Voice Biometrics

Embed Size (px)

DESCRIPTION

This report on voice biometrics looks at both speech recognition and speaker/voice recognition and provides a basic analysis of respective market sizes with a look at the main vendors controlling the space including Nuance Communications, ValidSoft Ltd and VoiceTrust.

Citation preview

  • 5/25/2018 Voice Biometrics

    1/12

    BIOMETRICS RESEARCH GROUP, INC.

    Speech and Voice Recognition

    White Paper

    Tis white paper differentiates between speech recognition and speaker/voice recognition and prvides a basic analysis of respective market size.

    Rawlson O`Neil King

    Lead Researcher, Biometrics Research Group, Inc.

    All inormation, analysis, orecasts and data provided by Biometrics Research Group, Inc. is or the exclusive uo subscribing persons and organizations (including those using the service on a trial basis). All such content copyrighted in the name o Biometric Research Group, Inc., and as such no part o this content may be repro-duced, repackaged, copies or redistributed without the express consent o Biometrics Research Group, Inc.

    All content, including orecasts, analysis and opinion, has been based on inormation and sources believed tobe accurate and reliable at the time o publishing. Biometrics Research Group, Inc. makes no representation oor warranty o any kind as to the accuracy or completeness o any inormation provided, and accepts no liabiliwhatsoever or any loss or damage resulting rom opinion, errors, inaccuracies or omissions affecting any part the content.

    2014, Biometrics Research Group, Inc.

  • 5/25/2018 Voice Biometrics

    2/12

    http://www.voicetrust.com/
  • 5/25/2018 Voice Biometrics

    3/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 3 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    TABLE OF CONTENTS

    Speaker/Voice Recognition and SpeechRecognition Differentiation

    Speech Recognition

    Speaker or Voice Recognition

    Research Methodology

    Biometrics Research Group, Inc. uses a combination o primary and secondary research methodologies to compile the necessary inormation or its research projections.

    Te conclusions drawn are based on our best judgment o exhibited trends, the expected direction the industrymay ollow, and consideration o a host o industry drivers, restraints, and challenges that represent the possibity or such trends to occur over a specic time rame. All supporting analyses and data are provided to the beo ability.

    Primary Research

    Biometrics Research Group, Inc. conducts interviews with technology providers, clients, and other organizatioas well as stakeholders in each o the technology segments, standards organizations, privacy commissions, andother inuential agencies. o provide balance to these interviews, industry thought leaders who track the implmentation o the biometric technologies are also interviewed to get their perspective on the issues o marketacceptance and uture direction o the industry.

    Biometrics Research Group, Inc. also applies its own proprietary micro- and macroeconomic modeling using regression analysis methodology to determine the size o biometric and related-industry marketplaces. Usingdatabases o both publicly and privately-available nancial data, Biometrics Research Group works to projectmarket size and market potential, in the context o the global economic marketplace, using proven econometr

    models.

    Secondary Research

    Biometrics Research Group, Inc. also draws upon secondary research which includes published sources such athose rom government bodies, think tanks, industry associations, internet sources, and Biometrics ResearchGroup, Inc.s own repository o news items. Tis inormation was used to enrich and externalize the primarydata. Data sources are cited where applicable.

    Market Size

    Nuance Communications

    ValidSof

    VoiceTrust

    4

    5

    5

    8

    9

    10

    11

  • 5/25/2018 Voice Biometrics

    4/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 4 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    Speaker/Voice Recognition and Speech RecognitionDifferentiation

    Speaker, or voice, recognition is a biometric modalitythat uses an individuals voice or recognition pur-

    poses. It is described as a process by which a machineor program receives and interprets dictation as well asunderstands and carries out spoken commands. Voicerecognition technologies help customers comply withprivacy, security, and saety requirements governed bylaw and by user expectations. It is a different technol-ogy than speech recognition, which recognizes wordsas they are articulated, which is not a biometric.

    Speaker or voice recognition processes rely on eaturesinuenced by both the physical structure o an indi-

    viduals vocal tract and the behavioral characteristics othe individual. Speaker recognition has been appliedmost ofen as a security application to control access tobuildings or sensitive data. Banking and nancial insti-tutions have employed speaker verication as a securi-ty mechanism on telephone-initiated transers o largesums o money. In addition to adding security, verica-tion is advantageous because it reduces the turnaroundtime on banking transactions. Speaker verication canalso be used by rms to limit data access to authorized

    personnel. Speaker recognition also provides a mecha-nism to limit the remote access o a personal worksta-tion to its owner or a set o registered users. In addi-tion to its use as a security device, speaker recognitioncan be used to trigger specialized services based on ausers identity.

    Speech Recognition

    Speech recognition, in contrast, is most ofen appliedin manuacturing or companies needing voice entry

    o data or commands while the operators hands areotherwise occupied. Related applications occur inproduct inspection, inventory control, command/con-trol, and material handling. Speech recognition alsonds requent application in medicine, where voiceinput can signicantly accelerate the writing o routinereports. Furthermore, speech recognition helps userscontrol personal workstations or interact with otherapplications remotely when touch-tone keypads are

    not available.

    Most systems or speech recognition include the ol-lowing ve components:

    1. A speech capture device or input.Tis usually cosists o a microphone and associated analog-to-digitconverter, which digitally encodes the raw speechwaveorm.

    2. A digital signal processing module.Te DSPmodule perorms endpoint (word boundary) detec-tion to separate speech rom non-speech, converts thraw waveorm into a requency domain representatiand perorms urther windowing, scaling, ltering, adata compression. Te goal is to enhance and retain

    only those components o the spectral representa-tion that are useul or recognition purposes, therebreducing the amount o inormation that the patternmatching algorithm must contend with. A set o thespeech parameters or one interval o time (usually10-30 milliseconds) is called a speech rame.

    3. Preprocessed signal storage.Here, the prepro-cessed speech is buffered or the recognition algorith

    4. Reerence speech patterns.Stored reerence pat-terns can be matched against the users speech samponce it has been preprocessed by the DSP module. Tinormation is stored as a set o speech templates or generative speech models.

    5. A pattern matching algorithm.Te algorithm mcompute a measure o goodness-o-t between thepreprocessed signal rom the users speech and all thstored templates or speech models. A selection procchooses the template or model (possibly more than

    one) with the best match.

    Five actors can be used to control and simpliy thespeech recognition task:

    1. Isolated words.Speech consisting o isolated wor(short silences betweenthe words) is much easier to recognize than continuous speech because word boundaries are diffi cult to

  • 5/25/2018 Voice Biometrics

    5/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 5 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    nd in continuous speech. Also, co-articulation e-ects in continuous speech cause the pronunciation oa word to change depending on its position relative toother words in a sentence. Error rates can be reducedby requiring the user to pause between each word.

    However, this type o restriction places a burden onthe user and reduces the speed with which inormationcan be input to the system.

    2. Single speaker.Speech rom a single speaker isalso easier to recognize than speech rom a variety ospeakers because most parametric representations ospeech are sensitive to the characteristics o the par-ticular speaker. Tis makes a set o pattern-matchingtemplates or one speaker perorm poorly or anotherspeaker. Tereore, many systems are speaker depen-

    dent - trained or use with each different operator.Relatively ew speech recognition systems can be usedby the general public. A rule o thumb used by manyresearchers is that or the same task, speaker-depen-dent systems will have error rates roughly three to vetimes smaller than speaker-independent ones. Oneway to make a system speaker independent is simply tomix training templates rom a wide variety o speakers.A more sophisticated approach will attempt to look orphonetic eatures that are relatively invariant between

    speakers.

    3. Vocabulary size.Te size o the vocabulary owords to be recognized also strongly inuences rec-ognition accuracy. Large vocabularies are more likelyto contain ambiguous words than small vocabularies.Ambiguous words are those whose pattern-matchingtemplates appear similar to the classication algorithmused by the recognizer. Tey are thereore harder todistinguish rom each other. O course, small vocabu-laries composed o many ambiguous words can be par-

    ticularly diffi cult to recognize. Te amount o time ittakes to search the speech model database also relatesto vocabulary size. Systems containing many patterntemplates typically require pruning techniques to cutdown the computational load o the pattern-matchingalgorithm. By ignoring potentially useul search paths,pruning heuristics can also introduce recognition er-rors.

    4. Grammar.Te grammar o the recognition domadenes the allowable sequences o words. A tightlyconstrained grammar is one in which the number owords that can legally ollow any given word is smalTe amount o constraint on word choice is reerred

    to as the perplexity o the grammar. Systems with lowperplexity are potentially more accurate than thosethat give the user more reedom because the systemcan limit the effective vocabulary and search space tthose words that can occur in the current input con-text.

    5. Environment.Background noise, changes in micphone characteristics, and loudness can all dramati-cally affect recognition accuracy. Many recognitionsystems are capable o very low error rates as long as

    the environmental conditions remain quiet and controlled. However, perormance degrades when noiseintroduced or when conditions differ rom the training session used to build the reerence templates. ocompensate, the user most usually has to always weaheadset-mounted, noise-limiting microphone.

    Speaker or Voice Recognition

    Voice recognition has a history dating back some

    our decades and uses the acoustic eatures o speechthat have been ound to differ between individuals.Tese acoustic patterns reect both anatomy (such asize and shape o the throat and mouth) and learnedbehavioral patterns (such as voice pitch and speakinstyle).

    Tere are two major applications between voice rec-ognition technologies and methodologies. I a speakclaims to be o a certain identity and the voice is useto veriy this claim, this is called verication or auth

    tication.

    On the other hand, identication is the task o de-termining an unknown speakers identity. In a sensespeaker verication is a 1:1 match where one speakevoice is matched to one template (also called a voicprint or voice model) whereas speaker identicatis a 1:N match where the voice is compared against Ntemplates.

  • 5/25/2018 Voice Biometrics

    6/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 6 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    Voice recognition is a popular choice or remoteauthentication due to the availability o devicesor collecting speech samples, such as telephonesystems and computer microphones. Due to its

    ease o integration, speaker recognition is differentrom some other biometric methods in that speechsamples are captured dynamically or over a periodo time, such as a ew seconds. Analysis occurs ona model in which changes over time are monitored,which is similar to other behavioral biometricssuch as dynamic signature, gait, and keystrokerecognition.

    Te physiological component o voice recognitionis related to the physical shape o an individuals

    vocal tract, which consists o an airway and thesof tissue cavities rom which vocal sounds origi-nate. o produce speech, these components workin combination with the physical movement o thejaw, tongue, and larynx and resonances in the nasalpassages. Te acoustic patterns o speech comerom the physical characteristics o the airways.

    Motion o the mouth and pronunciations are thebehavioral components o this biometric. Tere are

    two orms o speaker recognition: text dependentknown as constrained mode and text indepen-dent known as unconstrained mode.

    In a system using text dependent speech, theindividual presents either a xed or promptedphrase that is programmed into the system and canimprove perormance especially with cooperativeusers.

    A text independent system has no advance

    knowledge o the presenters phrasing and is muchmore exible in situations where the individualsubmitting the sample may be unaware o the col-lection or unwilling to cooperate, which presentsa more diffi cult challenge. Speech samples arewaveorms with time on the horizontal axis andloudness on the vertical access. Te speaker recog-nition system analyzes the requency content o the

    speech and compares characteristics such as thequality, duration, intensity dynamics, and pitch othe signal.

    In text dependent systems, during the collection

    or enrollment phase, the individual says a shortword or phrase -- reerred to as an utterance --typically captured using a microphone that can beas simple as a telephone. Te voice sample is con-verted rom an analog ormat to a digital ormat,the eatures o the individuals voice are extracted,and then a model is created.

    Most text dependent speaker verication sys-tems use the concept o Hidden Markov Models(HMMs), random based models that provide a

    statistical representation o the sounds producedby the individual. Te HMM represents the under-lying variations and temporal changes over timeound in the speech states using the quality dura-tion / intensity dynamics / pitch characteristicsmentioned above.

    Another method is the Gaussian Mixture Model,a state-mapping model closely related to HMM,that is ofen used or unconstrained text inde-

    pendent applications. Like HMM, this methoduses the voice to create a number o vector statesrepresenting the various sound orms, which arecharacteristic o the physiology and behavior o theindividual.

    Tese methods all compare the similarities anddifferences between the input voice and the storedvoice states to produce a recognition decision.Afer enrollment, during the recognition phase, thesame quality / duration / loudness / pitch eatures

    are extracted rom the submitted sample and com-pared to the model o the claimed or hypothesizedidentity and to models rom other speakers.

    Te other-speaker or anti-speaker models con-tain the states o a variety o individuals, not in-cluding that o the claimed or hypothesized identi-ty. Te input voice sample and enrolled models are

  • 5/25/2018 Voice Biometrics

    7/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 7 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    compared to produce a likelihood ratio, indicat-ing the likelihood that the input sample came romthe claimed or hypothesized speaker. I the voiceinput belongs to the identity claimed or hypoth-esized, the score will reect the sample to be more

    similar to the claimed or hypothesized identitysmodel than to the anti-speaker model.

    Te seemingly easy implementation o speaker rec-ognition systems contributes to the process majorweakness and susceptibility to transmission chan-nel and microphone variability and noise.

    Systems can ace problems when end users haveenrolled on a clean landline phone and attemptverication using a noisy cellular phone. Te in-

    ability to control the actors affecting the inputsystem can signicantly decrease perormance.Speaker verication systems, except those usingprompted phrases, are also susceptible to spoongattacks through the use o recorded voice. Anti-spoong measures that require the utterance o aspecied and random word or phrase are beingimplemented to combat this weakness.

    For example, a system may request a randomly

    generated phrase, to prevent an attack rom apre-recorded voice sample. Te user cannot antici-pate the random sample that will be required andthereore cannot successully attempt a playbackspoong attack on the system.

    Current research in the area o text independentspeaker recognition is mainly ocused on mov-ing beyond low-level spectral analysis. Althoughthe spectral level o inormation is still the drivingorce behind the recognitions, using higher-level

    characteristics with the low level spectral inorma-tion is becoming a popular laboratory technique.

    Speaker recognition characteristics such as rhythm,speed, modulation and intonation are based onpersonality type and parental inuence; andsemantics, idiolects, pronunciations and idiosyn-crasies are related to birthplace, socio-economic

    status, and education level. Higher-level character-istics can be combined with the underlying low-level spectral inormation to improve the peror-mance o text independent speaker recognitionsystems.

    Te great advantage o speaker verication is itswidespread acceptability and ease o use, as wellas the relative inexpensiveness o basic systemscompared to other biometric options. When voiceauthentication is integrated into a telephony sys-tem, it also creates a very riendly customer serviceenvironment. As a consequence, respondents innumerous surveys have indicated that they preerrecognition technology or biometric identica-tion. Voice recognition is characterized by non-

    contact and non-intrusiveness. Due to ease o use,voice recognition is a growing market segment,especially in the nancial sector.

    Financial institutions have identied voice biomet-rics as one o the best means to secure its client ac-counts and nancial inormation. Voice biometricscompares various characteristics drawn rom a per-sons voice such as inection, pitch, dialect, amongothers, and matches that with data captured. For

    voice recognition to work it requires banks andother nancial institutions to register their clientsvoice patterns and correlate them to personal dataor incorporation into a database.

    Voice biometrics solutions allow customers toveriy their identity simply by speaking, makingit easier and aster to gain access to secure bank-ing and insurance services by way o mobile apps,telephone and Web channels. Voice biometric solu-tions eliminate the need or PIN-based password

    or interrogation-based authentication methods,or can be used to add another level o security tothose systems.

    Banks that deploy voice biometrics to automate thelogin process not only enhance customer satisac-tion levels, but dramatically reduce their customercare costs through increased automation rates.

  • 5/25/2018 Voice Biometrics

    8/12

    BIOMETRICS RESEARCH GROUP, INC.

    Due to the versatility, along with consumer con-dence in voice biometric technology, the Biomet-rics Research Group expects voice biometrics tobe the asting growing technology modality in the

    banking sector.

    Surveys we have analyzed have ound that consum-ers preer voice recognition technology or biomet-ric identication. According to a survey conductedby I provider Unisys, the biometric modalitiesranked by consumer preerence are: voice recogni-tion (32 percent), ngerprints (27 percent), acialscan (20 percent), hand geometry (12 percent), andiris scan (10 percent). As a result, the BiometricsResearch Group projects that voice recognition will

    be widely adopted. We project the technology willnot only be implemented in bank calling centersthroughout the world, but ast growth will also bedriven by the continued rapid worldwide adoptiono mobile smartphone and superphone tech-nologies.

    Banks are in the preliminary stages o testing androlling out new voice biometric technologies ormobile devices. In North America last year, US-

    AAM, the independent bank and insurance bro-kerage that caters to the U.S. military, developeda voice recognition service that will eventuallyallow its entire mobile phone customer base tomake natural language inquiries or a wide range obanking services.

    USAAs voice recognition app is currently beingtested by a group o employees but is slated or useby the banks 6.3 million account holders early nextyear. Te bank has stated publicly that the applica-

    tion has tremendous potential to make banking:simpler, aster and more satisying on mobiledevices.

    Te bank cites the act that military personnel areofen quite mobile and would like to make greateruse o advanced smartphone and superphonetechnologies. Te bank also cited statistics rom its

    voice biometric technology supplier Nuance thatwhile over 50 percent o smartphone and super-phone owners have installed a mobile banking appon their device, only 27 percent actually use it ona regular basis. Consumers say improvements in

    a ew areas would increase the use o smartphoneand superphone banking apps signicantly. Tirty-our percent say they would appreciate a seamlessaccess to a live agent when they need one, and 21percent want their mobile apps to include moresel-service tasks. Eighteen percent would simplysettle or an app that was easier to use.

    echnology suppliers like Nuance are betting thatvoice biometrics will be the magic sauce that im-proves banking customer experiences. Other banks

    are also examining implementation o the voicebiometrics technology. Spanish bank BBVA is alsocurrently developing a Siri-like banking applicationor iPhones and iPads at its U.S. subsidiary.

    Further, as BiometricUpdate.com previouslyreported last year, major banks such as ANZ areseriously beginning to study implementing bio-metrics over the next three to ve years to improvethe quality o its banking services. ANZ has made

    public statements that it projects it will take two tothree years beore commercialization o biomet-rics in banking is achieved. However, the bank ispositioning itsel or the implementation o thenew technologies that will be designed to simpliyANZs distribution networks and its products andprocesses, while providing customers with addi-tional mobile and exible banking options, whileconcurrently improving the capability o ront-linestaff.

    Market Size

    Te market size or both speech recognition andvoice recognition continue to grow concurrently.While many research rms do not differentiate be-tween the speech recognition and speaker recogni-tion marketplaces in their analysis, the BiometricsResearch Group does.

    Page 8 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

  • 5/25/2018 Voice Biometrics

    9/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 9 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    We recognize that various industries such asairlines, banks, and brokerages depend on voicerecognition unctionality, not only to enhance theircustomer contacts, but also to comply with security

    requirements dictated by the law and the securityconscious expectations o customers. As a result,voice recognition as a biometric modality hasbeen growing. Biometrics Research Group, Inc.estimates that voice recognition will reach US$2.5billion in revenue by 2015, mainly driven by thebanking sector.

    Due to growing interest in providing consumerswith cutting edge technology, while concurrentlyenhancing banking security, Biometrics Research

    Group expects more nancial institutions to de-velop and deploy biometrics, and as a consequence,expects revenue growth or voice biometrics togrow. Our research estimates that at least US$200million was spent on voice biometrics in the bank-ing sector in 2012. We estimate that at least US$750million will be spent on voice biometrics in thebanking sector by 2015.

    While we estimate that growth will continue to

    occur, its pace will be dictated by technologicaldevelopments. One o the major actors that wehave determined is restraining the voice recogni-tion marketplace are system aults that cause alseinputs produced by poor communication linkages,loud external conversations, barking dogs, scream-ing children and the like. As these system errorsbegin to be addressed by technological innovation,the Biometrics Research Group anticipates expo-nential increases in year-over-year earnings orvendors in the voice recognition space.

    Biometrics Research Group also expects that theincreased use o mobile devices will also drivevoice recognition development. Currently, securitymeasures to lock smartphones, and the data con-tained within, include: our-digit passcodes, drawpattern unlock algorithms and increasingly nger-print impressions. Te most obvious and conceiv-

    able security measure or mobile devices howeveris voice recognition. We would expect that oncevoice recognition system error issues are resolved,major smartphone manuacturers will move to-wards wholesale implementation o the technology

    or device lock purposes. Such a move will expandcompound annual growth rates and increase totalmarket value, though we cannot estimate whenthese developments might occur.

    In differentiating the marketplace, BiometricsResearch Group notes that the speech recogni-tion market is much larger and more mature. Weestimate speech recognition sofware sales werevalued at US$11.5 billion in 2010 and will reachUS$20.1 billion in 2015. Sofware packages include

    automatic speech recognition and text-to-speechsystems. It should be noted that because we donot consider speech recognition itsel a biometric,we do not include revenue projections rom thissegment in our global biometric revenue estimates.Due to vendor crossover in concurrent marketshowever we do singularly identiy main champi-ons in the unied space. Biometrics ResearchGroup believes that the main vendors controllingthe space, based on market share, include: Nuance

    Communications, ValidSof Ltd, and Voicerust.

    Nuance Communications

    Nuance Communications is an American multina-tional computer sofware technology corporation,headquartered in Burlington, Massachusetts thatprovides speech and imaging applications. Currentbusiness products ocus on server and embeddedspeech recognition, telephone call steering systems,automated telephone directory services, medical

    transcription sofware & systems, optical characterrecognition sofware, and desktop imaging sof-ware. Te company also maintains a small divisionwhich does sofware and system development ormilitary and government agencies.

    Nuance was ounded in 1994 as a spin-off o SRIInternationals Speech echnology and Research

  • 5/25/2018 Voice Biometrics

    10/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 10 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    (SAR) Laboratory to commercialize the speaker-independent speech recognition technology devel-oped or the U.S. government at SRI. Initially basedin Menlo Park, Caliornia, Nuance deployed theirrst commercial large-scale speech application in

    1996. Teir initial route to market was through callcenter automation. Call centers had just central-ized the branch-offi ce telephone handling unctionthroughout many large companies. Te highestcost o running call centers is the cost o staff. Earlyprojects were completely developed by Nuance toprove commercial practicality and provide effi -ciency benets.

    Afer ormation, the company expanded as a resulto organic growth, mergers, and acquisitions. In

    October 2005, Nuance merged with ScanSof, aellow commercial large scale speech applicationbusiness. ScanSof a Xerox spin-off had its rootsin Kurzweil Computer Products, a sofware com-pany that developed the rst omni-ont characterrecognition system. In October 2011, unconrmedresearch suggested that Nuances servers powerApples iPhone Siri voice recognition application,reaffi rming the rms tradition o innovation. Term is best known in the consumer marketplace

    or Dragon, the worlds bestselling speech recogni-tion sofware or Macintosh and PC.

    ValidSof

    ValidSof is a UK-based security sofware company,providing telecommunications-based multi-actorauthentication, identity and transaction verica-tion technology. Te company was ounded in2003 and provides business-to-business mobilesecurity and cloud security products, includ-

    ing a multi-actor authentication platorm calledSMAR (Secure Mobile Architecture or Real-timeransactions), which uses mobile telecommunica-tion channels and devices and includes a propri-etary voice biometric engine. Tese solutions aredesigned to reduce electronic raud and saeguardconsumer privacy when using Internet and mobilebanking, credit, debit card and both mobile and

    xed line telephony channels. ValidSofs productsare designed to veriy the authenticity o bothparties to a transaction (mutual authentication),ensure the delity o telecommunication channels(secure communications), and conrm the integ-

    rity o transactions themselves (transaction veri-cation).

    VALid-SVP, otherwise known as the VALid Speak-er Verication Platorm, is ValidSofs proprietaryvoice biometrics solution, based on a modular andpluggable architecture, which allows organisationsto easily integrate voice verication into a broaderauthentication platorm. VALid-SVP is describedas a leading-edge voice biometrics engine thatsupports text-dependent, text-independent and

    conversational voice verication (biometric plusknowledge) models.

    It includes a ull enrolment module, incorporatesliveness validation, replay and synthesis attackmitigation techniques and pseudo device thefdetection as well as many other leading securitytechniques as part o its extensive layered securitymodel. It operates on any channel the VALid plat-orm supports, including Internet, mobile, IVR and

    contact-center and can also be used cross-channel.Additionally, VALid-SVP also utilizes more ad-vanced techniques to ensure the integrity o voiceverication in real-world usage. Tis includestechniques such as context aware voice vericationand dynamic threshold adjustments. As the VALidplatorm understand many acets o the overall au-thentication all o these aspects can be consideredwhen a conclusion is reached. Te rms aim is toallow the user to transact securely in the mannerthey wish to operate within in the most rictionless

    way.

    ValidSofs VALid-IMA is described as an in-band mobile authentication solution because theprimary channel o voice communication betweenthe app and the IMA Server is not actually thevoice channel, i.e. through a phone call to/romthe handset, but via the data channel on which

  • 5/25/2018 Voice Biometrics

    11/12

    BIOMETRICS RESEARCH GROUP, INC.

    Page 11 | Biometric Update Special Report | May 2014 | www.biometricupdate.com

    the app it is integrated with is communicating toits own host. In-band voice delivery, coupled withthe unctionality o the IMA platorm and theadvanced eatures o ValidSofs Voice Biometricengine, delivers a combination o service delivery

    cost reduction, low user riction, biometric per-ormance improvement, secure and simple enrol-ment/activation o the app and user-riendly ea-tures that can eliminate alse-negatives, i.e. denialo access to a legitimate customer. VALid-IMA canbe integrated into apps on any operating systemand on any handset because it simply uses a well-dened XML protocol or communication with theIMA Server. It can also be deployed as either anin-house implementation or sofware-as-a-service.

    VoiceTrust

    Voicerust is a global provider o voice biometricssolutions that enable highly secure and convenientauthentication. Te worlds largest banks, insurancecompanies, call centers, and enterprises rely on therms solutions to protect access to business andconsumer applications, prevent identity thef, anddeliver a more enjoyable authentication experience.Founded in 2000, the privately-owned company is

    headquartered in oronto, Canada with additionaloffi ces in the U.S., Germany, Netherlands, andUAE.

    Te companys voice biometrics platorm, knownas VAssure, can veriy identity based on text de-pendent and text independent authentication. Tisauthentication platorm consists o both an interac-tive call management platorm and voice verica-tion platorm, using a comprehensive multiactorauthentication algorithm and stored biometric

    templates. Voicerust also recently released Vinalk, a new solution or dynamic caller verica-tion. V inalk is an off-the-shel, ast deployablesolution that addresses the needs o call centerswanting to veriy the authenticity o a caller whileengaged in natural dialogue with the customer. Vinalk dynamically compares the speakers voicewith the existing claimant voice print to provide

    additional assurance that the caller is who he or sheclaims to be. Te solution will be integrated intothe Genesys GVP platorm. Te Genesys Interac-tive Voice Response Platorm is a sofware only,standards-based voice portal that enables busi-

    nesses to provide cost-effective customer interac-tions 247 or voice, video, and web-based interac-tions. Beyond traditional interactive voice response(IVR) systems, it provides touchtone access toapplications and incorporates speech recognitiontechnology and video or conversational exchangeto identiy and resolve customer requests.

    Voicerusts other recent clients and pilot projectsinclude a major U.S.-based bank, an Irish bank, aglobal U.S.-based apparel manuacturer, a Euro-

    pean insurance provider, a European governmentagency as well as several nancial service providersand a telecommunication rm in the Middle East.

    All o the above vendors are attempting to domi-nate the combined speech and voice recognitionmarketplaces by providing differentiated productsand services designed to provide a competitiveedge to their clients. Biometrics Research Groupexpects more stiff competition between these

    vendors as demand or voice recognition systemsgrows within the banking and mobile sectors.

  • 5/25/2018 Voice Biometrics

    12/12

    About the Biometrics Research Group, Inc.

    Biometrics Research Group, Inc. provides proprietary research, consumer and business data, custom consultinand industry intelligence to help companies make inormed business decisions.

    We provide news, research and analysis to companies ranging rom Fortune 500 to small start-ups through maket reports, primary studies, consumer research, custom research, consultation, workshops, executive coner-ences and our ree daily BiometricUpdate.com news service.

    Biometrics Research Group has positioned itsel as the worlds preerred supplier o pure-play market researchand consultancy services ocused on the biometric marketplace, which particular ocus on the law enorcemenand national security sectors. Our portolio o white papers and ull research reports is based upon high-qualquantitative analysis, allowing our clients to gain deeper understanding o the marketplace.

    We customize our research design, data collection, and statistical reporting using proprietary micro- and macreconomic modeling and regression analysis.

    Trough integration o our research results with qualitative analysis rom our BiometricUpdate.com news ser-vice, we provide actionable business analysis.