what is speech recognition - specialty answering … speech recognition technology has advanced to...

[Type text]

1

© Specialty Answering Service. All rights reserved.

Contents

1 Introduction to Speech Recognition ..................................................................................................... 2

2 Evolution of Speech Recognition Algorithms ........................................................................................ 3

3 Practical Applications of Speech Recognition Technology ................................................................... 5

3.1 Real Life Examples ......................................................................................................................... 5

4 Advantages of Using Speech Recognition in Call Centers ..................................................................... 7

4.1 Cost Savings Calculation ................................................................................................................ 7

4.2 Getting Ready for Speech Applications ......................................................................................... 8

4.3 Cost‐Benefit Analysis .................................................................................................................... 9

4.4 In‐house or Outsourced? .............................................................................................................. 9

4.5 Structured Approach ................................................................................................................... 10

4.6 Evaluating Service Providers ....................................................................................................... 10

4.7 New Opportunities for Speech Applications ............................................................................... 11

4.8 Identifying the Vendor ................................................................................................................ 12

5 Road Ahead for Speech Recognition as a Viable Option for Customer Service .................................. 15

6 Conclusion ........................................................................................................................................... 18

7 References .......................................................................................................................................... 19

2


1 IntroductiontoSpeechRecognition

Today, speech recognition technology has advanced to such a stage that you can have a realconversation with a computer where it not only understands what you say, but is also able tocommunicatebacktoyou.

Foralongtime,speechrecognitionwasintheresearchstageswherepracticaluseswereminimal.But, around the late 90s, the technology grew in a rapid pace and it became a financially viableoption for several industries, especially call centers, which started using speech recognition incombinationwithIVRsystemstonotonlyenhancethecustomerexperience,butalsotocutcosts.

Someofthekeyfactorsthatcontributedtotheriseofspeechrecognitionincludecomputers’fasterprocessingcapacitiesandtheirdropinprices,aswellasadvancesinspeechrecognitionalgorithms,whichimprovedtheiraccuracysignificantly.

In theearlydaysofspeechrecognition, itwas thoughtofasa technologywithagreatamountofpracticalpotential.Researchpredictedthatspeechrecognitionwouldbea$1billionindustrybytheendof the90s. Itwasexpectedthat inNorthAmericaalone,callcenterapplicationuseofspeechrecognitionwouldgrowfrom$60millionto$150millionintheperiod1996‐2001.

Today, most large call centers are actively considering speech recognition as part of theirtechnologystrategy,althoughnegativecustomerperceptionsandlowreallifeaccuracyrateshaveledtosomebadpressforspeechrecognitioninrecentyears.

There are several products and services that fall within the gamut of speech recognition. Theunderlyingtechnologyofspeechrecognitionisthespeechrecognitionengine–whichconvertsthespokenword into thewrittenword.Relatedproducts in this field include text tospeechengines,voiceplatforms,developmenttoolkitsthathelpsoftwareprofessionalstobuild,maintain,andfinetunespeechrecognitionapplications,aswellastoolstomonitortheaccuracyofthoseapplications.Today, thereare ‘plugandplay’packages thatallowcallcenters tocustomizeand implement thesolutionwithinamatterofafewmonths.

In the early days of speech recognition technology, the focus was on fine‐tuning the speechrecognitionenginewithdifferentkindsofalgorithmsinanattempttoimprovetheaccuracyoftheengine.However,thishasbeensuccessfullyconquered,andapplicationprovidersare focusingonprovidingvalueaddedfeaturessuchaseaseofusetoenablefasteracceptanceintheindustry.

3


2 EvolutionofSpeechRecognitionAlgorithms

Speechrecognitionhascomealongwayintermsofunderstandingnormalconversations.Thishasdrasticallycutdowntheerrorrateasalgorithmicimprovementsinthelastdecadehavemanagedtobringdownerrorsbyabout30%everyyear.Thishasmadespeechrecognitionaviablepracticalsolution.

Leadingspeechapplicationscannowachieveaccuracyintherangeof90%‐98%dependingonthenature of the transaction. One of the key reasons why accuracy has improved is because thevocabularyofmost current applications includesmore than30,000wordsas compared to a fewdozenwordsintheearly90s.Thisinturnenablesdeveloperstoincorporatetheabilitytorespondtocomplexcustomerrequeststhatcanbephrasedinmorethanoneway.Contextualunderstandinghasalsocontributedtoimprovingtheuserexperiencewhileusingtheseapplications.

Anotherkeyalgorithmicchangeistheabilitytocorrectlyinterpret‘naturallanguage’orcontinuousspeech. In the initial days, users had to speak specific phrases in clipped form for the engine torecognize and respond effectively. However, the latter versions of the software allow people tospeakinfullsentencesinanaturalmanner,thusmakingtheinteractionmorerealandlifelike.Thisalso helps to drastically reduce call time by allowing users to request multiple servicessimultaneously. Practically, call centers have reported asmuch as 30% to 40% reduction in calltimeduetothisfeature.

Even after the advent of speech recognition technology, for a long time call centers remainedcontent with just IVR systems. But common standards such as VoiceXM and SALT, as well assupportfromestablishedvendorssuchasIBMandMicrosoft,madesurethatapplicationsbasedonspeechrecognitiontechnologywereadoptedbythemainstream.Someofthekeycostsavingsarosefromincreasedautomationofcalls,partialautomationofcallsroutedtoagents,andmoreaccurateskills‐basedroutingofcalls.Indirectbenefitsalsoincludedincreasedcustomersatisfaction,betterbrandawareness,andtheabilitytoretaincustomerswhowouldotherwiseswitchtocompetitionwiththiscutting‐edgetechnology.

Initially, speech applications were deployed in industries such as financial services and airlineswhereitwasdifficulttoentertickersymbolsandtraveldatesontoakeypad.Thehugesuccessoftheseimplementationsresultedinotherindustriesadoptingthetechnology.However,intheearlydays,speechsystemsonlyreplicatedtheIVRmenustructurewithpromptssuchas“Pressorsay2”.Thus, insteadofflatteningthemenustructure, itresultedinlongercallsduetothelongprompts.Soon,theindustryrealizedthatinordertofullybenefitfromthetechnology,awell‐designedvoiceuserinterface(VUI)wasnecessary.Awell‐designedVUIhadtocatertoafirsttimecalleraswellasanadvanceduserwhomaywanttoskipseveralstepsforwardorgetmultipleinformationrequestsprocessedsimultaneously.

Nowadays, sophisticated applications employ statistical language modeling and open grammaralgorithmstostatisticallyinferthereasonofthecall.Thistechniquehelpsineffectivecallrouting,by asking callers open ended questions such as “Please explain the reason for your call”. These

4


algorithmsrelyonastatisticalanalysisofauser’s inputrather thanaone‐to‐onematchofeveryinputtoapre‐determinedsetofresponses.Thisenablesthesystemtounderstandamuch largersetofinputsandprovideaprobableinterpretationandresponse.

Awell‐designedVUIshouldhaveapersonalityandtonethatispleasingtothecallerascallersoftenfail to distinguish between a VUI and a live agent. The first step in designing a robust VUI is toevaluateinboundcallstounderstandthewaycustomersinteractwithliveagentsandthewaytheyphrasetheirquestionsandresponses.Afterthis,variousprototypesaredesignedandtestedpriortochoosingtherightonetoimplement.ThekeythingtorememberisthatdesigningagoodVUIisan ongoing process that requires initial planning aswell as ongoingmonitoring and refinement.This ensures that an optimal system is established to meet the callers’ needs while avoidingfrustrationscausedbyincorrectresponsesfromthesystem.Inturn,thiswillhelptheapplicationtoreachitsmaximumpotential,providingagoodreturnoninvestment.

5


3 PracticalApplicationsofSpeechRecognitionTechnology

Someofthekeyindustrieswherespeechrecognitionfindspracticalapplicationsincludebanksandfinancialinstitutions,airlinesandtelecommunications.Telecomcompaniesusespeechrecognitionfor offering services such asdirectory assistance and to complete collect calls.Airlinesuse it forflightreservationbypassengersaswellasflightschedulingbyemployees.Logisticscompaniesuseit for trackingpackages in transit. In financial institutions, speech recognitionhas beenused forhandling balance inquiries, responding to customer requests, helping customers execute simpletransactions such as applying for mortgages, getting stock quotes, and so on. Some of the keyfinancial institutions like American Express, Charles Schwab & Co., and NationsBank havesuccessfullyusedspeechrecognitiontechnologytobetterservetheircustomers.

Studieshaveshownthatcallersarelesslikelytoaskforaliveagentwhencateredtobyaspeechapplication versus a touch‐tone menu. Two key factors contributing to the wide acceptance ofspeechapplicationsbythecallersareimprovedvoiceinterfacesandtheuseofcellphones,whichmaketouchtoneresponsescumbersometohandlewhileonthemove.

Oneof thekeyadvantagesofusing speech recognition technology is thatwhenused in the rightmannerincanhelptocutoperationalcostsbyalargemargin.Forexample,ahumanagentwouldcost nearly $30,000 per yearwhereas voice recognition can cost around $3,000 per phone line.Thus,thereturnoninvestmentcanbeseeninasfastasafewmonths.

Althoughbrokerage firmswere among the early adopters of the technology, bankswerenot tookeenonintroducingthesystem.Thiswasduetointegrationissueswiththelegacysystems,andthemultiple applications that were typically used to handle different products and sales channelsofferedbythebank.

Oneofthekeysuccessfactorsofapracticalapplicationisthatitismodularandcanbeintegratedwithothersoftwareusedinthecallcenter.Thisisbecausecallcentersoftenwanttoimplementthesolution inphases,andunless theapplicationcan talk to the databasesandotherapplications, itwillnotbewidelyused.

3.1 RealLifeExamples

Inthissection,wewill lookatsomeofthereallifeexamples,wheresuccessfulimplementationofspeechapplicationshasprovidedbenefitsforcallcenters.

OneoftheearliestcommercialdeploymentsofspeechrecognitiontechnologywastheVoiceBrokerapplication implementedatCharlesSchwab&Co in1996. Itwasusedtoprovidecustomerswithstockquotesandinformationaboutothermarketindicators.Thankstotheeffortsofthecompanytoanalyzeeacherrorandconstantlytweakthesystem,theerrorratehasdroppedfromnearly30%to about 4%‐5%. Schwab has also implemented biometric speech software to verify a caller’sidentity with the help of his or her voice. This is done by matching the voice against a storedelectronicvoiceprint.Thefinancialserviceindustryisquiteaptforspeechrecognitiontechnology

6


useasitreceivesalargevolumeofcallseveryyear,andevenamarginalreductionincalldurationcanprovidehugecostsavings.

Fidelity Investments has used a natural language application called FAST to help users performtransactionssuchasobtainstock,option,mutualfundandindexquotes,reviewaccountbalances,andevenexecutetrades.Infact,morethanthreequartersofcallsthatreachtheFidelitycallcenterarehandledautomaticallythroughtheIVRandspeechrecognitionplatforms,thusfreeingupagenttimetofocusonsalesandefficiencyimprovements.

Standard Life uses speech recognition for its life insurance and pension businesses, making itpossibletoauthenticatecallers,identifytheirrequirements,andproperlyroutecalls.Italsorelayscallerinformationtotheagent,sothatcallersdonothavetorepeattheinformationagain.Thishashelpedthemincreasecall‐handlingcapacityby25%andcallroutingaccuracyby66%.AtStandardLife,theVUIispersonalizedandknownas‘Sheila’.Thevoiceapplicationhasnotonlyimprovedcallrouting,butalsoincreasedcustomersatisfactionlevels.

ThegeneralinsurancecompanySuncorphasusedNaturalLanguageSpeechRecognition(NLSR)toreplaceitsIVRsystem.Thegrammaroftheapplicationincludesover0.1millionphraseswhicharespecifictothefinancialservicesindustry.Thecompanyhasmanagedtoreduceitscallwaittimestoabouthalfaminuteandthemisdirectedcallsarevirtuallynon‐existent.

Ladbrokes, a betting giant, has managed to successfully divert calls based on their nature.Transactionssuchasplacingabetoraninquiryofoddswerehandledautomaticallywhilerequestsformore‘customized’betswereroutedtoagents.Thecompanyhasalargedatabaseofhorsesandfootballplayersthatareupdatedonarealtimebasisandfeedsthedataintothevoiceapplication.

7


4 AdvantagesofUsingSpeechRecognitioninCallCenters

Theinitialinvestmentrequiredforapracticalapplicationofthesoftwareisconsiderablylargeanditcanbetothetuneof0.5–1million.However,foralargecallcenterthathandlesseveralmillioncallsperyear,evenasavingsof10centspercallcanbeaconsiderablylargeamount.Thebiggestcostsavingscomesfromautomatingalargernumberofcalls,thusreducingtheneedforliveagents.

4.1 CostSavingsCalculation

A sample calculation for expected cost savings and revenue enhancements that a call center cangeneratebyusingvoicerecognitionsoftwareisgivenbelow:

Assumptions:

No.ofincomingcallsperyear:3.6million

%ofcallsthatgotoaliveagent:34%of3.6million=1224000

% of live agent calls handled by a speech recognition application: 65% of 1224000 =795600

Averagedurationpercall:3minutes

Averagecostperliveagent:$30,000peryear

Telecommunicationlinecharges:$0.10perminute

Beforeintroductionofspeechrecognitionapplication:

No.ofcallsthataliveagentcanhandleinaday(360minutesofcalltimeperday):360/3=120calls

No.ofcallsperyearperagent:120*22*12=31680

Totalliveagentsrequired:1224000/31680=39agents

Beforeintroductionofspeechrecognitionapplication:

Savingsincalldurationduetospeechrecognitionapplication:20%

Average duration of calls which pass through speech recognition engine: = 3*80% = 2.4minutes

No.ofcallsinwhichtimesavingisobtained=795600

8


No.ofsuchcalls thata liveagentcanhandle inaday(360minutesofcall timeperday) :360/2.4=150calls

No.ofcallsperyearperagent:150*22*12=39600

Totalliveagentsrequiredforsuchcalls:795600/39600=20agents

Remainingcalls=122400–795600=428400calls

Totalliveagentsneededforremainingcalls=428400/31680=14agents

Totalagents:=20+14=34agents

No.ofagentssaved=39‐34=5

Savingsinpersonnelcosts:5*$30,000=$150,000

Savingsintelecommunicationcosts:=477360*$0.1=$47,736

Similarly, ifyouassumeretailrevenueof$600peraccount,evena1%increaseinsalesachievedfromtheincreaseinsalestimeavailablewillworkouttobealargeamount.

Oneof thechiefadvantagesofspeechrecognitionapplications is the fact that theytakeawaythecumbersomenatureofIVR,makingitunsuitableforcarryingoutcomplextransactionswithoutthehelpof a live agent. In the initialdaysof its success, callers also liked the fact that itoffered theabilitytoobtaininformationandconductcomplextransactions likethosethatyouwoulddoovertheInternetorthroughaphonecall.However,withthewidespreadpopularityofsmartphoneswithbuilt‐in internet browsing capabilities and the spread of 3G and 4G, the advantage of speechrecognition is slowly ebbing away. This is because customers today conduct transactions eitherthrough the web or through mobile commerce rather than waiting for the speech recognitionengineinthecallcentertounderstandandrespondtotheirrequests.

WhatspeechrecognitionhasdoneistomaketheIVRsysteminacallcentermoreuserfriendlybyallowingthecustomertodomoretransactionsinafasterandmorenaturalmanner.

4.2 GettingReadyforSpeechApplications

Ifyouareconsideringimplementingspeechrecognitionsystemsinyourcallcenter,thenansweringsomeofthesequestionsmayhelpyoureachadecision:

1. Is your average wait completion time for simple transactions such as routine balanceinquirytoohigh?

2. Doagentsspendtoomuchtimeinnon‐valueaddingactivitiessuchasvalidatingacaller’sidentity,routingthecalltothecorrectdepartment,andansweringsimplequestions?

9


3. Doyourcallerscallwhileonthemove,makingitdifficulttonavigatethroughatouch‐toneIVRmenu?

4. Areyouoperatinginahighlycompetitiveindustrywhereswitchingcostsforyourcallersislessthanyourcustomeracquisitioncosts?

If you answered any of these questions in the affirmative, then evaluating speech recognitionapplicationsasameanstoenhanceservicelevelsandproductivityinyourcallcenterwouldbeofbenefit.

4.3 Cost‐BenefitAnalysis

Beforeimplementingspeechapplications,thecallcentershouldperformacost‐benefitanalysistochoosebetweenacompleteredesignofthecallcenterversusatacticalpiecemealadd‐onapproach.AhighROIapplicationthatiseasytoimplementwouldhelptogetthenecessarybuy‐insfromallstakeholders for future rollouts ofmore applications. It is important to have a complete plan inplace so that the infrastructure and vendors chosen for the initial phases are strong enough tohandlesubsequentphasesoftheproject.

Evaluate theprojectnot just fromanROIperspective,butalso fromthepointofwhether it is inalignment with your organizational objectives, and whether it would help you achieve yourstrategic, financial,marketing, and customer service goals. Some of the other questions that youshouldanswer,beforeyougoaheadwithaspeechapplicationaregivenbelow:

1. Isyouroverridingfocusoncuttingcostsoronimprovingcustomerservice?Whilemostcallcenters would want a mixture of both, it is always better to identify which is moreimportantthantheother,intheeventthatyouneedtomakesometrade‐offs.

2. Arecallerssatisfiedwiththecurrentlevelofserviceoffered?

3. Domanagementteamsandagentsfeelcomfortableaboutswitchingapplications?

4. Whowillbeelectedintheorganizationtochampiontheassociatedchangemanagement?

5. Dothespeechapplicationsgiveyouacompetitiveedgeoverotherplayersintheindustry?

6. Doyouhavesufficientbudgetstopayfortheapplicationanditsmaintenance?

4.4 In‐houseorOutsourced?

Onceyouhavedecidedtogoaheadwithspeechapplications,thenextstepistodecidewhetheryouwouldwanttorunitin‐house,orwouldhaveitoutsourcedtoathirdparty.Someofthequestionstoconsiderinthisregardaregivenbelow:

1. Do you have the in‐house expertise to customize the application for grammar, dialoguedesignandotherhumanfactors?

10


2. Doyouhaveenoughbandwidthtomanagetheapplication24/7?

3. Doyouhavemulti‐sitebackupsystemsinplace?Ifnot,areyouwillingtoinvestinone?

Eachapplicationisdifferent,butasystemthatcanspeech‐enablemostofyourcallswillcostyouanywherebetween$1.5millionto$4million.Ifyoudonothavethatkindofbudgets,thenitisbesttocontractouttheprojecttoanoutsourceproviderwhohasademonstrateddialoguedesignandtechnical expertise in this area. By outsourcing your speech application rollout, you can startreapingtheassociatedbenefitsmuchfasterthanifyouhandleitin‐house.Additionally,youwillnothave toworry about software upgrades andmaintenance, enabling you to focus your efforts onimprovingthecallcenter’scoreoperations.

4.5 StructuredApproach

Ifyouplan togoaheadwith the implementationofaspeechapplication inyourcall center, thenfollowingastructuredapproachwillimprovethechancesofyourproject’ssuccess.Theguidelinesbelowcanassistwiththeproject‐planningphase:

1. Formasteeringcommitteewithrepresentationsfromuserdepartmentssuchasmarketingandcustomerserviceaswellasdepartmentssuchas thecall centeroperations teamandthetechnicalteam.

2. Analyzecalldatabycollectingdetailsoncallvolumes,andcalltypesfromyourIVRandcallrouting systems. You should be able to identify specific points where the caller hasproblemswith the IVR system and opts for the live agent. Youmust also analyze agent‐handledcallsindetailtoidentifyspecificcallsthatareidealcandidatesforautomation.

3. Develop the business case for the project by doing a cost benefit analysis. Compare theinvestmentsneededontheprojectagainstthecostsavingsandrevenueenhancementsthatyou expect. While factoring the costs of the system, you must also include the cost ofintegratingthespeechapplicationwiththebackendcustomerdatabase,inordertobeabletoderivethemaximumbenefits.Includeintangiblebenefitssuchasbettercustomerserviceandaloweringofcalldroprates.

4. Developaprojectplantorolloutthespeechapplicationinphases.Identifythecallsthatareeasiest to automate such as address changes and balance inquiries for the initial phase.Transactionsthataremorecomplexcanbesetasideforalaterphase.

4.6 EvaluatingServiceProviders

Oneofthekeyareasoftheplanningprocessistoidentifytherightvendor.Inthissection,wewilllookatsomeofthethingsitishelpfultoconsiderwhileevaluatingpotentialvendors.

1. TechnicalExpertise:Askyourpotentialvendor fora ‘dialoguespecification’documentandtheprofileofthepeoplewhowouldbeworkingonyourprojecttojudgethedialoguedesign

11


experienceofyourvendor.AskforspecificexperienceintheareaofCTIintegrationforcallcentersaswell.

2. DataAnalysis:Askyourvendorhowtheyplantocollectandanalyzecalldata,asthisisanintegralpartofdesigningarobustsystem.

3. ImplementationandTraining:Askyourvendortosharetheimplementationplanwithyou.Also,askaboutwhattrainingwillbeofferedtoyourstaffandthecostsassociatedwithit.

4. References:Request referencesofprevious implementations so that you can talk tootherpeoplewho have gone through the implementations and learn about the challenges thattheyfacedintheprocess.

4.7 NewOpportunitiesforSpeechApplications

Althoughusingspeechapplicationstofurtherautomateexistingservicescanyieldsignificantcostsavings,todayvendorsaredesigningnewapplicationsthatcanautomatecomplextransactionsandprovide better operational efficiencies. The transactions that best lend themselves to speechautomation are those that have high volumes and are highly repetitive in nature. These includeaddress change requests, requests for product information, and requests for switching of tariffplans. Caller identification is another area where speech automation can be used. It can enrollcustomers with a voiceprint to enable an added layer of security, as well as enhance customerservicebyavoidingtheneedforcallerstorepeathardtorememberaccountdataeverytimeacallismade.Yetanotherareawherespeechapplicationscanbeusedistoconductcustomersatisfactionsurveysfollowingcustomers’calls.

Someofthekeyadvantagesofimplementingspeechapplicationsaregivenbelow:

Reduce customer wait time by as much as 35%Better customer segmentation andconsequentroutingofcalls,resultinginhigherfirst‐callresolutionrates

Highercallcompletionrates

Abilitytohandlehighercallvolumes

Enhancedcustomersatisfaction

Abilitytooffer24/7service

Integrating the speech applicationwith customer relationshipmanagement software has severalbenefits. The speech recognition system can be configured to greet callers by their name andcommunicateinthelanguageoftheirchoice.Inaddition,thespeechapplicationcanalsokeeptrackof the caller’s preferences in terms of preferred products and services, nature and frequency oftransactions.

12


Recent surveys indicate that most callers enjoy being greeted by name and are happy that thesystemcouldremembertheirlanguagepreferenceandthelasttransactiondetails.

Successful deployment of a speech recognition application has always been a challenge as themajorityoftheworkinvolves‘tuning’theapplicationforyourspecificcallerprofiles.Thisincludesupgradingthevocabularyofthespeechenginewithkeywordsandphrasesmostcommonlyusedby your callers. This is not a one‐off process and must be repeated frequently to maintain theapplication’slevelofaccuracy.Inturn,thismeansthatyouwouldbespendingtimeandeffortonanongoingbasiswiththeconsultantswhodesignandinstallyourspeechapplication.However,youcandoseveralthingstoreducethetime,effort,andcosts involvedindeployment,someofwhicharelistedbelow:

Implementationofpackagedapplications

Expandedgrammarsetsincludingindustryspecificjargons

Improving easeof useof design and tuning tools. For example,Microsoft has adapted itsVisualStudioWebdevelopmenttooltobeusedwithitsSpeechServer,thusmakingiteasyfordeveloperswithVisualStudioexpertisetodesignandtunespeechapplications.

In fact, today’s software is able to perform many functions that previously required expensivehardware. Speech recognition applications deployed over a network not only enable moreautomation, but also reduce the hardware required, and even support business continuity byautomaticallydistributinglicensesandworkloadfromafailedservertoabackupserver.

4.8 IdentifyingtheVendor

Speech recognition technology can be used to reduce costs and enhance revenue. While somecompaniesusespeechapplicationsentirelyforupsellattempts,othersusethetechnologytogaugea caller’s interest for a new product or service before routing the call to a live agent for salesclosure.

Some vendors also offer value‐added services such as business analytics tomine automatic calldetailsforbusinessintelligencethatcanleadtoincreasedrevenues.

Themostpopularvendorswhooffersoftware,hardware,andconsultingservicesforimplementingspeechrecognitionincallcentersarelistedbelowinalphabeticalorder:

1. AspectSoftware

2. Avaya

3. Edify

4. Empirix

13


5. EnterpriseIntegrationGroup(EIG)

6. Genesys

7. IBM

8. Intervoice

9. LumenVox

10. Microsoft

11. Nuance

12. SterlingAudits

13. Syntellect

14. TuVox

15. UnveilTechnologies

16. Voxify

Inordertoselecttherightspeechapplicationforyourneeds,keepthefollowingaspectsinmind:

TechnologyStandards:ProprietaryversusStandard

Oneofthekeychangesthatthespeechrecognitionindustryhaswitnessedinrecentyearsisamovetowards open standards and convergencewith the other IT assets in the call center. Unlike IVRtechnologies,whichlargelyexistedinsilosandmadeintegrationwithbackenddatabasesdifficult,speechapplicationsaredesignedtoenableeasycommunicationwith theother ITapplications inthecallcenter.Itisimportantthereforetoselectlanguageandplatform‐independentapplicationsso that your current IT equipment and infrastructure need not be changed to accommodate theapplication.

DevelopmentOptions:PackagedVersusCustom

In the initial ages of speech recognition application, each application was custommade for thespecific requirements of a call center. This resulted in high cost solutions that needed a longimplementation time and significant effort for maintenance. Today, there are tried and truepackaged applications that can be easily customized to result in low risk, low cost and quickdeployments. Thus, unless you are in a niche industry with a large set of jargon and industryspecificneeds,itisalwaysbettertostartfromapackagedapplication.

14


DeploymentOptions:HostedVersusCustomerPremisesEquipment(CPE)

IfyoudonothaveastrongITteamoryouhavebudgetaryconstraints,thenahostedsolutioncouldbeagoodalternative.However,CPEsolutionswillofferyoumorecontrolastheycanaccommodatehighercallvolumeswhenyourbusinessexpands.

VendorExpertise

Before you finalize the vendor, complete your due diligence and choose a vendorwith a strongrecordofaccomplishmentthatcanofferyoucustomerreferences.Youmustideallyselectavendorthatcangiveyouthesoftware,equipment,andconsultancyservices foryourproject.ThismeansthatthevendorshouldbeabletodothefeasibilityandscopingphasesofyourprojectaswellastheimplementationphasewhereyouwillbeofferedassistanceintheareasofVUIdesign,tuning,andtraining.Besuretorequestforpostimplementationsupportintheformofperformanceauditsandongoingtuningsupport.

Flexibility

Ultimately,theapplicationshouldalsoofferenoughflexibilitytobetweakedtocatertochangesinyour business environment and customer profile. Thus, the system should allow the call centerpersonneltoeasilymakechangesthroughauserinterfaceandsystemadminconsole.

The main objectives of using speech application software are to offer a service that callers arehappy about, that is able to achievemeasurable results, andoffers a good returnon investment.Most call centers arenowable to realize anROI from speech technologywithin the first year ofimplementation. However, to reap the maximum benefits, you must go through the stages ofdiscovery,design,implementation,andoptimization.

1. Discovery:Inthisstage,youwouldprepareabusinesscaseforspeechapplicationsinyourcall center. The first step is to define your business, your caller, the IT system that youcurrentlyhave,andthewaysinwhichspeechapplicationscanimproveeachofthese.

2. Design:Inthisphase,youwilldesigntheuserinterfaceforthespeechapplication.Unlessitis easy to use, callerswill not use your speech application for self‐service.Make sure toreinforceyourcompany’sbrandthroughyourVUI.YoucanevenpersonalizeyourVUIwithanagentprofileandaname.

3. Implementation: In this phase, the actual speech application is customized, tested, anddeployedinyourcallcenter.

4. Optimization:Thisisakeyphasethatisoftenoverlookedbycallcenters.Inordertokeepyourapplicationintunewiththechangingcallerbehavioraswellasthechangingnatureofyourbusiness,ongoingoptimizationisessential.

15


5 RoadAheadforSpeechRecognitionasaViableOptionforCustomerService

Inrecenttimes,callcentershavereallygivenathumbs‐downtospeechapplications.AccordingtoanIVRSurveycarriedoutbytheCallCentreHelperwebsiteinFebruary2012,onlyanaverageof10%ofcall centersusespeechrecognition.Asexpected, the larger thecallcenter, thehigher theadoptionof speechrecognition.Only6%ofsmallercall centersusespeechapplications,whereasthefigureis7%formediumsizecallcentersand23%forlargecallcenters.Theadoptionofspeechrecognitionhasalsovariedwidelybetween industries.Asurveyconducted in2009revealedthatwhiletelecomandutilitiesindustrieshavealmost33%adoptionrates,otherssuchasretailandIThave not opted for speech recognition at all. The average adoption rate hovered around 8% forspeech recognition in 2009 versus 10% in 2012. This shows that there has been a shift fromTouchtoneIVRtovoicerecognition,althoughataverysmallpace.

Themainproblemwithspeechapplications is itsaccuracy,especiallywhen it comes tocorrectlyinterpretingregionalaccents.Mostsystemswhichclaimtoofferashighas90%accuracyperformmuch below expectations in real life and offer only up to 70% accuracy. However, surveyrespondents also agree on the importance of careful tuning to improve accuracy rates,which isfoundtoimproveupto90%iftunedproperly.Thefactremainsthatthereisawidegapbetweenrecognitionrates fora limitedvocabulary inacontrolledenvironmentandthoseobtained inreallifesituationswitha lotofbackgroundnoise,vocabularyvariationsandaccentvariations.Wordsthatarenotpartofthespeechengine’svocabulary,knownas‘outofgrammar’errors,contributetoasmuchas15%oftheproblems.

That being said, this doesnot signal the endof speech recognition technologybeing used in callcenters.Itsimplementationwillundergochangesinthefuture.Somepotentialchangeswouldbeinthefollowingareas:

1. PlaybackInformation:Customerswhowantfastaccesstoinformationandareinahurrydonotalwayswanttospeaktoaliveagent.Theywouldratherhaveashorterwaittimeandgettheinformationthattheywantmorerapidly.ThishasbeensuccessfullyusedintheDublinAirport,wherea30%increase inpassengernumbershasbeenhandledwithout theneedfor additional agents. The incoming calls are filtered according to the requirements, andcalls that require info on ‘departures’ or arrivals’ are directed to a speech recognitionsystem,handling the requirements effectivelywithouthaving to transfer the call toa liveagent.The systemhasbeen fine‐tuned to recognize the Irish accent,whichhas improvedaccuracyrates.Theaveragecalltimeforthesecallsisjust53seconds,thusfreeingupagenttimeformorecomplexcalls.

2. CallRouting: Callersno longerwant to speak to aperson,butwant to speak to the ‘rightperson’. This iswhere speech recognition applications can play amajor role by allowingcallers to ‘say’what theywant so that they can be routed to the right operator, therebyleadingtoimprovedfirstcallresolutionrates.

16


3. CallerAuthentication:With increased instances of identity fraud being reported from callcenters,voicebiometricsmayprovidetheanswerforcallerauthenticationwithouthavingtoaskforpersonaldata.Creatingavoiceprint isrelativelyeasyand it takes lessthantwominutesonaveragetocreateone.Thisisthenstoredagainstthecustomer’srecordandthenext timehe calls, thevoice ismatchedagainst thevoiceprint stored in thedatabaseanddirectly routed to anagent.This feature cutsdowncall timespendondetailed IDchecksusingpasswords,address,accountdetails,andsoon.

4. ReplacingComplicatedIVRMenus:Speechrecognitionapplicationswillfindusesintheareaof ‘intelligentcallsteering’byaskingcustomerswhattheywant insteadofaskingthemtonavigatethroughacomplicated‘pushbutton’menu.

5. DealingwithSpikesinCallVolumes:Inindustriessuchasbettingwheremostcallsoccurjustbefore a race, speech recognition applications can help in handling sudden spurts in callvolumes.

6. CustomerSurveys: After an agent has handled a call, a speech application can be used toconduct customer satisfaction surveys. This makes it more comfortable for the callerespecially in instanceswhere the agent is receivingnegative feedback. In addition, itwillalsofreeupexpensiveagenttime.

Some of the key reasonswhy speech applications have not been successful in certain cases arelistedbelow:

DifficulttoUse:Ifcallersareunabletocompletethetransactionthroughavoiceapplication,then they will surely opt for a live agent. Another factor is the length of the call. Ifcompletingthetransactiontakeslongeroverthevoiceapplicationthanthroughtheagent,thecallerswillpreferanagenttotheapplication.

BadDesign:AbadlydesignedVUIisoneofthemainreasonswhycallersmayoptout.BadlydesignedmenustructureswerethebiggestproblemwithIVRsystems,andthisistrueforvoicesystemsaswell.Insteadofhavingmultiplelayers,theapplicationmustbeconfiguredto ask open‐ended questions such as “Whatwould you like to do today?” and be able tointerpretallpossibleanswers.Youmayalso includesupplementaryquestions to improvethe interpretationaccuracyofyour speechengine.Forexample, insteadofasking for justthepostalcode,thespeechsystemcanbeconfiguredtoaskforbothpostalcodeaswellasthestreetnamesothatthesystemisabletoaccurately interprettheaddressbygettingauniquematchinalargersetofsituations.

LackofCallerPrompts: Callersmust be prompted to provide information in a consistentmanner.Forexample,ifyouhavetocaptureanilvalue,thenthecallermaysay“nil”,“no”,“nothing”,“zero”andsoon.Insuchacase,insteadoftrainingthespeechenginetointerpretall possibilities, it is better to train the callerwith aprompt such as, “If the amount is 0,pleasesayzero”.

17


Inordertoimprovethecustomerexperience,thefollowingaspectsshouldbeincorporatedinyourspeechapplication:

1. Havehelpmessagesatthestartofthecall:Callersshouldhaveanoptiontorequestthehelpmenuatanytimeduringthecallbysaying,“help”.Youcanhaveapromptatthebeginningofthecallthatsays,“Atanypointintime,pleasesay,‘Help’,ifyouneedmoreinformation”.

2. Ensure that the prompts are easy to understand: Instead of saying, “State your accountnumber”, say, “State your account number that is in the upper right corner of your bankstatement.Ifyoudonothaveyouraccountnumber,stateyourtelephonenumber”.

3. Have sufficient time lag: Give users ample time to respond to queries and provide theinformationthatyouareaskingfor,beforereplayingtheprompt.

4. Avoid endless loops: Placing callers in an endless loop of having to repeat a request orinformationmultipletimesisbothersome.Ifyourspeechenginefailstorecognizetheinputaftertwotries,transferthecalltoaliveagent.Makesurethatalltheinformationcapturedduringthecallisavailabletotheliveagentandcallersarenotrequiredtorepeateverythingfromthestart.Youcanalwaysanalyzethecalllatertotweakyourspeechengine.

18


6 Conclusion

Whileself‐serviceapplicationsaren’talwaystherightchoiceforeverybusinessorineverysituation,advancedspeechapplicationscanhelpcompaniesachievethedelicatebalancebetweengivingcustomersasatisfyingexperiencewhilesuccessfullymanagingcosts.Easeofuse,therefore,becomesthemostvaluablewayofdifferentiatingyourbusinessfromcompetitors.Voicetechnologyhasreachedthestagewhereonlysmallimprovementsinsystemspeedcanbemade.Yetthereisplentyofscopeforimprovingtheuserexperience–howevercomplexormissioncriticaltherequirement.

19


7 References

1. "Speechrecognitionenhancescall‐centercapabilities",2000,CommunicationsNews,vol.37,no.11,pp.44‐44.

2. Barbetta, F. 1996, "Unisys deliver speech recognition for call centers", BusinessCommunicationsReview,vol.26,no.12,pp.60‐60.

3. Maselli,J.2001,"SpeechrecognitioncutsAmtrak’scall‐centercosts",InformationWeek,,no.866,pp.24‐24.

4. Zager,M. 2005, "SpeechRecognition GoesMainstream",CallCenterMagazine,vol. 18, no.10,pp.20‐24,26,28‐30.

5. Sherter,A.1999,"Speechrecognitionspeaksvolumes",BankTechnologyNews,vol.12,no.8,pp.1‐1,33+.

6. VoiceGenieandBBNTechnologies toMarketSpeech‐RecognitionProduct forLargeBusinessCallCenters2002,,NewYork,UnitedStates,NewYork.

7. ALTechandAmarexPartnerToDeliverSpeechRecognitionForNetwork‐BasedCallCenters1998,,NewYork,UnitedStates,NewYork.

8. "SpeechRecognition",1999,CommunicationsToday,vol.5,no.145,pp.1‐1.

9. Kolor,Joanna.(Oct1997)."Speechrecognitiongetsserious"BankTechnologyNews,1,43+.

10. Call Centers: Speech Recognition Centers Reduce Call Time by 35% PR Newswire [NewYork]16Aug2005:1.

11. Pollock,B.&Calabro,T.2002, "Ready forspeechrecognition?",PublicUtilitiesFortnightly,vol.1,no.1,pp.44‐44.

what is speech recognition - specialty answering … speech recognition technology has advanced to...

Documents