Home
>
Lucent Technologies
>
Communications System
>
Lucent Technologies INTUITY CONVERSANT System Version 6.0 Guide
Lucent Technologies INTUITY CONVERSANT System Version 6.0 Guide
Have a look at the manual Lucent Technologies INTUITY CONVERSANT System Version 6.0 Guide online for free. It’s possible to download the document as PDF or print. UserManuals.tech offer 413 Lucent Technologies manuals and user’s guides for free. Share the user manual or guide on Facebook, Twitter or Google+.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-7 Speech Recognition 2 FlexWord sp eec h rec og nition and WholeWord sp eec h rec og nition c an b e used tog ether for g reater flexib ility in your ap p lic ations. You c an d efine a Promp t & Collec t ac tion to use WholeWord sp eec h rec og nition If you want c allers to say “ yes” or “ no,” a series of d ig its, or sing le d igits. If you want c allers to say a word or phrase from your voc ab ulary, you would d efine the Promp t & Collec t to use FlexWord sp eec h rec og nition and sp ec ify the word list. FlexWord Speech Recognition Capabilities This sec tion d esc rib es the FlexWord sp eec h rec og nition uniq ue c ap ab ilities. FlexWord Speech Recognition Languages The FlexWord Sp eec h Rec og nition software is used to rec og nize sp oken inp ut for the following lang uag es: nFren c h nGerman nJap anese nSp a n is h nUS English Only one FlexWord sp eec h rec og nition lang uag e c an b e installed on your system at a time. FlexWord Speech Recognition Vocabularies As mentioned b efore, FlexWord sp eec h rec og nition uses a c ustom, tailored voc ab ulary that is sp ec ific to eac h ap p lic ation. FlexWord sp eec h rec og nition voc ab ulary items c an b e sing le word s or p hrases. Eac h item in a voc ab ulary is known as a word, even if the item is a p hrase of several word s. For examp le, in a FlexWord sp eec h rec og nition voc ab ulary, “ c hec king ” and “ mutual fund ” are b oth c onsid ered word s, even though “mutual fund ” is a p hrase. The words in a vocabulary must be divided into groups called wordlists. A word list inc lud es all the word s that c an b e sp oken at a p artic ular p rompt. See Tab le 2-1 for an examp le of how word s are g roup ed into word lists to form a c omp lete vocab ulary. Eac h word list c an c ontain up to 2000 word s, and you c an have up to 200 word lists. However, the total numb er of word s c annot exc eed 5000. It is p ossib le for ap p lic ations to share wordlists.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-8 Speech Recognition 2 There are two ways to c reate your voc ab ularies: nUse the FlexWord Toolkit software. The FlexWord Toolkit software allows you to b uild your own voc ab ularies q uic kly and easily. This op tional software p ac kag e g ives you a p oint-and -c lic k, g rap hic al environment in whic h to ad d , d elete, or c hang e word s on new or existing word lists. You c an then use these word lists in your ap p lic ation as soon as you c reate them. nUse Luc ent Tec hnolog ies c ustom voc ab ulary servic e. The c ustom voc ab ulary servic e is also an availab le op tion to build your voc ab ularies, for a fee. Consid er the size of your voc ab ularies and how many FlexWord ap plic ations you p lan to have when d ec id ing to p urc hase the FlexWord Toolkit or use the c ustom voc ab ulary service. 1. This samp le voc ab ulary c onsists of 53 word s d ivid ed into 8 word lists. Table 2-1. Sample Vocabulary1 Wordlist 1 (11 words)Wo r d l i s t 2 (6 words)Wo r d l i s t 3 (6 words)Wo r d l i s t 4 (8 words) skirt b louse dress pants shorts sandals shoes soc ks belt help attendantsmall medium larg e extra large help attendantpetite small medium larg e help attend antsky b lue c ranb erry brown natural teal g reen rust help attend ant Wordlist 5 (6 words)Wo r d l i s t 6 (5 words)Wo r d l i s t 7 (5 words)Wo r d l i s t 8 (6 words) black brown lig ht tan olive help attendantwomen’s men’s child’s help attendantnarrow medium wid e help attend antvisa master card americ an exp ress discover help attend ant
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-9 Speech Recognition 2 FlexWord Speech Recognition and Key Word Spotting FlexWord sp eec h rec og nition sup p orts key word sp otting . As exp lained ab ove in “ WholeWord Sp eec h Rec og nition and Key Word Sp otting ,” key word spotting is the ability of the rec og nizer to isolate a key word out of a sequenc e of other word s or noises. As with WholeWord key word sp otting , the c aller d oes not have to say the key word b y itself to b e rec og nized . For example, if the c aller says “ I want my c hec king ac c ount,” the rec og nizer c an isolate the key word “ c hec king .” Rememb er, thoug h, that the rec og nizer find s a key word most ac c urately when the word is said all b y itself. FlexWord Speech Recognition and Barge-in Unlike WholeWord sp eec h rec ognition, FlexWord sp eec h rec og nition d oes not sup p ort b arg e-in. Callers must wait until the end of a p romp t to b eg in sp eaking their resp onse. FlexWord Speech Recognition Accuracy Like WholeWord speec h rec og nition, FlexWord sp eec h rec og nition is most suc c essful when ap p lic ations first offer c allers a c hanc e to ind ic ate that they have a touc h-tone telep hone (usually b y p ressing one on the keyp ad ). If no tone is d etec ted , the app lic ation p romp ts the c aller to respond with sp oken inp ut (instead of transferring the c all to an attend ant). In this manner, c allers who want to p rovide spoken inp ut c an be served b y the system, instead of req uiring an attendant. You should c ontinue to p rovid e some attend ants for your c allers to maintain c ustomer satisfac tion. As with WholeWord sp eec h rec og nition, b e c areful about always allowing sp oken inp ut. If your c allers may often b e sp eaking from a noisy environment like an automob ile or an airp ort, touc h-tone inp ut will b e faster and more ac c urate.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-10 Speech Recognition 2 Speech Recognition Accuracy Measurement The c onc ep t of rec og nition ac c urac y is similar for b oth WholeWord and FlexWord sp eec h rec og nition. Sinc e WholeWord voc abulary is fixed at only a few items in eac h lang uag e, we c an measure and then exp ec t ac c urac y rates. However, sinc e FlexWord voc ab ularies are c omp letely c ustomized for eac h app lic ation, there is no ab solute ac c urac y rate. For FlexWord sp eec h rec og nition, ac c urac y c an only b e measured on eac h ap p lic ation, b ec ause eac h ap p lic ation will have a uniq ue set of words and p hrases that are g roup ed into wordlists. The rec og nition ac c urac y of eac h sep arate word list c an b e measured with some effort. Rec og nition accuracy assumes that c allers are saying something the recognizer c an rec og nize. When a c aller g ives an invalid inp ut, the rec og nizer should rejec t it. Ac c urac y is measured over many speakers, d ifferent typ es of ap p lic ations, various environmental c onditions, and reg ional ac c ents. The average WholeWord ac c urac y rate for a sp eaker saying a sing le d ig it is 97%. A 97% ac c urac y rate means that for a larg e numb er of different typ es of c allers over a long time and und er d ifferent c onditions, 97% of valid inputs will be c orrec tly rec og nized. It d oes not mean that every c aller sp eaking the d ig it will b e c orrec tly rec og nized 97% of the time. Nor d oes it mean that 97% of p eop le who c all in will b e rec og nized c orrec tly. Ac c urac y d ec reases slig htly as dig it sequenc es g et long er, b ec ause of the d iffic ulty to d etec t b ound aries b etween d ig its. When two or more d ig its are sp oken in a row (know as c onnec ted d ig its), the ac c urac y rate is 0.97n, where n is the number of d ig its that the rec og nizer exp ec ts c allers to say. For examp le, a three-d ig it seq uenc e is exp ec ted to have an ac c urac y rate of 0.97 x 3, or 91%. In g eneral, you c an sub trac t ab out three p erc entag e points for eac h ad ditional d ig it sp oken in a row. Improving Speech Recognition Accuracy The g uid elines in Chap ter 4, ‘‘Desig ning a Voic e Resp onse Ap p lic ation,’’ will help you op timize the c aller-ap p lic ation interfac e d esig n, make c hoic es on whether to use custom grammars, create affective FlexWord wordlists, and inform callers ab out what to say.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-11 Text-to-Speech 2 Text-to-Speech This advanc ed sp eec h tec hnolog y allows your app lic ation to c onvert ASCII text into sp oken word s that are said to the c aller; simp ly p ut, it is a read ing machine. Text-to-Speech Uses Text-to-Sp eec h is esp ec ially useful when it is not prac tic al to p rerec ord information to be spoken, for examp le, if the information is leng thy, c hang es freq uently, and /or c omes from a d atab ase. Information you c ould c onsider sp eaking to your c allers with Text-to-Sp eec h inc lud es: nNews wire stories nRules and reg ulations nNames and telep hone numbers nTelephone d irec tory entries In most of these c ases, the c ost of having a p rofessional sp eaker p rerec ord all the information is p rohib itive. Text-to-Sp eec h is an ec onomic al way to p rovid e c allers with ac c ess to extensive or quic k-c hang ing information. Text-to-Sp eec h c an also b e used in p lac e of p rerec ord ed sp eec h when p romp ting c allers, thus saving you time when you d emonstrate your ap p lic ation for testing purposes. Text-to-Speech Capabilities This sec tion d esc rib es the Text-to-Sp eec h uniq ue c ap ab ilities. Text-To-Speech Languages Text-to-Sp eec h is sup p orted only for US Eng lish. Text-to-Sp eec h speaks word s in a male voic e with a US Eng lish ac c ent. Most c allers find the speec h easy to und erstand . Text-to-Speech and Abbreviations The Text-to-Sp eec h feature exp and s ab b reviations. If the text you want sp oken c ontains ab b reviations suc h as “ Ave.” and “ Mr.,” Text-to-Sp eec h c an sp eak them out as “ Avenue” and “ Mister.” Further, Text-to-Sp eec h also allows you to sp ec ify the typ e of information that is b eing sp oken. If you prep are Text-to-Sp eec h to read an ad d ress, it will read “ Dr.” as “ Drive,” b ut if you p rep are it to read a name, it will read “ Dr.” as “ Doc tor.”
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-12 Text-to-Speech 2 Text-to-Speech and Touch-Tone Recognition Text-to-Sp eec h is fully c omp atib le with touc h-tone rec og nition. An ap p lic ation d esig n c an allow c allers to p ress a touc h-tone b utton to d ial throug h a Text-to-Sp eec h p romp t or announc ement the same as if it were a p rerec orded p romp t or announc ement. Similarly, you c an also allow c allers to d ial ahead , and respond to several prompts in a row. Text-to-Speech and Speech Recognition Text-to-Sp eec h is also c omp atib le with WholeWord and /or FlexWord sp eec h recognition. An application can be designed so that callers can respond during a Text-to-Sp eec h p romp t or announc ement the same as if it were a p rerec ord ed p romp t or announc ement. However, when using Text-to-Speec h p romp ts with FlexWord sp eec h rec og nition, b arge-in is not availab le and the c aller must wait until the entire p romp t is c omp lete to sp eak. Text-to-Speech Accuracy Text-to-Sp eak assumes that the information it is read ing is struc tured in stand ard US Eng lish sentenc es (c omp lete with p unc tuation, c ap italization, sub jec t, ob jec t, and verb ). If the information you want sp oken is not written in c omp lete sentenc es (for example, a series of d atabase field s like name, ad d ress, and telep hone numb ers), the outp ut will not b e as und erstand able as full sentenc es. Se e ‘‘Text-to-Speec h in App lic ations’’ in Chap ter 4, ‘‘Designing a Voic e Resp onse Ap p lication,’’ for more information.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-13 Script Builder FAX Actions 2 Script Builder FAX Actions You c an p rovid e fax servic e to your c allers in your ap p lic ations with the Sc rip t Build er FAX Ac tions software p ac kag e. This software p rovid es you with the c ap ab ility to have your ap p lic ation send fax messag es, and rec eive fax messag es from c allers. FA X A ct io n s U s es You c an send g rap hic and textual information to c allers’ fax mac hines up on their req uest. Your ap p lic ations c an g ive c allers the op p ortunity to request information suc h as: nBlank forms (suc h as tax forms, ap p lic ations, or entry blanks) nCustomer servic e information nFinanc ial or med ic al rec ord s nReservation c onfirmations nAd vertising b rochures FAX Actions Capabilities The FAX Ac tions offer the following c ap abilities for fax app lic ations: nCustomize your c over pag es. nSend either p re-stored faxes or faxes d ynamic ally c reated d irec tly b y the ap plic ation. nSc hed ule fax messag es so that they are d elivered when telep hone rates are low. nRed ial the c aller’s fax mac hine if it is b usy. nCreate an ap p lic ation to exec ute a UnixWare shell c ommand to c reate text files. (Ap p lic ations c an fax anything that c an b e p ut into a text file, inc lud ing host sc reens and ORACLE datab ase rep orts. nRec eive faxes from c allers into the system. The FAX Ac tions c an b e used in the same ap p lic ation as touc h-tone input, DPR, WholeWord speec h rec og nition, FlexWord sp eec h rec og nition, and Text-to-Speech. This allows you many options for designing fax applications. For example, you c ould d esig n an ap p lic ation that allows physic ians ac c ess to hosp ital p atients’ med ic al rec ord s.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Voice Response Advanced Technologies Page 2-14 Script Builder FAX Actions 2 Onc e the p hysic ian d ials into the system and g ives a p ersonal id entific ation numb er (using either WholeWord sp eec h rec og nition or touc h-tone inp ut), he or she enters a p atient id entific ation numb er to either listen to med ic al rec ord s sp oken using Text-to-Sp eec h or to have the med ic al rec ord faxed to his or her office.
Planning a Voice Response Application Page 3-1 Overview 3 INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 3 3Planning a Voice Response Application Overview This c hap ter d esc rib es some g uid elines to keep in mind when p lanning the d esig n of your voic e resp onse ap p lic ations. These are g eneral g uid elines and are not sp ec ific to the ad vanc ed tec hnolog ies of the I NTUITY™ C O N VERSA N T® system. Sp ec ific guid elines are d esc rib ed in Chap ter 4, ‘‘Desig ning a Voic e Resp onse Ap p lic ation.’’ Purpose The p urp ose of this c hap ter is to p rovid e you with information to p lan your voic e response application designs.
INTUITY CONVERSANT System Version 6.0 Application Design Guidelines 585-310-670 Issue 1.0 December 1996 Planning a Voice Response Application Page 3-2 Plan Your Application Design 3 Plan Your Application Design This sec tion offers g uidelines to use as you p lan the d esig n of your voic e response ap p lic ations. Use Human Factors/Usability Engineering Resources Human fac tors and usab ility c onsultants are exp erts in the d esig n and testing of interfac es b etween p eop le and c omp uters. Some c onsultants sp ec ialize in telep hone-b ased user interfac es. If your c omp any d oes not have p eople trained in usab ility or human fac tors, ind ep end ent c onsultants are availab le to assist you in application design, testing, and deployment. Contrac ting p rofessional d esig n and usability testing servic es c an b e c ost effec tive as well as save you time and frustration. Customer satisfac tion c an also b e inc reased , sinc e you will b e more likely to offer your c ustomers a well-desig ned , extensively tested app lic ation. Luc ent Tec hnolog ies offers human fac tors d esig n c onsultation servic es on a Professional Servic es b asis. For more information, c ontac t your Luc ent Tec hnolog ies rep resentative. For a d irec tory of ind ivid uals and c omp anies offering ind ep end ent c onsulting in human fac tors and usability, c all the Human Fac tors and Ergonomic s Soc iety at (310) 394-1811. Use Prerecorded Speech Prerec ord ed p romp ts and announc ements sp oken b y a p rofessional speaker (suc h as a voic e-over artist or a d isc joc key) c an enhanc e the quality of your ap plic ation. It is usually ap p ropriate to use the same rec ord ed voic e throughout the app lic ation. An exc ep tion is when you have a g ood reason to try to d isting uish b etween d ifferent p arts of the ap plic ation. For instanc e, you may want to have “help” speech recorded by a male voice and the remainder of your p romp ts and announc ements in a female voic e. This c onsistenc y may help your c allers to know their loc ation within an ap p lic ation. This sec tion d esc rib es two methods to inc lud e p rofessional sp eec h in your ap plic ations.