Introduction

Welcome to the DeepID Web Platform, a state-of-the-art Deepfake detection solution developed by Deep Media. Our platform harnesses cutting-edge machine learning algorithms to accurately identify Deepfakes across multiple modalities, including audio, video, and images. This comprehensive user guide will provide you with a detailed walkthrough of the platform's features, functionalities, and best practices to ensure you get the most out of DeepID.

Getting Started

In this section, we'll walk you through the basics—like how to set up your account, get familiar with the dashboard, and explore other important features.

Creating an Account

To access the DeepID Web Platform, you need to create an account:

1. Visit our website's login page and click on the "Sign Up" button.

2. Fill in the required information, including your name, email address, and desired password.

3. Once your account is created, and your usage is approved by Deep Media Administrators, you can log in using your credentials.

Dashboard Overview

Note: If you are interested in learning more about how to navigate the dashboard scroll down to the Analyzing Uploaded Content section.

Upon logging in, you will be directed to the DeepID dashboard. The dashboard provides an overview of your recent activity, including the number of files analyzed, the modalities used, and the detection results. You can also access various sections of the platform from the sidebar on the left side of the screen

Filtering Options

To quickly find the files you need, you can utilize our robust filtering features.

• Start by filtering the content based on its status, which gives you a clear view of the progress of each file in the analysis process.

  • You can see exactly what stage the file is going through, whether it's queued, processing, or completed.

• Additionally, you can search for specific files using titles or keywords, making it easy to locate and analyze individual pieces of content.

Analyzing Images

When analyzing images, the dashboard provides essential information at a glance, including the upload date, overall falsehood probability, modality, and the user who uploaded the file. By clicking on a specific image, you'll open a detailed dashboard with a "See Details" button. Upon clicking this button, you'll access a wealth of additional information.

One of the key features is the detailed heatmap that highlights the areas our detector has identified as potentially manipulated. The red regions indicate possible manipulation, while the green areas suggest authenticity. This heatmap provides a visual representation of where our algorithms have detected anomalies or inconsistencies in the image. You can toggle between the heatmap view and the original image using the buttons at the top of the screen, allowing you to compare and contrast the detected manipulations with the original content.

Analyzing Videos

When you click "See Details" for a video file, you'll be presented with a comprehensive video analysis interface. The full video player allows you to watch and navigate through the content, providing you with complete control over playback. Additionally, you'll find relevant subject information displayed alongside the video, giving you context about the content being analyzed.

One of the most powerful features of DeepID's video analysis is the ability to examine the video on a second-by-second basis. Below the video player, you'll see the live output from our ensemble detectors. These detectors are state-of-the-art algorithms that work together to identify potential Deepfake manipulations throughout the video.

The live output presents a timeline view of the video, with each second represented by a specific data point. As the video progresses, our ensemble detectors analyze each second independently, providing real-time insights into the authenticity of the content. You can hover over any data point to view the detailed results for that particular second, including the falsehood probability and specific manipulation indicators.

This second-by-second analysis is incredibly valuable for pinpointing the exact moments where manipulations may have occurred. It allows you to identify specific frames or segments that require further attention, enabling a highly targeted and efficient investigation process.

In addition to the live output, DeepID generates an overall falsehood probability for the entire video. This probability takes into account the collective results from our ensemble detectors, providing a comprehensive assessment of the video's authenticity.

Analyzing Audio

Audio analysis in DeepID offers a range of powerful tools to examine and understand the audio content. Upon clicking "See Details," you'll be greeted with a Mel Spectrogram, a visual representation of the audio spectrum that highlights important features. The Mel Spectrogram converts the audio signal into a two-dimensional representation, with time on the x-axis and frequency on the y-axis. The intensity of the colors in the spectrogram indicates the strength or energy of the audio at each time-frequency point.

You can also switch to an FFT (Fast Fourier Transform) spectrogram, which provides an alternative perspective on the audio frequencies. The FFT spectrogram represents the audio signal in terms of its frequency components, allowing you to identify specific frequencies and patterns within the audio.

These spectrograms serve as valuable tools for identifying anomalies, inconsistencies, and patterns in the audio that may indicate manipulation. By visually analyzing the spectrograms, you can detect sudden changes, unexpected frequency patterns, or other signs of tampering.

In addition to the spectrograms, DeepID provides a comprehensive audio player that allows you to listen to the entire audio file and navigate through it using intuitive controls. This enables you to examine the audio both visually and auditorily, giving you a multi-sensory approach to analysis.

On the left side of the screen, you'll find a robust transcription algorithm that converts the audio into written text. DeepID's advanced speech recognition technology accurately transcribes the audio content, making it easier to review and analyze the spoken words. The transcription feature supports a wide range of languages, ensuring that you can effectively analyze audio content from diverse sources and regions.

Below the transcription, you'll see the live feed output from our ensemble audio Deepfake detectors. Similar to the video analysis, these detectors operate on a second-by-second basis, providing real-time insights into potential manipulations within the audio. Each second of the audio is analyzed independently, and the detectors' outputs are displayed as a timeline view.

You can hover over any data point in the live feed output to view detailed results for that specific second, including the falsehood probability and manipulation indicators. This granular analysis allows you to identify the exact moments where audio manipulations may have occurred, enabling targeted investigation and verification.

Ultimately, DeepID generates an overall falsehood probability for the entire audio file based on the collective output of the ensemble detectors. This probability provides a high-level assessment of the audio's authenticity, taking into account the results from each second of the analysis.

Exporting Reports

For audio and video content, DeepID offers the option to export comprehensive reports. By clicking the "Export Report" button in the top right corner, you can generate detailed visualizations and insights that go beyond simply identifying manipulation.

These reports include in-depth breakdowns of the second-by-second analysis, highlighting the specific moments where manipulations were detected. The visualizations provide clear and concise representations of the detection results, making it easy to understand and communicate the findings.

Moreover, the exported reports offer valuable insights into the individual components of the audio or video that may have been manipulated. This includes identifying specific speakers, objects, or segments within the content that exhibit signs of tampering. By isolating these elements, the reports enable targeted investigations and provide compelling evidence for further analysis or legal proceedings.

The ability to export these comprehensive reports is a game-changer in the field of Deepfake detection. It allows you to share the findings with stakeholders, collaborate with team members, and present the results in a professional and accessible format.

Conclusion

The DeepID Web Platform is a powerful tool designed to help you combat the challenges posed by Deepfakes. By following the instructions outlined in this user guide, you will be well-equipped to navigate the platform, upload media, select modalities, collaborate with team members, and gain valuable insights into your Deepfake detection efforts.

If you encounter any issues or have further questions, our dedicated support team is here to assist you. Feel free to reach out to us through the "Support" section in the platform or via email at support@deepmedia.ai

We are committed to continuously improving and updating the DeepID Web Platform to stay ahead of the evolving landscape of Deepfakes. Regular updates and enhancements will be made available to ensure you have access to the latest technologies and features.

As you utilize DeepID's content analysis features, remember to leverage all the available tools and visualizations to gain a thorough understanding of the content's authenticity. Don't hesitate to explore the second-by-second analysis, examine the visual representations, and dive deep into the live detector outputs. The more you engage with the platform's capabilities, the more effectively you'll be able to identify and combat Deepfakes.

Thank you for choosing DeepID as your trusted Deepfake detection solution. We look forward to supporting you in your mission to identify and combat Deepfakes effectively.