether they are real or not.Deepfake technology is evolving rapidly, and we are now moving into an era where even individuals can easily create fake media. While the social impact and concerns are growing, it is said that measures need to be taken urgently, especially in the financial and media industries, where personal authentication and the credibility of information are important. Here, we will introduce an overview of deepfake detection technology and the solutions that NABLAS possesses.



What is a deepfake?

This is a general term for technology that uses deep learning to create fake videos and audio with a high degree of accuracy. Not only can it create images and voices of non-existent people, but it can also generate images and voices that look exactly like a specific individual, and the technology has advanced to a level where humans can no longer distinguish them. There are concerns that it could be used for crimes such as impersonating others or fabricating evidence, and the number of cases that develop into criminal cases is increasing worldwide.

How deepfakes are generated


The process of generating deepfakes



Main Usage Scenarios


Check for Fakes on Social Media and the Internet



News Materials


Solution Features


Detect Fakes in Various Formats, Including Images, Videos, and Audio


Detect Fakes Created Using Unknown Generation Methods.


Detect Fakes Even with

Audio Containing Noise

NABLAS Deepfake Identification Solution

NABLAS plans to provide an authentication solution that can be applied to media suspected to have been generated using deepfake technology. There are two methods of provision: "simple authentication" that allows for quick authentication using API, and "detailed authentication" that is highly reliable and conducted by authentication experts. Please use the method according to your purpose. Please contact us for details.


The details of the detection technology are explained below.

Deepfake Detection Technology for Images and Videos

Techniques for detecting artifacts


A classic method is to determine whether an image is a fake by focusing on specific parts of the image and detecting unnatural points.

For example, in the image on the left, the pupil shape is a perfect circle, whereas in the generated image on the right, the shape is distorted. By paying attention to the shape of the pupil, it is possible to determine whether the image is fake or not. In addition, the following features are known to be effective in detecting fake images:

Loss or partial loss of eye color or light reflexes

◼︎ Uneven color and grayscale in the video itself

◼︎ Pupils, blinking patterns, and frequency

◼︎ Facial expressions and head movements

◼︎ Characterize the correlation and actions of each part

Deep Learning Detection

ディープラーニングによる検知 プロセス

It is also known that techniques such as deep learning that automatically extract features that may become artifacts are useful. By learning
real , and by detecting whether the image data contains those features, it is possible to determine whether it is real or fake.
Videos can also be captured as a collection of images, so by breaking them down into images for different scenes, an image classification approach can be implemented for each image.

Demo of Deep Fake Detection for Images and Videos

Green indicates real prediction, red indicates fake prediction

Deepfake Audio Detection Technology

Techniques for detecting artifacts


Extract voice features and use machine learning technology to detect them

Classify audio features using a machine learning model

◼︎Classifying  raw waveform characteristics using DL technology

◼︎ Approach to converting audio into images and classifying them

Convert to spectrogram for classification and detection


Convert audio into mel spectrograms etc. and classify using pre-trained models

◼︎ Calculate speech features and take a statistical approach

Demo of Deep Fake Audio Detection

Real Voice

A recording of the spoken voice

00:00 / 00:03

Convert to Spectrogram


Fake Voice

Mimicking the Characteristics of Real Voice

Voice Generated by Generative Deep Learning Technology

00:00 / 00:04

Convert to Spectrogram


White Paper

Deepfake technology is an imminent threat to society, but the underlying technology, generative deep learning, is of great industrial value and is a promising technology that is expected to bring about change in many industries. NABLAS has published a white paper that not only discusses the threats and negative aspects of deepfakes, but also highlights the underlying technology, generative deep learning, and fake detection technology.

No posts published in this language yet
Once posts are published, you’ll see them here.
bottom of page