DOC PREVIEW
CMU CS 15492 - Speech Translation Case study: Transtac Details

This preview shows page 1-2-3-4-5-6 out of 17 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 17 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Speech Processing 15-492/18-492Speech TranslationCase study: TranstacDetailsTranstac: Two S2S SystemDARPA developed forDARPA developed forCheck points, medical and civil defenseCheck points, medical and civil defenseRequirementsRequirementsTwo wayTwo wayEyesEyes--free (no screen)free (no screen)PortablePortableUsable by real Usable by real usersSusersSTranstac SystemLaptop secured in BackpackOptional speech controlPush-to-Talk ButtonsClose-talking MicrophoneSmall powerful SpeakersTranstac System DetailsTwo way systemTwo way system2 ASR systems: English and Iraqi2 ASR systems: English and Iraqi2 way statistical translation2 way statistical translation2 synthesizers2 synthesizersPushPush--toto--talk systemtalk system(Users don’t like “translate everything mode”)(Users don’t like “translate everything mode”)Echo back ASR resultEcho back ASR resultAnd then translationAnd then translationIraqi LanguageIraqi Arabic is a dialectIraqi Arabic is a dialectMost Iraqi’s write Modern Standard ArabicMost Iraqi’s write Modern Standard ArabicMost Iraqi’s do not write their own dialectMost Iraqi’s do not write their own dialectNo standardized spellingNo standardized spellingTranstacTranstacproject invented oneproject invented oneBut Iraqi’s may not be used to itBut Iraqi’s may not be used to itArabic (MSA and dialects)Arabic (MSA and dialects)Do not write short vowels in wordsDo not write short vowels in wordsData for TrainingCollected human mediated dialogsCollected human mediated dialogsHuman acts as a machineHuman acts as a machinePassed a microphone back an forwardPassed a microphone back an forwardTry to get people not to talk at same timeTry to get people not to talk at same timeLarge number of collections (over 4 years)Large number of collections (over 4 years)650 thousand sentences pairs650 thousand sentences pairsMany different speakersMany different speakersHand transcribed by experts (in Iraqi spelling)Hand transcribed by experts (in Iraqi spelling)Hand translate (Source sentences and Interpreter’s)Hand translate (Source sentences and Interpreter’s)Iraqi ASRAcoustic model from Iraqi dataAcoustic model from Iraqi dataBased on MSA Based on MSA phonesetphonesetNeeds to be small fast modelsNeeds to be small fast modelsDiscriminative TrainingDiscriminative TrainingSpeaker specific adaptationSpeaker specific adaptationLexiconLexiconBased on LDC provided lexiconBased on LDC provided lexiconMultiple pronunciations/typos still a problemMultiple pronunciations/typos still a problemStatistically trained LTS rulesStatistically trained LTS rulesLanguage ModelLanguage ModelTrained on Iraqi input (and translated output)Trained on Iraqi input (and translated output)English ASRAcoustic modelAcoustic modelOriginally using other modelsOriginally using other modelsThen trained from collected dataThen trained from collected data(Mostly military personnel)(Mostly military personnel)LexiconLexiconExisting lexicon but needed to add Military speak: Existing lexicon but needed to add Military speak: MRAP, IEDMRAP, IEDLanguage modelLanguage modelTrained from data providedTrained from data providedTrained from “similar” data found on the webTrained from “similar” data found on the webTraining from hand created “typical” examplesTraining from hand created “typical” examplesTTSStandard English TTSStandard English TTSAppropriate “command” voiceAppropriate “command” voiceUnit selectionUnit selectionAdded lots of military vocabularyAdded lots of military vocabularyIraqi TTSIraqi TTSRecorded from Iraqi radio announcerRecorded from Iraqi radio announcerBased on example sentences in the domainBased on example sentences in the domainLDC lexicon and LTS rules (same as ASR)LDC lexicon and LTS rules (same as ASR)Hand tunedHand tunedS2S Interface IssuesHow do you teach people to use the systemHow do you teach people to use the system““TranstacTranstacsay instructions”say instructions”Not really sufficientNot really sufficientHow can you tell it translated correctlyHow can you tell it translated correctlyGive (speech) feedback.Give (speech) feedback.BacktranslationBacktranslationASR echo backASR echo backS2S Interface IssuesHow do you translate namesHow do you translate namesA correct translation/transliteration is hard to A correct translation/transliteration is hard to understandunderstandMark names in translationsMark names in translations“My name is … Abdullah”“My name is … Abdullah”“He lives on … al“He lives on … al--AqarAqar… street”… street”S2S Evaluation (Transtac)Offline testsOffline testsASRASR-->Text and Text>Text and Text-->Text>TextCompare to translation referencesCompare to translation referencesWER and “BLEU” scoreWER and “BLEU” scoreOnline testsOnline testsConcept transfer (through defined scenarios)Concept transfer (through defined scenarios)Speed (number of concepts per minute)Speed (number of concepts per minute)(English speech masking)(English speech masking)Utility testsUtility testsDoes it really workDoes it really workTranstac ParticipantsDeveloper groupsDeveloper groupsIBMIBMSRISRIBBNBBNCMUCMUUSCUSCEvaluationsEvaluationsTwice a year in Iraqi (somewhere in DC)Twice a year in Iraqi (somewhere in DC)One surprise language (Farsi, One surprise language (Farsi, BahasaBahasaMalay)Malay)Other evaluations with military groupsOther evaluations with military groupsDoes it work??Yes, mostlyYes, mostly27 concepts out of 3027 concepts out of 30--ish turnsish turnsSystems are mostly similarSystems are mostly similarBut some better than othersBut some better than othersOther techniquesOther techniquesBelt/holster based PC with handheld speakerBelt/holster based PC with handheld speakerSmall PC in pouchSmall PC in pouchChest mounted array microphoneChest mounted array microphoneS2S ASR Advanced issuesTight


View Full Document
Download Speech Translation Case study: Transtac Details
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Speech Translation Case study: Transtac Details and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Speech Translation Case study: Transtac Details 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?