Although we already know that NSA relies on fingerprints and facial images to identify targets, according to an institutional document in 2008, voiceprint is "the dominance of the NSA."
In the most severe period of the Cold War in the United States and the Soviet Union, it was the winter of 1980. The FBI agent recorded a call—a man was arranged to have a secret meeting with the Soviet ambassador in Washington, DC. However, on the day of the appointment, the agents of the Bureau of Investigation did not see who entered the embassy. At the time, the agents had no way to find out his name based solely on his voice, so the spy was able to continue hiding his identity and sold the details of some US secrets to the Soviet Union over the next five years.
It was not until 1985, based on information provided by a Russian defector, that the FBI finally determined that the caller was former National Security Agency (NSA) analyst Ronald Pelton. The following year, Ronald Pelton was sentenced to espionage.
Today, FBI and NSA agents can identify his true identity within a few seconds of Pelton's first call to the Soviet Union. The NSA's January 2006 confidential memo showed that NSA analysts used a "sound-to-speech technology" to successfully match Pelton's previous audio files to phone monitoring. The memo describes: "If this technology has been in existence for 20 years, it will be able to detect Pelton's espionage and arrest it in advance, which will greatly reduce Pelton's significant damage to national security."
The confidential documents provided by former NSA employee Snowden show that NSA develops technology that not only turns private conversations into text, but also automatically identifies people in the conversation.
Americans often use this technique called voiceprint recognition when they wake up Amazon's Alexa or call the bank. But long before the language instructions such as "Hello Siri" and "OK Google" entered the homes of ordinary people, the NSA used this voiceprint recognition technology to monitor terrorists, politicians, drug lords, spies and even some ordinary people. Employees of government agencies.
The principle of this technique is to analyze the unique physical and behavioral characteristics of the individual's voice to distinguish the voices of different people, such as the pitch of the vocalization, the shape of the mouth, the length of the pharynx, and so on. The algorithm then creates a dynamic computer model of the individual's sound characteristics, also known as the "soundprint" model. The whole process—capturing the words, translating them into voiceprints, and comparing that representation to other “soundprints†in the database—can be done in an instant. Although we already know that NSA relies on fingerprints and facial images to identify targets, according to an institutional document in 2008, voiceprint is "the dominance of the NSA."
We can easily see the reason. The NSA, whether licensed or not, intercepted calls from millions of US citizens, including transoceanic, video, and Internet calls, creating an unparalleled voiceprint library. The files from Snowden show that after the analysts have provided some people's recordings to the voiceprint recognition algorithm, even if these people use unknown numbers, secret codes or different languages ​​in other calls, the algorithm can The existing audio matches.
As early as Iraq’s freedom of movement, analysts used voiceprint recognition technology to confirm that the “records of Saddam’s leader who was suspected of being deposed†were indeed Saddam’s own, not as forged as the public thought. The NSA's memorandum further indicates that NSA analysts have built voiceprints for this bin Laden. “In a few broadcasts, his voices are very clear and consistentâ€; with Al Qaeda's current leader Ayman al-Zawahri and Al Qaeda The three-handed Abu Musab al-Zarqawi is significantly different. They also used Zarqawi's voiceprint to find him from several online audio releases.
According to confidential documents from 2004 to 2012, the NSA has made more complex iterations of its voiceprint recognition technology. The document confirms that voiceprint recognition is used in both anti-terrorist operations and anti-drug operations. The document also recommends that more state agencies deploy this technology, not only to trace espionage like Pelton, but also to prevent whistle-blowers like Snowden.
Algorithm that is always listening
(On March 4, 2015, a man leaned on a smartphone at a public phone booth in New York)
Some civil rights experts worry that voiceprint recognition technology and the extended application of the technology will infringe citizens' privacy. "The voiceprint recognition technology has created a new intelligence capability, an ability that is easily abused," explains TImothy Edgar, a former director of the White House National Intelligence Agency. "Our voice replaces us through various channels to complete communication. In the age of mass surveillance, this ability has a profound impact on the privacy of all of us."
Edgar and other experts point out that the relative stability of vocals makes it difficult to change or disguise compared to names, addresses, passwords, phone numbers, and personal identification numbers. According to Jamie Williams, an attorney at the Electronic Frontier Foundation (Electronic FronTIer FoundaTIon), this makes tracking "easy." "As long as you can identify someone's voice," she said, "you can find them in the monitor record or recording."
Sound is a unique and easily accessible biometric: unlike DNA, it can be passively collected and is not limited by distance, without the need to know or obtain their consent. Although the accuracy of the recognition is similar to the radio conditions, in a controlled environment—low ​​noise floor, familiar acoustic environment, and good communication quality—this technique can be accurately matched in a few sentences. personal. The more different sound samples of the same person owned by the computer model, the more powerful the model becomes, and the more mature it is.
In the business environment, voiceprint recognition technology is closely related to call center fraud review, voice assistant conversations such as Siri, and personal banking password verification. And the use of this technology is growing, according to market research firm TracTIca, by 2024, the revenue of the voice biometric technology industry is expected to reach $5 billion a year, and its use will extend to border checkpoints, medical care, and credit card payments. And in wearable devices.
One of the main concerns of civil libertarians is that voiceprint recognition technology has the potential to "cool down" conversations. Trevor Timm, executive director of the Press Freedom Foundation, pointed out that NSA's voiceprint recognition technology could be used to track journalists, expose sources, and block anonymous messages. Although people now know that they should encrypt their phones when dealing with sensitive materials, Timm points out that there are so many channels for secretly recording sounds from TVs to headsets to Internet devices. Timm said that we carry a microphone with us 24 hours a day, that is our mobile phone. We know that the government has a way to break into mobile phones and computers to turn on these microphones.
He continued, "Despite the great changes in legislation after Snowden’s leak, the American people still have only one-sided understanding of the tools that the government uses to monitor millions of people around the world. This is a worthwhile The debate in the public sphere.†But he pointed out that if the public lacks a meaningful understanding of the use of technology, this debate will be very difficult – even if it exists, it is debatable.
A former defense intelligence official could not discuss confidential documents due to policy restrictions. An Intercept said anonymously that he believes that this technology has always been hidden. “The government avoids discussing this technology because it raises some serious questions that the government is not willing to answer,†the official said. “This is an important shift in our personal and our rights since the events of 9/11.†And to get into the scope of technical monitoring, officials pointed out, “You don’t have to do anything, just talk and talk.â€
Civil rightsists worry that if there is no open discussion and supervision of the government's secret collection of our voice patterns, we may enter an increasingly silent world.
New sound tool
(October 9, 2012, Boulder, Colorado, National Institute of Standards and Technology, Institute of Speech (NIST) Building)
Beginning in 2013, Americans already knew that NSAs were collecting large amounts of phone data from home and abroad, but the technology of how to turn raw data into useful information is still little known. In 2015, according to The Intercept, the NSA established a series of “human language technologies†to handle the large amount of audio collected by the government. Automatically translate speech into text through development programs - what analysts call "voice-based Google" - government departments can use keywords and "selectors" to search, read, and index recordings instead of sending people to listen Save a lot of manpower.
Voiceprint recognition technology derived from voice-to-text items provides analysts with an additional tool that can intercept and classify countless theater audio. The NSA and the Department of Defense have invested heavily in developing this technology and increasing its reliability. Prior to the digital age, voiceprint recognition was part of forensic science. During World War II, human analysts compared the visual output of the sound frequencies from the radio. According to Harry Hollien, author of The Voiceprint Recognition of the Court, these machines that can “read the voiceâ€â€”that is, the spectrogram technique—are even used to refute the rumors that Adolf Hitler was assassinated and replaced.
As the federal leader in the standardization of voiceprint recognition in the court, James Voiceman, chief voiceprint recognition expert, explained: "The voiceprint can be seen," he pointed out, although the word "soundprint" has been used by commercial companies, but In fact, it is somewhat misleading. Because "printing" means that the information captured is physical, not behavioral. He said: "In fact, what you have is an equation in the software program that can output different numbers."
These equations have evolved from simple averaging to dynamic algorithm models. Since its reliance in 1996, the NSA has funded the National Institute of Standards and Technology's Institute of Speech (NIST) to develop and test "the most dominant and promising algorithms for solving voiceprint recognition problems." Also working with NIST is the leading biometrics companies and researchers, some of whom have received funding from the NSA and the Defense Advanced Research Projects Agency (DARPA).
The NSA has been ignoring its voiceprint recognition project, so it is difficult for the public to know how much it can currently play. But given the close links between NSA-funded academic institutions and private companies, what is being done in other countries and what the vendors are selling can roughly estimate the progress of the NSA.
For example, industry leader Nuance sells “a nationwide bio-soundprint recognition system to the government, military, and intelligence agencies that can quickly and accurately identify and distinguish one's voice from a system containing millions of voiceprints.†In 2014, the Associated Press reported that Turkey's largest mobile phone company has used Nuance's technology to collect voice data for about 10 million users.
In October, Human Rights Watch (HRW) reported that the Chinese government is establishing a national-level voiceprint database to automatically identify people on the phone. The government aims to link the voice biometrics of thousands of people with their ID numbers, ethnic and family addresses. According to HRW, companies that manufacture Chinese voice software have even applied for a patent for a system that can accurately locate audio files for "monitoring by public opinion."
In November, according to news released by Interpol, an international voiceprint recognition project funded by the European Union passed the final test. More than 100 intelligence analysts, researchers and law enforcement officers from more than 50 countries, including Interpol agents, the British Metropolitan Police and the police in the Portuguese justice system, came to the show floor, and the researchers proved Their program can identify "unknown speakers who speak different languages ​​in social media or legally intercepted audio."
The NSA file reviewed by The Intercept depicts the outline of a similarly evolving system – in the years following the 9/11 event, the development of this system allowed “voice analysts to be able to count hundreds of hours in seconds. The voice clips are filtered to filter useful information based on keyword or speaker voiceprint recognition."
"dramatic" results
(On November 22, 2008, a Sahwa member called on a cell phone near the checkpoint in downtown Baghdad)
Part of the history of NSA voiceprint recognition technology can be reconstructed by reading the Internal Intelligence of the Signal Intelligence Agency (SID) for nearly a decade. Excluding those self-proclaimed and deliberately skipped parts, SIDtoday's memo details the evolution of voiceprint recognition—from a less reliable forensic science that requires human examiners to an automated algorithm that relies on a large amount of voice data. program. In particular, these memos highlight how US analysts work closely with their counterparts in the UK Government Communications Headquarters (GCHQ) to handle a large number of recordings related to counter-terrorism efforts from Iraq and Afghanistan. Although GCHQ refused to answer the specific details involved in this article, it boasted in its internal communication that its system "has played an important role in the cooperation with the NSA."
While it is sometimes difficult to distinguish between SIDtoday and the future of information dissemination and technology, it is clear that at least since 2003, NSAs have been using automatic voiceprint recognition technology to locate and label "the voices of those focused objects." news". A SIDtoday memo explained that whenever a sound signal is intercepted, voiceprint recognition technology models it and compares it to other known sources to answer this question: "This is the horror we have been tracking. Molecules? Is Osama bin Laden?"
But the function of the NSA system goes far beyond answering "yes" or "no". In a series of newsletters in 2006, a project called "Voice RT" was reported that not only automatically recognizes speaker identity in voice interception, but also identifies their language, gender and accent. Analysts can categorize interceptions by category, perform real-time searches with keywords, and set up automatic alerts to notify them when incoming intercepts meet certain label conditions. An NSA PPT further confirmed that the Voice RT program converted its “ingested†Iraqi voice data into voiceprints.
The NSA memo provided by Snowden does not describe the scope of deployment of Voice RT, as mentioned in the minutes of the GCHQ Voice/Fax User Group. The UK agency's memo details how the NSA's voiceprint recognition program is deployed for foreign targets. In the fall of 2007, when their voice/fax user group met with an NSA representative, NSA members introduced an efficient Voice RT system that provides linguists and analysts with speaker recognition and language category recognition capabilities. Convert speech to text and search for speech. “In essence,†the meeting record described Voice RT as “a one-stop shop... they have put a lot of effort into improving the system's deployability.†By 2010, NSA's Voice RT program can handle more than 25 Foreign language audio. In Afghanistan, the NSA uses speech analysis and mapping software to locate clusters of Arabic-speaking towers – used to discover new Al Qaeda training camps.
GCHQ uses a project called Broad Oak to identify the target based on the sound. The British government has established a voiceprint recognition system in the Middle East, identifying leaders from Saudi Arabia, Pakistan, Georgia and Iraq. GCHQ's minutes said: "If you think we can help you find the target you are interested in in the massive flow of information, please feel free to contact us, we are happy to discuss your needs with you and hope to provide a quick and accurate solution."
This is not a blank check. In 2009, when identifying Adil Abdul Mahdi, one of Iraq's vice presidents, GCHQ boasted that they did better than their rival NSA. “Since we have been reporting on him (Vice President) faster than they are, the NSA has given up on participating... This good performance has also improved our reputation in the NSA.†2010, GCHQ Research Overview It shows that the two organizations have cooperated extensively on the joint experiment of the voiceprint analysis project.
But the development of voiceprint recognition tools is not always smooth. In its early stages, this technology was far less powerful and effective than it is today. Former defense intelligence officials recalled that while analysts were able to play voice samples on their workstations, searching for important samples was a big challenge because the audio was not indexed. In a 2006 letter to the editor by SIDtoday, an analyst complained that the introduction of the voice tool made it very crashy and compared its initial speed to "the syrup that drools in January."
However, by 2007, voiceprint recognition technology has matured significantly. The NSA's practical utility for the Iranian President Mahmoud Ahmadinejad's trip to the United Nations General Assembly in New York City details the practical utility of the technology. After obtaining legal authorization, the analyst configured a dedicated system to track as many 143 Iranian representatives as possible. On all of these incoming traffic, they run a voice activity detection algorithm to avoid the analysts wasting time tracking when they are not talking; searching for "email address delivery and discussion of important people" by keyword; According to the voiceprint recognition of the speaker, the dialogue of "important people including the Iranian Foreign Minister" was successfully positioned.
In the same year, a new audio forensics laboratory opened by the NSA in Georgia in the same year pointed out in a statement that the NSA plans to provide these voice technologies to more analysts throughout the organization. A 2008 SIDtoday memo report stated that the upgrade of the system would allow analysts to “find new audio clips that match the target's past audio.â€
These tools evolve as targets find strategies to evade voiceprint recognition. In 2007, analysts noted that the frequency of interception of the two targets they identified as Al Qaeda was beyond the scope of normal people's conversation. In the next few years, analysts in Yemen, Afghanistan, Iraq and elsewhere also discovered the modulated target of this sound. "This is most likely to avoid being identified by intelligence agencies." Some of the audio clips they observed distorted the speaker's tone, so it sounded like "Evan and Chipmunk" talking. Analysts have speculated that AQAP members who participated in the December 2009 bombing of Detroit escaped the government's identification by using brand new phone numbers and masking their voices.
In 2010, the organization's technical staff developed a solution to these modulated sounds—using HLT Lite—a software for searching for modified or unusual sounds. According to SIDtoday, the program found at least 80 modified voice examples in Yemen after scanning more than 1 million audio. According to reports, this led the agency to find a target person who used several new phone numbers at the same time.
As the system's capabilities improve, their monitoring range is also expanding. A September 2010 newsletter details the “dramatic†development of the upgraded voiceprint recognition system in Mexico City – the head of the website said it is comparable to an additional scanner. By searching for the word "bomba" in audio interception, analysts can separate and detect conversations about bomb threats.
The voiceprint recognition system can also be easily reconfigured for use elsewhere. GCHQ's October 2008 meeting minutes described the process of establishing a “high-level network of people involved in the Afghan drug tradeâ€, which was later “invested in unexpected applicationsâ€. Analysts even “carried a carpet search in areas with large drug flows†to identify more targets.
From the battlefield to the organization
(network equipment in the server room)
NSA quickly realized that their recording processing capabilities could be used to identify employees within the NSA. As explained in the January 2006 memo discussing Ronald Pelton's audio, "sound matching technology is being applied to the newly proposed Inside Threat program in an attempt to catch "spy in us."
After leaking secrets by US military informant Chelsea Manning, the plan was publicly announced by the Obama administration, which closely monitored the lives of government employees. But the document seems to indicate that the initiative was implemented before Obama issued an executive order in 2011.
According to Trevor Timm of the Press Freedom Foundation, the NSA's use of biotechnology to detect external threats to detect dissidents inside members is not new. “In the past 15 years, we have seen a series of examples in which law enforcement agencies use invasive tools that are acting on terrorists – whether it is location tracking or face recognition, or like this voiceprint. Identification technology - used in a variety of other crime investigations."
Timm pointed out that in the past few years, informants, intelligence personnel and journalists have taken stricter security measures to avoid revealing their identity. However, "if the phone number used by the reporter is not associated with his identity, then the government will scan their calls by means of a license, etc. This technology will also potentially be used to curb the development of the journalism industry."
For Timothy Edgar, the first civil rights lawyer in the intelligence circle, these “risks†can be attributed to the question: “Are they looking for legitimate goals or abuse of power? For example, trying to monitor reporters or informers?â€
Edgar says that in some ways, voiceprint recognition helps protect personal privacy. This technology allows analysts to screen calls so they can focus on the voice of the target person and eliminate the voices of others. A 2010 SIDtoday memo emphasizes that by ensuring that “the speaker is a country leader rather than a waiter at a donut shop,†the technology can reduce the amount of calls that intelligence personnel need to monitor.
“In fact,†this level of precision became the “shield of the NSA's interpretation of the massive collection of metadata,†Edgar explained. “One of the means by which they defend the project is to claim that they don’t collect all the data, but to collect the information through filters.â€
At the same time, identifying the targets of specific individuals from large amounts of data often proves the need to continue collecting more data. Identification can help analysts narrow down the range of calls, but this technique seems to encourage them to search on a larger scale, because essentially the goal of this task is to listen to the dialogue in which the target sound appears, no matter what number he dials. . Or as pointed out in the Pelton Memo, this technology allows analysts to "discriminate that sound anywhere."
While these documents indicate that the agency does attempt to apply this technology to its employees, the documents reviewed by The Intercept do not clearly indicate whether the institution has created voiceprints in the dialogue data of ordinary citizens in the United States.
The Foreign Intelligence Surveillance Act (FISA) provides that organizations can freely collect audio data transmitted over foreign servers and infrastructure, as well as voice communications between Americans and foreigners. Because of the provisions, Edgar said that the transoceanic phone was "basically" recorded by the system. "I would be surprised if they didn't use these techniques for overseas calls. In a way, that's what they do."
However, the experts expressed a disagreement about whether the NSA has the right to scan the US citizens' calls and make voice prints without permission. This divergence stems in part from the incompleteness of the relevant laws of supervision, and the law in this area fails to keep up with the development of digital technologies such as voiceprint and speech recognition.
Although the United States has enacted strict laws prohibiting the recording of telephone content on US territory without a license, no federal agency is responsible for supervising the collection and processing of sound data.
Part of the reason for the lack of supervision is that although the government needs permission to obtain “content,†should voiceprint be classified as “contentâ€? Or, as the NSA claims, voiceprints are nothing more than “metadataâ€â€”there is less legal protection for this part of the information. On this issue, the law basically did not make any specific provisions on this part of the information, which led some experts to speculate that the NSA is using this legal gray area to implement voiceprint establishment.
In response to a series of detailed questions, the NSA responded as follows: "Based on long-term policies, the NSA will not confirm or deny the accuracy of the US government information referred to in this article."
Do not use it
On Thursday, the Senate voted to expand the 702 of the Foreign Intelligence Surveillance Act (FISA), which gives the NSA the right to conduct secret surveillance of Americans who communicate with foreigners and does not require a permit. This reauthorization, similar to last week's action in the House of Representatives, confirms the views of those critics who believe that the NSA's interpretation of its legal power is increasingly tougher – and increasingly blurred.
Andrew Clement, a computer scientist and monitoring researcher, has been investigating NSA's unlicensed eavesdropping activities before Snowden disclosed the information. He is convinced that the agency has not received any restrictions on the use of voiceprint recognition by US citizens. He explained that the agency often chooses to classify all the information it collects until it reaches the human analyst's ability to listen or read it as metadata. "That's just a huge loophole," he explained. "It seems that they will simply classify any information obtained from the content algorithmically into metadata."
Clement is analogous to how NSA treats phone numbers and email addresses, and how the NSA will legalize the act of establishing voiceprints. The XKeyscore project, which Snowden disclosed in 2013, allows staff to extract email addresses (which are classified as metadata) from intercepted emails. These workers can also perform full-text searches on keywords, which are also classified as contextual information rather than content.
However, Edgar believes that if the government counts our voices as metadata, then he will be shocked. “You can try to argue that the sound characteristics are not equal to the content of the speech,†Edgar said. “But for voiceprint recognition, you still have to collect the contents of a domestic call and analyze it to extract the sound.â€
It is not known how many domestic calls are collected, sampled, or retained by the NSA. But EFF's Jamie Williams pointed out that NSAs don't have to make Americans' voiceprints by collecting American phone recordings, because private companies have been recording our voices. They are having more and more audio resources. Cars, thermostats, refrigerators, light bulbs, and even trash cans have become "smart" (ie, Internet-based) voice devices. Gartner predicts that one-third of human-computer interactions this year will be through conversations with the voice system. Recently, Google and Amazon's "smart speakers" have introduced a voiceprint recognition system to distinguish the voices of family members. “Once the company has these sound data,†says Williams, “in theory, law enforcement agencies can get it, as long as they have an effective legal process.â€
The former government official pointed out that the original voice data can be stored in private enterprises and accessed by the NSA through a secret agreement. For example, in the Fairview program, the agency is partnering with AT&T. Although the US Congress has tried to control the NSA's collection of domestic telephone records, the agency has been seeking to obtain the raw data we provide to the corporate database. (For example, the partnership with Verizon and AT&T, the penetration of the Xbox game system, and the implicit collection of online metadata for thousands of netizens are just a few recent examples.) "Telecom companies hold data, and nothing They can stop them from running an algorithm," the former official said.
What Clement wants to know is that for NSAs, voiceprint recognition is more important than its content recognition capabilities. “This allows them to connect you with your other identity and determine your relationship with others,†he said.
This seems to be the ultimate goal of the NSA. At a conference in 2010 on “An unprecedented opportunity to understand how NSAs use their creative energy for personal tracking,†NSA executives talked about how to incorporate “whole life†strategies into their goals. . They describe the need to integrate biometric data (such as voiceprints) with biographical data (such as social networks and personal history). In the agency's own words, “Everything is about the continuity of personal positioning, tracking and maintenance in space and time. We are not limited to traditional communication methods – we do everything.â€
Molded Waterproofing Cable Assemblies
We specialize in waterproofing products overmolding. We can custom build, custom mold, and over-mold your cable designs.
We specialize in molded cable manufacturing for the widest diversity of
cable and connector types, across the whole spectrum of industries. Rich expeirence in developing and proposing solution Special for IP67, IP68 waterproofing products.
Molded waterproofing cable assemblies, waterproof wire harness, waterproofing cables overmolding
ETOP WIREHARNESS LIMITED , https://www.oemwireharness.com