supervisor: dr. eddie jones co-supervisor: dr martin glavin electronic engineering department final...
TRANSCRIPT
Supervisor: Dr. Eddie JonesCo-supervisor: Dr Martin Glavin
Electronic Engineering Department Final Year Project 2008/09
Development of a Speaker Recognition/Verification System
for Security Applications
ContentsBackgroundInitial objectives Development stepsOutcomes/ConclusionQuestions
BackgroundUniversal method of communication.Unique to each user.Speech as a user interface:
Telephone banking.Call centre routing.
BackgroundWhat is Speaker Recognition?
Recognition of who is speaking based on characteristics of their speech signal.
Speaker Identification: Determines which registered speaker has spoken.
Speaker Verification: Accept or reject a claimed identity of a speaker.
Enrolling in the system with speech samples.
Basic flow diagram of the system
Feature/characteristics
Extraction
Similarity
Similarity
Similarity
Maximum Resemblance
selected
Matrix of speaker 2 characteristics
Matrix of speaker 1 characteristics
Matrix of speaker N characteristics
Input Speech Speaker identified
Initial ObjectivesResearch into speaker
recognition/verification.Simulation of Front End Processor in
MATLAB.Simulation of Classifier (Neural
Networks).Investigation of Speaker Recognition
over the internet.Investigation and development of a real-
time version of the system
Development stepsResearch:
MatlabSpeaker recognitionMFCC (Mel Frequency Cepstral
Coefficients)ANN (Artificial Neural Networks)VoIP technology & speaker recognition
Front End ProcessorClassifier
Front End ProcessorPreparing the signal for analysis: Endpointing,
framing, windowing, overlapping, analysis using MFCC fitlerbank, timewarping.
What are MFCCs:Close representation to the human auditory
system.Triangular filters spaced linearly and
logarithmically at low and high frequencies respectively.
Triangular filter are used to weigh a piece of the spectrum, and then the weighted values are summed together to give the overall filter output.
Preparing utterance data for training to the neural network.
ClassifierANN (Artificial Neural Network) -
Interconnected group of artificial neurons which processes information using a connectionist approach to information processing.
Multilayer Perceptron:Input nodes of the database of speakers.Hidden layer to weigh each connection to
show the behaviour of the network.Output node matches to the input data.A high output value will appear on the correct
node
Outcomes/ConclusionDevelopment of knowledge on Speaker
Recognition software.Development of the MATLAB
programming language skills.Speaker characteristics extracted from
speech.
Questions?
MFCC Filter bank
0 20 40 60 80 100 120 1400
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
Artificial Neural NetworkHidden Units
Input Units Output units
i
j
k
Connection weights