Speech Processing 15-492/18-492Spoken Dialog SystemsCase-study: Personal Digital AssistantsSpeech-based Personal Digital AssistantBuild a speech enabled PDABuild a speech enabled PDASpeech in/out for individual useSpeech in/out for individual useGoalsGoalsControl scheduleControl scheduleControl messagingControl messagingReplace personal assistantReplace personal assistantAny similarity to any existing product is purely Any similarity to any existing product is purely coincidentalcoincidentalDisclaimer: Much of this is relevant to Apple’s Disclaimer: Much of this is relevant to Apple’s SiriSiri, but this information is , but this information is general and may or may not be what is in general and may or may not be what is in SiriSiri..SPDA:ScopeScheduleScheduleCalls (in and out?)Calls (in and out?)NavigationNavigationFinding local businessesFinding local businessesWith reviewsWith reviewsOpen questionsOpen questionsReminders/AlarmsReminders/AlarmsSPDA: Scope“Call John”“Call John”“Call John, Bill and Mary and setup a meeting “Call John, Bill and Mary and setup a meeting sometime next week about Plan B that’s fits my sometime next week about Plan B that’s fits my schedule”schedule”“Make a reservation at a local Chinese restaurant “Make a reservation at a local Chinese restaurant for 4 at 8pm.”for 4 at 8pm.”“You should call your mom as its her birthday”“You should call your mom as its her birthday”“I have sent flowers to your mom as its her “I have sent flowers to your mom as its her birthday”birthday”CALO (DARPA)Cognitive Assistant that Learns OnlineCognitive Assistant that Learns OnlineDARPA project (2003DARPA project (2003--2008)2008)Led by SRI (involved many sites, including CMU)Led by SRI (involved many sites, including CMU)Personal Assistant that Learns (Pal)Personal Assistant that Learns (Pal)Answers questionsAnswers questionsLearn from experienceLearn from experienceTake initiativeTake initiativeSpinSpin--off company off company --> SIRI> SIRIAquired by Apple in April 2010Aquired by Apple in April 2010SPDA: PlatformDesktopDesktopComputational powerComputational powerPhone (nonPhone (non--smartphonesmartphone))General Magic General Magic Was handheld, became phone basedWas handheld, became phone basedLed into GM’s Led into GM’s OnStarOnStarSmartphoneSmartphoneLocal to deviceLocal to deviceWith CloudWith CloudSmartphone + CloudSmartphoneSmartphoneKnow about userKnow about userContacts, Schedule etcContacts, Schedule etcSame speaker Same speaker Some computation possible on deviceSome computation possible on deviceCloudCloudLearn from multiple examplesLearn from multiple examplesRetrain acoustic/language/understanding Retrain acoustic/language/understanding modelsmodelsVoice Search and User FeedbackVoice SearchVoice SearchGoogleGoogle, Bing, , Bing, VlingoVlingo, Apple, AppleGet users to help label the dataGet users to help label the dataListen to userListen to userShow best optionsShow best optionsThey select which on is correctThey select which on is correctFind out how users actually speakFind out how users actually speakFull sentences Full sentences vsvs“search terms”“search terms”How do English speakers say ethnic namesHow do English speakers say ethnic namesVoice Search: SimplificationsToo many words …Too many words …ContextContextWhere you are (location: home/not home)Where you are (location: home/not home)What is on your phone (contacts)What is on your phone (contacts)What you’ve said beforeWhat you’ve said beforePersonalityHave a characterHave a characterCalls you by name (you choose)Calls you by name (you choose)Pushy, helpful, nagging …Pushy, helpful, nagging …Allow user choiceAllow user choicePersonalize itPersonalize itMay form better relationship with itMay form better relationship with ite.g. e.g. SiriSiriUS and UK are female/maleUS and UK are female/maleMake it do things wellTargeted appsTargeted appsChose what it will do wellChose what it will do wellSay, 12 different appsSay, 12 different appsHave target (hand written) interactionHave target (hand written) interactionChose what fields you need, and how to Chose what fields you need, and how to intereactintereactwith with the back end datathe back end dataIf all else fails dump result in If all else fails dump result in GoogleGoogleHardware aidHardware aidInfraInfra--red detector for VADred detector for VADMarketingMake sure people know its thereMake sure people know its there(Voice search has been on PDA’s for years)(Voice search has been on PDA’s for years)Get a *lot* of people to use itGet a *lot* of people to use itGive “silly” examplesGive “silly” examplesPeople will repeat them, you can adapt your system People will repeat them, you can adapt your system and expect them to say themand expect them to say themKnow Your UsersYoung educatedYoung educatedStandard English speakersStandard English speakers(Non(Non--native too?)native too?)Can you train them to use it betterCan you train them to use it betterGet them to adaptGet them to adaptWhat is Missing?Add an SDKAdd an SDKOther app developers will want to allow speechOther app developers will want to allow speechMay make it harder to distinguishMay make it harder to distinguishDialog contextDialog contextWhat was said in the previous utteranceWhat was said in the previous utteranceOthers …Others …Will it work?Will people talk in publicWill people talk in publicTalking on the phone is now acceptableTalking on the phone is now acceptableTalking to the phone … Talking to the phone … Will people continue to use itWill people continue to use itCool at first, but easier to use menusCool at first, but easier to use menusOnly use for setting alarmsOnly use for setting alarmsLong term use …Long term use …But others may join in anywayBut others may join in
View Full Document