Lab 2: Isolated WordRecognition
Lab 2: Isolated Word Recognition
Architecture of an ASR systemAcousticLanguageModelModel"Goodmorning'FeatureDecoderExtractionWordsSpeechFeature vectorsXW3/13/2017HUMANCOMPUTERINTERACTIONPage2
Architecture of an ASR system 3/13/2017 HUMAN COMPUTER INTERACTION Page 2
Isolated word recognition based on HMMTrainHMM国选择要打开的文件XHC.SpeechRecognitio..TvoSearch SpeechRecoqnition-h..main.mOrganizeNewfolderODesktopNameTitleContributing arDocumentsRecord atest speech2mySpeech.wav+DownloadsHonor6XspeechRecord.mJMusicPicturesg.interfacesVideosRunGUI开始1CWindows7_OS(OCDDrive(D:)Hisg_interfaces.m号Lenovo_RecovelNehaunri0.8Filename(C.wav)OpenCancel0.6国0.40.2识别波形3.00.10.20.30.40.50.60.70.80.90推款3/13/2017HUMANCOMPUTERINTERACTIONPage3
Isolated word recognition based on HMM Train HMM • main.m Record a test speech • speechRecord.m Run GUI • g_interfaces.m 3/13/2017 HUMAN COMPUTER INTERACTION Page 3 1. 2. 3
Assignment1.Run SpeechRecognition-hmm.Getfamiliar with theprogram.2.TrainanotherHMMusingyourownspeech/wordsandmodifyGUlaccordingly3.Adda new function:Recordtestaudios using Stat/Endbutton4.Savetraining/testaudios indifferent subdirectories3/13/2017HUMANCOMPUTERINTERACTIONPage4
Assignment 1. Run SpeechRecognition-hmm. Get familiar with the program. 2. Train another HMM using your own speech/words and modify GUI accordingly. 3. Add a new function: Record test audios using Stat/End button 4. Save training/test audios in different subdirectories 3/13/2017 HUMAN COMPUTER INTERACTION Page 4
ReportDescribe what you have done in the training and test progress.The report should answer thefollowing questions:The words to recognizeThenumberoftrainingsamplesThemodificationsto GUlandthecodesTheaccuracyofspeechrecognitionandhowtoimproveit,ifpossible?AnalyzetheimpactofdifferentnumberofhmmstatesonrecognitionaccuracySubmityourwork(codeandreport)toTAPrepareareadmefileto illustratehowtorunyourprogramCompressthecode(includingall your.wavfiles andthereadmefile) and the report into a zip file: ID_name_lab2.zip3/13/2017HUMANCOMPUTERINTERACTIONPage5
Report Describe what you have done in the training and test progress. The report should answer the following questions: • The words to recognize • The number of training samples • The modifications to GUI and the codes • The accuracy of speech recognition and how to improve it, if possible? • Analyze the impact of different number of hmm states on recognition accuracy. Submit your work (code and report) to TA • Prepare a readme file to illustrate how to run your program • Compress the code (including all your .wav files and the readme file) and the report into a zip file: ID_name_lab2.zip 3/13/2017 HUMAN COMPUTER INTERACTION Page 5