Voice RecognitionOverviewSpeech Recognition Must…Two Categories of Voice RecognitionIsolated Word RecognitionConnected Word RecognitionAnalysis Of Discrete Voice RecognitionContinuous Voice RecognitionSlide 9Speaker Dependent SystemsSpeaker Independent SystemsComponents Of A Speech Recognition SystemVoice Interactive SystemApplications Of Voice RecognitionVoice VerificationDifficulties In Voice RecognitionFuture Of Voice RecognitionVoice RecognitionVoice RecognitionBy: Scott OrphanBy: Scott OrphanOverviewOverviewWhat is voice/speech recognition?What is voice/speech recognition?Types of recognitionTypes of recognitionAdvantages & disadvantages of different Advantages & disadvantages of different recognition approachesrecognition approachesApplications of voice recognitionApplications of voice recognitionDifficulties of recognitionDifficulties of recognitionFuture of voice recognitionFuture of voice recognitionDemonstrationDemonstrationSpeech Recognition Must…Speech Recognition Must…Identify the sound of a human voiceIdentify the sound of a human voiceUses the physics of soundUses the physics of soundFactor out environmental noise Factor out environmental noise Convert the acoustic signal to a stream of Convert the acoustic signal to a stream of wordswordsAccept messages as input for controlling Accept messages as input for controlling the systemthe systemTwo Categories of Voice Two Categories of Voice RecognitionRecognitionDiscrete speech recognitionDiscrete speech recognitionIsolated word and phrase recognitionIsolated word and phrase recognitionConnected word recognitionConnected word recognitionContinuous speech recognitionContinuous speech recognitionIsolated Word RecognitionIsolated Word RecognitionMost simple formMost simple formUses pattern matchingUses pattern matchingSingle words, separated by pausesSingle words, separated by pausesSpeech compared to list of word templatesSpeech compared to list of word templatesUsed by automated operator systemsUsed by automated operator systemsConnected Word RecognitionConnected Word RecognitionContinuation of isolated word recognitionContinuation of isolated word recognitionSystem “learns” fluid sequences of its System “learns” fluid sequences of its vocabulary wordsvocabulary wordsExamples: Credit card numbers Examples: Credit card numbers Telephone numbersTelephone numbersAnalysis Of Discrete Voice Analysis Of Discrete Voice RecognitionRecognitionSpeech not natural or easySpeech not natural or easySpecific commands (limited vocabulary)Specific commands (limited vocabulary)No grammatical or syntactic interpretationNo grammatical or syntactic interpretationRely only on phonological inputRely only on phonological input““Accept” vs. “Except”Accept” vs. “Except”Continuous Voice RecognitionContinuous Voice RecognitionA more complex systemA more complex systemAbility to speak in an everyday mannerAbility to speak in an everyday mannerTries to recognize and understand speechTries to recognize and understand speechNo specific or learned commandsNo specific or learned commandsMay use hidden Markov modeling, neural May use hidden Markov modeling, neural networks, dynamic time warpingnetworks, dynamic time warpingContinuous Voice RecognitionContinuous Voice RecognitionError proneError proneExpensive Expensive Requires a lot of computational powerRequires a lot of computational powerTwo types: Two types: speaker dependentspeaker dependentspeaker independentspeaker independentSpeaker Dependent SystemsSpeaker Dependent SystemsText read, voice & speech pattern analyzed Text read, voice & speech pattern analyzed Lacks flexibility, cannot be sharedLacks flexibility, cannot be sharedLess costlyLess costlyMore accurateMore accurateUsed by most commercial softwareUsed by most commercial softwareSpeaker Independent SystemsSpeaker Independent SystemsUnderstands multiple users of a certain Understands multiple users of a certain language typelanguage typeNo enrollment periodNo enrollment periodGreater flexibilityGreater flexibilityMore error prone & expensiveMore error prone & expensiveTends to be used for specialized, single-Tends to be used for specialized, single-task systemstask systemsComponents Of A Speech Components Of A Speech Recognition SystemRecognition SystemVoice Interactive SystemVoice Interactive SystemApplications Of Voice RecognitionApplications Of Voice Recognition$40 billion market:$40 billion market:Post office for speed mail deliveryPost office for speed mail deliveryWal-Mart warehouse facilitiesWal-Mart warehouse facilitiesWord processing / DictationWord processing / DictationViaVoiceViaVoiceVoice XpressVoice XpressSpeechworksSpeechworksVoice VerificationVoice VerificationMany more…Many more…Voice VerificationVoice VerificationDifficulties In Voice RecognitionDifficulties In Voice RecognitionUnpredictable errors, signal & acoustic Unpredictable errors, signal & acoustic variabilityvariabilityPhonetic variabilityPhonetic variabilityWithin speaker variabilityWithin speaker variabilityAcross speaker variabilityAcross speaker variabilityPeople’s fears and expectationsPeople’s fears and expectationsMulti-modal communicationMulti-modal communicationSpoken language is differentSpoken language is differentFuture Of Voice RecognitionFuture Of Voice RecognitionBetter rejection of extraneous speechBetter rejection of extraneous speechBetter recognition of embedded commandsBetter recognition of embedded commandsBetter efficiency on low cost processorsBetter efficiency on low cost processorsStandards for performance evaluationStandards for performance evaluationIncreased portabilityIncreased portabilityLower error ratesLower error ratesImprove overall robustnessImprove overall
View Full Document