projects in speech and information processing systems

Projects in Signals and Projects in Signals and Information Processing Information Processing Systems (2007)Systems (2007)

Offered by

Dr. Roberto Togneri Room 4.10, [email protected]

Offered byOffered by

Dr. Roberto Togneri Dr. Roberto Togneri Room 4.10, Room 4.10, [email protected]@ee.uwa.edu.au

CIIPS Signals and Information Processing Systems

SIP FYP Projects (pgs. 26SIP FYP Projects (pgs. 26--29)29)

•• Spoken Language Systems / IDEAL House (4A)Spoken Language Systems / IDEAL House (4A)

•• Speech Processing (4B, 4C, 4D)Speech Processing (4B, 4C, 4D)

•• Intelligent Information Processing (4E, 4F)Intelligent Information Processing (4E, 4F)

•• Biomedical Engineering (4G)Biomedical Engineering (4G)

•• Pattern Recognition (4H)Pattern Recognition (4H)

•• Cryptography (4I, 4J)Cryptography (4I, 4J)


Spoken Language SystemsSpoken Language Systems

4A. Voice4A. Voice--activated Speech and activated Speech and Speaker Recognition Speaker Recognition Interactive SystemsInteractive Systems

–– Command and ControlCommand and Control–– Speaker AuthenticationSpeaker Authentication–– Others: voice activity detection, Others: voice activity detection,

keyword spotting, continuous speech keyword spotting, continuous speech recognition, language understandingrecognition, language understanding

–– More than one student (on More than one student (on different subdifferent sub--projects)projects)


IDEAL HouseIDEAL House

•• VoiceVoice--activated Assistance and Sound Monitoringactivated Assistance and Sound Monitoring–– You are home, how many things you need to manipulate:You are home, how many things you need to manipulate:

•• Remote controls (at least 2 or 3!)Remote controls (at least 2 or 3!)•• Telephone dialling Telephone dialling •• Switch on/off appliances, lamps, etc.Switch on/off appliances, lamps, etc.

–– ANSWERANSWER: Use the power of your voice!: Use the power of your voice!

–– You are still home, is it listening?You are still home, is it listening?•• Fridge beeping (door left open), microwave beeping (food Fridge beeping (door left open), microwave beeping (food

cooked), but you are not in the vicinitycooked), but you are not in the vicinity

–– ANSWERANSWER: Detect the sounds, classify, and react : Detect the sounds, classify, and react accordingly (house knows where you are and tells you: accordingly (house knows where you are and tells you: ““food is cooked, come and get it!food is cooked, come and get it!””))


Speech ProcessingSpeech Processing

Time Time Time

Freq

uenc

y

Clean Mel-Spectrogram Corrupted by white noise (SNR: 10dB)

Spectrographic Maskthreshold: SNR=0dB

•• 4B. Reconstruction of Noise Corrupted Spectrogram4B. Reconstruction of Noise Corrupted Spectrogram–– Additive noise dominates some timeAdditive noise dominates some time--frequency regions and will frequency regions and will

adversely affect recognition (by machine and humans)adversely affect recognition (by machine and humans)

–– Identify regions where noise dominates and attempt to Identify regions where noise dominates and attempt to ““reconstructreconstruct”” the damaged regions by removing the noise and the damaged regions by removing the noise and using the known properties of speech and the reliable parts of using the known properties of speech and the reliable parts of the spectrogram.the spectrogram.

–– Listen to the reconstructed speech and perform recognitionListen to the reconstructed speech and perform recognition



•• 4C. Single Channel Blind Source Separation4C. Single Channel Blind Source Separation–– Two signals (speech + music), need to extract the Two signals (speech + music), need to extract the

speech signal, how?speech signal, how?–– Two microphones, easy: use beamTwo microphones, easy: use beam--forming or BSS to forming or BSS to

spatially separate signalsspatially separate signals–– Single channel recording, harder: but more Single channel recording, harder: but more

interesting. Train basis functions on signal of interest interesting. Train basis functions on signal of interest and use to detect signal (signaland use to detect signal (signal--space analysis!)space analysis!)



•• 4D. Performance Evaluation of Auditory Models4D. Performance Evaluation of Auditory Models–– Use the available MATLAB software (Auditory Image Use the available MATLAB software (Auditory Image

ModelingModeling (AIM) and/or Development System for Auditory (AIM) and/or Development System for Auditory Modellings (DSAM)Modellings (DSAM)

–– Implement and evaluate different models for the Implement and evaluate different models for the identification of important perceptual cues that can be identification of important perceptual cues that can be exploited by exploited by speechspeech recognisers.recognisers.


Intelligent Information ProcessingIntelligent Information Processing

•• 4E. Music Classification and Summarisation4E. Music Classification and Summarisation–– Use classification paradigms (neural networks, Use classification paradigms (neural networks,

support vector machines (SVM), support vector machines (SVM), neuroneuro--fuzzy fuzzy networks (NFN), etc.) on different music genres networks (NFN), etc.) on different music genres (rock, jazz, classical)(rock, jazz, classical)

–– Use higher level semantic and knowledge based Use higher level semantic and knowledge based processed to segment and summarise music pieces.processed to segment and summarise music pieces.


Intelligent Information ProcessingIntelligent Information Processing

•• 4F. Nonlinear Function Mapping using Neural 4F. Nonlinear Function Mapping using Neural Networks (also fuzzy/evolutionary methods)Networks (also fuzzy/evolutionary methods)–– VTR tracking: f(12d features) = 4 VTR tracking: f(12d features) = 4 VTRsVTRs

–– Generation model: f(4 Generation model: f(4 VTRsVTRs) = 12d features) = 12d features

–– We have the VTR data (just released) and the feature We have the VTR data (just released) and the feature data, now we need the mapping!data, now we need the mapping!

–– Universal Function Universal Function ApproximatorsApproximators: MLP, RBF, NFN, GA: MLP, RBF, NFN, GA–– Microsoft Research (USA) is interested in this workMicrosoft Research (USA) is interested in this work


Biomedical EngineeringBiomedical Engineering4G. 4G. Real time EEG processing Real time EEG processing

for interactive ERP and TMSfor interactive ERP and TMS•• Apply a stimulus to produce an Evoked Apply a stimulus to produce an Evoked

Response Potential (ERP)Response Potential (ERP)

•• Stimulus timing is dependent upon the Stimulus timing is dependent upon the recognition of a transitory Brain Wave recognition of a transitory Brain Wave State in the Electroencephalogram (EEG)State in the Electroencephalogram (EEG)

•• In this project we aim to apply signal In this project we aim to apply signal processing and syntactic pattern processing and syntactic pattern recognition for automated identification recognition for automated identification and classification.and classification.

INTERESTED?INTERESTED?

•• Contact Roberto Togneri directly so a Contact Roberto Togneri directly so a visit with Dr. Greg Price from visit with Dr. Greg Price from CCRN CCRN (Centre for Clinical Research in (Centre for Clinical Research in Neuropsychiatry)Neuropsychiatry) can be arranged.can be arranged.


Pattern RecognitionPattern Recognition

•• 4H. Etch Pit Density (EPD) of Semiconductor 4H. Etch Pit Density (EPD) of Semiconductor WafersWafers–– Image processing to enhance defects (e.g. Image processing to enhance defects (e.g. thresholdingthresholding, mask , mask

processing, 2D FFT, etc.)processing, 2D FFT, etc.)–– Detection of defects (feature extraction, clustering and Detection of defects (feature extraction, clustering and

classification)classification)–– Counting the number of defectsCounting the number of defects–– Jointly Supervised with MRGJointly Supervised with MRG


CryptographyCryptography

•• 4I. Performance Analysis of Cryptographic Algorithms4I. Performance Analysis of Cryptographic Algorithms–– Maths/CS/IT Majors: Maths/CS/IT Majors:

•• evaluation of ECC based schemes (strength, computations, etc.)evaluation of ECC based schemes (strength, computations, etc.)•• analysis of timing attacks based on CPU, cache profilinganalysis of timing attacks based on CPU, cache profiling

–– Otherwise: Implement and evaluate publicly available private Otherwise: Implement and evaluate publicly available private and public key encryption algorithms for computational and and public key encryption algorithms for computational and memory resource requirements, what do you recommend?memory resource requirements, what do you recommend?

–– Supported by Motorola Research AustraliaSupported by Motorola Research Australia


CryptographyCryptography

•• 4J. Evaluation of Identity4J. Evaluation of Identity--Based Encryption SchemeBased Encryption Scheme–– With publicWith public--key encryption maintaining the most upkey encryption maintaining the most up--toto--date date

publicpublic--keys of your recipients is an issuekeys of your recipients is an issue–– With IBE all you need to do is know the email address of the With IBE all you need to do is know the email address of the

recipient and encrypt the email using that as the key!recipient and encrypt the email using that as the key!–– Only the recipient needs to worry about authenticating Only the recipient needs to worry about authenticating

himself/herself so to obtain the corresponding decryption keyhimself/herself so to obtain the corresponding decryption key–– Implement a prototype IBE based scheme, and equivalent PKI Implement a prototype IBE based scheme, and equivalent PKI

scheme and evaluate the strengths and weaknessesscheme and evaluate the strengths and weaknesses–– Supported by Motorola Research AustraliaSupported by Motorola Research Australia


Want to know more?Want to know more?

•• SIP FYP 2007 Projects PageSIP FYP 2007 Projects Page– http://www.ee.uwa.edu.au/~roberto/research/projects2007.html

– Also lists 2006 projects (most are still available) and 2008 projects (which didn’t make it this year) in case you are interesting in more projects in this area, or

– Suggest your own project in the speech, information and signals area!

•• Contact me: Contact me: [email protected]

•• Interested students can get more information on each project incInterested students can get more information on each project including:luding:–– Reading list of the key articles and textbooksReading list of the key articles and textbooks

–– Selected WWW pages and resourcesSelected WWW pages and resources

–– Links to software and manualsLinks to software and manuals

–– Contact emails of collaborators who are involved with the projecContact emails of collaborators who are involved with the projectt

http://www.ee.uwa.edu.au/~roberto/research/projects2007.html

mailto:[email protected]

projects in speech and information processing systems

Documents