GCHQ Visit to NSA on Voice Analytics Systems
Open PDFDocument date: 2008-01-01
Pages: 1
Genre: report
Banner: SECRET STRAP1
Classification: SECRET STRAP 1
Dissemination: -
This document is a GCHQ report detailing a late-November (2007?) visit to NSA’s R64 division focused on collaboration in language identification (LID) and speaker identification (SID) systems, including sharing models and research insights. It also describes NSA tools like Hotzone and Voice RT for real-time voice analysis, and mentions an upcoming visit from the CLEAR consortium to address challenges in speech intelligibility and audio recovery for intercepted communications.
HOTZONE, B3M, Monte Vista, Voice RT, NUCLEON, ROSECROSS, HLT
People mentioned:
-
Countries:
Afghanistan
Tags:
voice analytics, speaker identification, language identification, LID, speech activity detection, SAD, speech-to-text, phonetic search, Human Language Technology, academia, audio analysis, speech processing, CLEAR, Centre for Law Enforcement Audio Research, intelligibility, intercepted communications, TDY, model exchange, real-time analysis, research collaboration, linguist support
Identifier:
Pages-From-VFUG-minutes-Jan-2008
PDF SHA-256:
d4ed633ad1c744b69bd406b65b8e0b9a63e3553cab40ea250f6435cb378c6f04
My notes
SECRET STRAP1
a) Visit to R64, NSA
[REDACTED] and [REDACTED] from B14 visited our opposite numbers in R64 at NSA during the last week in November. The aim of the visit was threefold:
i. to learn about the NSA process for building language ID (LID) and speaker ID (SID) systems whilst sharing our own experiences;
ii. to establish closer collaboration on research tasks;
iii. to move forward the exchange of datasets/models for use in LID and SID systems.
An opportunity for closer collaboration was identified for LID, and it is proposed that [REDACTED] will spend a 4-week TDY in R64 probably in March.
The sharing of models with NSA is possible, but the legal position there currently prevents them sending us raw datasets, although it is hoped that this may be overcome. We undertook to provide them with our Afghan languages LID corpus, which R64 have now received.
Hotzone is NSA's audio tool, which we evaluated some years ago and assessed to be unsuitable for use at GCHQ, due to its inherent latency and inability to accept plug-in filters. Since then this tool has undergone a complete software re-write, with impressive results. We have asked for a copy of the source code to evaluate this further with the initial intention of integrating it into the B14 Monte Vista operational prototype. Hotzone is coded in Java and will run natively on a Windows desktop, so potentially could be considered as an eventual replacement for Rosecross.
We were shown NSA's Voice RT (Voice in Real Time) system, which provides content-based information to NSA linguists/analysts used in conjunction with Nucleon and Hotzone (NSA's equivalent of B3M and Rosecross). Voice RT provides Speech Activity Detection (SAD), SID, LID, Speech to Text (STT) and Phonetic search. Essentially it is a one-stop shop for what we term Voice Content Related Information (CRI). This is a major element of the HLT Programme and a massive effort has been expended to improve deployability of this system.
b) CLEAR
On 31 Jan, B14 will be hosting a visit from the CLEAR (Centre for Law Enforcement Audio Research) consortium. CLEAR is a multi-agency endeavour sponsored primarily by us, SS, HMGCC, SOCA and the Home Office to become a centre of excellence in the science of speech cleaning and recovery. Four academics and a senior scientist from HMGCC will be visiting and wish to see the problems that we face at GCHQ with intelligibility/audibility of our voice intercept. OPI~MENA and OPI~SC have already been asked to demonstrate some of the problems they encounter, and other IPTs are welcome to contribute too. More about CLEAR here: [REDACTED]
3 of 6
This information is exempt from disclosure under the Freedom of Information Act 2000 and may be subject to exemption under other UK information legislation. Refer disclosure requests to GCHQ on [REDACTED] (non-sec) or email [REDACTED]
SECRET STRAP1
Published in the Same Report
(10 of 21 documents)
-
RT10 Initiative Overview [pdf]
Overview document for RT10, "Real-time, distributed architecture for mobile (tactical) and fixed (national) operational environments"
Programs: RT10 · COBRA FOCUS · GREENDRAGON · GOLDMINER · JUGGERNAUT · MATTERHORN · OBELISK · SORTINGHAT · INFOVIS · SHARPFOCUS2 · TOGA · SORTINGLEAD · GLOBALREACH · VOICERT · QUADRASPECTRE · RENOIR · ANALYSTSNOTEBOOK · OCTAVE · HLT · DRT · GEOT · InnoVisions -
NSA Georgia Opens New Audio-Forensics Lab SIDtoday [pdf]
The document discusses the launch of a new audio-forensics lab at NSAG (National Security Agency Georgia) designed to enhance the processing and analysis of intercepted voice communications. The lab, funded through the Senior Language Technology Authority (SLTA), will focus on resolving language processing challenges for language analysts by improving the intelligibility and quality of foreign-language voice intercepts.
People: Mercer, Marti Hix -
Alert: Voice Masking Is Discovered in SIGINT SIDtoday [pdf]
This March 2007 post from the internal NSA newsletter SIDToday draws attention to voice-masking technology used by subjects of surveillance to frustrate the agency’s voice-recognition efforts.
-
Tips for a Successful Quick Reaction Capability SIDtoday [pdf]
In response to Iranian President Ahmadinejad’s 2007 visit to New York, the NSA’s Iran Government and Leadership branches launched a Quick Reaction Capability (QRC) to gather time-sensitive intelligence. The QRC focused on enhanced collection techniques, collaboration across departments, and efficient data management to analyze the extensive communications traffic during the event, resulting in valuable leads on Iranian leadership activities and diplomatic plans.
People: Ahmadinejad, Mahmoud -
New RT-RG Overview Video Available on NSANet SIDtoday [pdf]
The NSA's Meade Operations Center released a 10-minute video overview of the Real-Time Regional Gateway (RT-RG), a tool suite designed for forward-deployed analysts. The suite enables faster access to real-time metadata and reporting, improving tactical support for military operations, with new functionalities being developed for use in Afghanistan and beyond.
-
Voice Biometrics Capability Proven During Recent Technology Demonstration SIDtoday [pdf]
The Coordinate Operational Resources for Voice Exploitation Technology (CORVET) project, managed by the Expeditionary SIGINT Services Office (ESSO), demonstrated a promising capability for voice biometrics that could be used in security operations to identify individuals based on voice characteristics. Similar to the long-established method of fingerprinting, CORVET aims to create a database of voice models that could be used for identification in various contexts, such as detention and force-protection scenarios.
-
Alvin, Simon, and... Al Qaeda? Finding Modified Voice in SIGINT Traffic SIDtoday [pdf]
This SIDtoday article discusses the use of voice modification technology by Al-Qaeda and other extremist groups to evade identification in SIGINT traffic. It highlights the development of the HLT Lite software by R64, which can automatically detect modified voices in intercepted communications, enabling the identification of targets even when traditional voice recognition fails.
-
Come to SID's Identity Intelligence (I2) Day Conference and See Your Target in a Whole New Light! SIDtoday [pdf]
This SIDtoday article promotes NSA's Identity Intelligence (I2) Day Conference, highlighting the importance of using SIGINT tools and techniques to track individuals across time and space. It emphasizes the growing role of biometrics, biographic data, and digital footprints in identifying and targeting foreign persons of interest, with insights from key speakers and case studies.
-
SIGINTers Use Human Language Technology to Great Advantage, Isolate Conversation about Threat to US Official SIDtoday [pdf]
In 2010, the Special Collection Service in Mexico City began using Human Language Technology (HLT) to streamline voice collection and analysis, significantly improving efficiency. By rapidly filtering through large volumes of audio data, HLT helped isolate critical intelligence, such as identifying a conversation related to a bomb threat targeting a U.S. Embassy official, all within minutes instead of hours.
Programs: HLT -
Innov8 Voice Analytics Experiment Profile [pdf]
This document outlines Phase 1 of a Voice Analytics experiment conducted by GCHQ in partnership with the NSA, testing the ingestion and processing of voice data through the DISTILLERY platform on GHOSTMACHINE hardware. The experiments evaluate independent, parallel, and linked analytics (SAD, GID, LID), as well as dynamic resource management, to ensure accurate, scalable, and compliant voice processing in a Joint Collaboration Environment.
Marking: TOP SECRET
Related Documents
-
Voice Fax User Group Minutes December 2010 [pdf]
This GCHQ document is a NSA/GCHQ meeting record from December 2010 detailing the deployment of voice analytics, including speech activity detection, speaker identification, and language identification, integrated with tools like Google Earth to map locations of target communications. It outlines ongoing and planned speaker ID operations against specific targets, including militant leaders and foreign officials, demonstrating the operational use of voice recognition for intelligence collection and geolocation.
-
Voice Fax User Group Minutes March 2009 [pdf]
The document is a TOP SECRET GCHQ meeting record from March 2009 detailing updates to B14/MCS voice analytics systems, including new language identification (LID) models for Pashto, Dari, Brahui, and Baluchi languages and a first-pass speaker ID (SID) system for General Kiyani of Pakistan. It also highlights operational successes, such as improved reporting on Iraq's Vice President Adil ’Abd-al-Mahdi, which enhanced GCHQ’s reputation with NSA and reduced unnecessary workload for linguists.
-
Finding Nuggets -- Quickly -- in a Heap of Voice Collection, From Mexico to Afghanistan SIDtoday [pdf]
A 2011 article by an NSA Texas analyst describing the deployment of Human Language Technology analytics — including speech-to-text, speaker identification, gender recognition and language identification — to Special Collection Service sites in the Americas and to the Afghanistan Remote Operations Cryptologic Center, with operational examples ranging from tunnel detection in Tijuana to intelligence support in Kandahar.
Programs: HLT -
Alvin, Simon, and... Al Qaeda? Finding Modified Voice in SIGINT Traffic SIDtoday [pdf]
This SIDtoday article discusses the use of voice modification technology by Al-Qaeda and other extremist groups to evade identification in SIGINT traffic. It highlights the development of the HLT Lite software by R64, which can automatically detect modified voices in intercepted communications, enabling the identification of targets even when traditional voice recognition fails.
-
For Media Mining, the Future Is Now! (conclusion) SIDtoday [pdf]
In the future, thanks to Human Language Technology (HLT), an analyst will be able to sort through millions on intercepted phone calls each day "and focus on only the small percentage that is relevant". He or she will also be able to get precise answers to queries such as "Where is the mayor of Baghdad?"
People: Picone, Joseph“How the NSA Converts Spoken Words Into Searchable Text” - The Intercept ↗ (2015-05-05) · The Intercept ↗ (2019-05-29) · Froomkin, Dan