It’s a rainy April afternoon in Taguig City when police spring into action. Nearly one million people call this bustling financial center of the Philippines home. It’s loud and crowded and easy for a child to become lost and maybe even forgotten—but not today. Local law enforcement has been methodically building their case for half a year, scouring the internet, and analyzing every scrap of data available. And it all pays off. The highly coordinated operation is a success … but how did we get here?
Technology: Learning to Wield a Double-Edged Sword
Our interconnected, digital world has brought immense benefits, advancing every sector of society; however, some with malevolent intent have taken advantage of this connectivity to gain access to those most vulnerable. Unfortunately, traffickers, abusers, and pedophiles use technology to exploit children. Criminals use social media, gaming platforms, chat applications, and video conferencing software to sell recorded videos, images, and livestream feeds of rape and abuse at an alarming rate. Online child exploitation demands decisive and unflinching action. Fortunately, there are ways to fight back.
Transnational organized crime (TOC), which includes human trafficking, poses a direct threat to public health, public safety and national security. Countering TOC efforts are carried out in three main ways: Interdiction, Intelligence, and Investigation. Numerous investigators and analysts are working tirelessly to detect, disrupt, and identify offenders, but the sheer volume of data and digital media to comb through can be overwhelming. The Department of Homeland Security (DHS) is fiercely dedicated to countering human trafficking across the country and around the world, no matter how difficult or time consuming the mission may be. That’s why the Science and Technology Directorate (S&T) has teamed up with Homeland Security Investigations (HSI) and others to develop the enhanced analytic platforms, tools, and capabilities required to do this important work.
“The digital forensics work S&T is doing with HSI has directly led to children rescued and traffickers arrested,” explained S&T Digital Forensics Subject Matter Expert Patricia Wolfhope. “I am overjoyed and deeply thankful to see such a significant impact being made.”
StreamView: Putting the Puzzle Pieces Together with Livestream Software
The investigation of individuals who sell livestreamed webcam shows of child sexual abuse and exploitation is a painstaking process. This is because disrupting and dismantling TOC networks is so complex. Agents must identify connections between various perpetrators, including the abusers who commit physical acts, groomers who manipulate victims online, multimedia distributors, and paying consumers of the resulting material.
In 2019, a strategic partnership between HSI and S&T was formed to design and develop StreamView, a prototype livestream investigation tool that would help reduce victimization and availability while increasing apprehensions. After just 90 days, the software was created and piloted operationally at the HSI Portland, Maine, Field Office. The innovative program allows agents to streamline their investigative process, aggregating and organizing the huge amount of unstructured data from legal warrant returns to generate and prioritize investigative leads, identify relationships, and build a cohesive case. The new capability not only significantly enhances agents’ effectiveness and efficiency; it’s also saving lives. StreamView is helping combat and dismantle child exploitation TOC networks worldwide. To date, this technology has contributed to over 40 rescued victims, more than 30 arrests and over 60 new case leads.
In fact, the highly coordinated operation described earlier was one of two successful operations that recently took place on April 5, 2022, in Taguig City, Philippines, after six months of international law enforcement collaboration between HSI and the Philippine National Police’s Women and Children Protection Center. Thanks in part to StreamView technology, three suspected traffickers were arrested and 23 victims—22 of whom are minors—were rescued.
SpeechView: Finding the Signal in the Noise Through Language Translation
HSI recognized the need for an effective tool to identify language spoken in videos for investigative lead purposes and requested assistance filling this capability gap. A key aspect of the monumental effort of investigating worldwide TOC networks is language detection, transcription, and translation. Agents are dependent on linguists for assistance, but this expertise is not always immediately available. Multiple federal, state and local agencies and DHS’s Federal Law Enforcement Training Centers identified additional applications for such a technology, including transcription and translation of 911 calls, witness and suspect interviews, jailhouse calls, seized media, and bodycam videos.
To meet this vast and diverse need, S&T developed SpeechView, which uses advanced machine learning and natural language processing (NLP) to build language-specific models, process the information, and provide an output for human review and analysis. This language processing and text analytics tool provides numerous vital capabilities:
- Speech Activity Detection – automatically detect speech within audio/video files
- Language Identification – help narrow down geographic origin and/or assist in finding a translator
- Diarization – partition individual voices into distinct audio segments
- Gender Estimation – determine gender from spoken language
- Speech-to-Text Transcription – convert speech into text while maintaining the exact wording used
- Translation – convert text in one language to another
- Speaker Recognition – identify if an individual is known or unknown to a law enforcement database
SpeechView is in operational prototype assessment by multiple federal, state and local entities. The Speech Activity Detection capability currently recognizes over 65 languages. The Speech-to-Text capability supports seven languages and Translation supports six.
Audio and video files can be many hours long, so capabilities like Speech Activity Detection are a gamechanger in terms of efficiency. Even a 5-minute call can take agents hours to manually transcribe, but SpeechView is able to generate a draft for analysts’ review in minutes. The ability to quickly triage and analyze audio content greatly increases the productivity of investigators and can be applied to a range of cases and investigations to combat TOC, including online child sexual exploitation and abuse.
Horus: Finding a Needle in a Haystack with Media Analytics
Media collected during an investigation frequently contains faces—of suspects, victims, associates, and others. Key questions typically arise related to who the individuals are and whether or not they are known to law enforcement. These questions are not quickly or easily answered through manual review, especially when the volume of media is prohibitively large.
Horus is a media analytics platform incorporating facial recognition (FR) algorithms that can detect faces within a set of still and video imagery to perform triage, cluster (group) faces belonging to the same subject, or search for a specific face from within a gallery of faces. It uses artificial intelligence and machine learning to enhance investigative processes through automation and analytics, greatly reducing agent/analyst time. For instance, a set of seized imagery can be quickly processed to find and group images containing faces, or a face can be extracted from a security video and matched against a gallery of mugshots. Most recently, Horus has been used to help investigate document fraud and identify war criminals.
This web-based application incorporates unconstrained FR algorithms originally developed through the Office of the Director of National Intelligence's (ODNI) Intelligence Advanced Research Projects Activity (IARPA) Janus program. In collaboration with ODNI, the Department of Defense (DoD), and the Federal Bureau of Investigation (FBI), as well as end users, S&T developed and expanded the Horus application to provide the user interfaces, workflow, and features required to operationalize the capability for investigator and analyst use. Since 2019, Horus has been transitioned to six federal users, with their feedback continuing to be used to refine the tool.
“Horus is a shining example of what can be accomplished when federal agencies work together,” said Wolfhope. “Together, we were able to tackle a difficult issue and create a media forensics tool that can benefit all of our collective stakeholders, without duplicating our efforts.”
Mission: Persisting Until Every Victim Is Saved and Every Perpetrator Brought to Justice
Analysts use any information that can possibly be gleaned from illicit multimedia to uncover leads and bring unconscionable criminals to justice, and any tool or capability that can increase efficiency is a welcome addition to the arsenal. These are just three examples of ways S&T is helping colleagues on the frontlines of this battle to fight back against online child sexual exploitation and abuse.
Anyone who has information about online sexual exploitation of children happening is encouraged to report the crime immediately. Please use the electronic Cyber Tip Line or call 1-800-THE-LOST (1-800-843-5678). The Cyber Tip Line is operated by the National Center for Missing and Exploited Children in partnership with the FBI and other law enforcement agencies.