Speech Processing 15-492/18-492Speech ProcessingCurrent Topics and Future challengesCommercial and ResearchCurrent and FutureWhat are the hot topics in SpeechWhat are the hot topics in SpeechWhat currently worksWhat currently worksWhat could work soon (5What could work soon (5--10years)10years)What are the industry hot topicsWhat are the industry hot topicsWhat are the research challengesWhat are the research challengesSpoken Dialog: NowIndustry:Industry:Location based queryingLocation based queryingGoogleGoogle: 411, : 411, smartphonesmartphoneMicrosoft Live Search: Microsoft Live Search: smartphonesmartphoneYahoo (Yahoo (VlingoVlingo))Blackberry, Blackberry, IPhoneIPhone(Owners have money)(Owners have money)How do you make money out of this …How do you make money out of this …Spoken Dialog: NowResearchResearchError recoveryError recoveryAdaptive systemsAdaptive systemsRapid deploymentRapid deploymentLearning dialog structure from dataLearning dialog structure from dataASR: NowIndustryIndustryMoving from grammar based to NMoving from grammar based to N--gram basedgram basedBroadcast news transcription of IRBroadcast news transcription of IRRobust speech recognition:Robust speech recognition:In car, outside, in noisy officeIn car, outside, in noisy officeLM adaptation from other sourcesLM adaptation from other sourcesUsing click through and search queriesUsing click through and search queriesPronunciation variants (“wrong” ones too)Pronunciation variants (“wrong” ones too)Medical transcriptionMedical transcriptionASR: NowResearch:Research:Discriminative trainingDiscriminative trainingAcoustic parameter projections to discriminate Acoustic parameter projections to discriminate between the correct answers and competitorsbetween the correct answers and competitorsRobust recognitionRobust recognitionFar field microphonesFar field microphonesBlind source separationBlind source separationOut of vocabulary wordsOut of vocabulary wordsUnsupervised trainingUnsupervised trainingTTS: NowIndustryIndustryBuilding custom voices (and your voice)Building custom voices (and your voice)Multilingual on small devicesMultilingual on small devicesE.g. for GPS Navigation over EuropeE.g. for GPS Navigation over EuropeEasy methods to build new languagesEasy methods to build new languagesTTS: NowResearchResearchImproving statistical synthesisImproving statistical synthesisRapid support in new languagesRapid support in new languagesEmotional speech synthesisEmotional speech synthesisAutomatic building of voices from dataAutomatic building of voices from dataWithout any human interventionWithout any human interventionSynthesis beyond the sentenceSynthesis beyond the sentenceSynthesis with more text analysisSynthesis with more text analysisSpeech to Speech TranslationIndustryIndustryOne way systems, domain limited systemsOne way systems, domain limited systemsSimple targeted cell phone systemsSimple targeted cell phone systemsResearchResearchTwo way systems, large domainsTwo way systems, large domainsOne way lecture/broadcast newsOne way lecture/broadcast newsVC and SID: NowVoice conversionVoice conversionCross Lingual Voice ConversionCross Lingual Voice ConversionEmotion/style conversionEmotion/style conversionConversion without training dataConversion without training dataSpeaker IDSpeaker IDAccuracy on large data sets (> 1000 speakers)Accuracy on large data sets (> 1000 speakers)Cross channel/language IDCross channel/language IDMore information in ID (prosody, More information in ID (prosody, vocabvocab))CALL: NowIndustryIndustryPronunciation trainingPronunciation trainingScenario practicingScenario practicingResearchResearchGame based toolsGame based toolsMeasuring educational contributionMeasuring educational contributionSpeech Processing FutureHard challenges (PhD topics and beyond)Hard challenges (PhD topics and beyond)All on the research sideAll on the research sideBut maybe in Research LabsBut maybe in Research LabsSpeech Reco without SpeechUsing other modalitiesUsing other modalitiesLip movement, muscle movementLip movement, muscle movementSilent speechSilent speechNo generated audioNo generated audioJust think about the wordsJust think about the wordsGesture recognitionGesture recognitionConversational SystemsParticipant in a meetingParticipant in a meetingTrue conversational speechTrue conversational speechAppropriate nonAppropriate non--word speech generationword speech generationKnow when to speak, when to laugh, when to listenKnow when to speak, when to laugh, when to listenAppropriate timing conversationAppropriate timing conversationAble to interrupt when having something to sayAble to interrupt when having something to sayHave something to sayHave something to saySummaries and DiscussionsDescribe a paper/movie/eventDescribe a paper/movie/eventAppropriate summaryAppropriate summaryAllow questionsAllow questionsKnow when to use style/emotionKnow when to use style/emotionNot just speech<Not just speech<-->text >text Understand more of the text contentUnderstand more of the text contentFinal NotesDon’t forget to fill in Faculty Course Don’t forget to fill in Faculty Course EvaluationEvaluationFinal Homework dueFinal Homework dueMonday 8Monday 8thth3:30pm3:30pmFinal examFinal examTuesday 16Tuesday 16thth1pm1pm--4pm 4pm