Welcome to the homepage of the project

Vector Quantization in Speaker Recognition

University of Joensuu, Department of Computer Science

PROBLEM STATEMENT:

Speaker recognition is usually a general name referring to two different subtasks: speaker identification (SI) and speaker verification (SV).
Our research primaly concentrates on the identification task. The aim in SI is to recognize the unknown speaker from a set of known speakers (closed-set SI).
Referring to the Figure 1 below, we can see that a speaker recognition system is composed of the following modules:

Front-end processing - the "signal processing" part, which converts the sampled speech signal into set of feature vectors, which characterize the properties of speech that can separate different speakers. Front-end processing is performed both in training- and recognition phases.
Speaker modelling - this part performs a reduction of feature data by modelling the distributions of the feature vectors.
Speaker database - the speaker models are stored here.
Decision logic - makes the final decision about the identity of the speaker by comparing unknown feature vectors to all models in the database and selecting the best matching model.

Fig.1: Schematic diagram of the closed-set speaker identification system.

THE GOAL OF OUR RESEARCH:

Algorithm development:

Feature extraction
Speaker modeling using clustering and VQ [VQ research page]
Discriminative matching
Real-time issues, optimization

Applying speaker recognition technology in real world [PUMS project]

HISTORY AND NEWS OF THIS PROJECT:

August 2004 - First journal paper got accepted to IEEE Trans. Speech and Audio Processing
February 2004 - Guess who got the degree of Ph.Lic. today (20.2.2004) !?! :-)
Autumn 2003 - PUMS project has started, project manager Juhani joins the research group
Spring 2003 - VQ-based classifier fusion prototype was designed and a program prototype developed. Preliminarty results look promising. The program is in test use at the University of Helsinki, Department of Phonetics.
Autumn 2002 & Spring 2003 - An algorithm to speed up speaker identification performance was introduced. Further development is currently under work.
Spring 2002 - Two summer workers join the project two papers got accepted. Sounds good to me!
Autumn 2001 - Many "promising approaches" to be implemented and tested (the phrase used by all researches in many occassions). Wait until coded, tested and reported.
Spring 2001 - Development of a discriminative training/matching method. => Clearly better results in all cases! :-)
Autumn 2000 - The previous results indicated that maybe there aren't any clusters in the speech data. => Statistical analysis (F-test criterion and PCA-transform visualization) of the feature vectors. The results: clear indication that indeed, there aren't any clusters in the speech! Surprised the authors little.
Spring 2000 - Analysis of different clustering algorithms in speaker identification. The results: no difference!
Year 1999 - Tomi Kinnunen writes his MSc thesis about the subject with the knowledge virtually nothing about acoustics, phonetics or DSP - but very enthusiastic about general pattern recognition.

CURRENT RESEARCH TEAM AND RESPONSIBILITIES:

[In alphabetical order - "corresponding author" marked by *]

Core team:
	Pasi Fränti (Ph.D., Professor)	Project leader, Clustering/VQ algorithms, pattern recognition
	Ville Hautamäki (MSc)	Classifier fusion, programming
	Evgeny Karpov (M.Sc.)	Real-time speaker identification, programming
	Tomi Kinnunen (PhD)*	Feature extraction and matching algorithms
	Ismo Kärkkäinen (M.Sc.)	Clustering algorithms, programming
	Jussi Parkkinen (Ph.D., Professor)	Pattern recognition and DSP support
	Juhani Saastamoinen (Ph.Lic.)	Project Manager (PUMS project)
Students:
	Zdenek Fiedler	Empirical study of the recognition accuracy in practise
	Olga Grebenskaya	Adaptive speech recognition / GUI development
	Rosa González Hautamäki	F0 estimation and modeling for speaker recognition
	Timo Viinikka	Effect of GSM coding to speaker recognition performance

CO-OPERATIVE PARTNERS:

PUBLICATIONS:

Theses:

T. Kinnunen: Spectral Features for Automatic Text-Independent Speaker Recognition, Ph.Lic. thesis, Department of Computer Science, University of Joensuu, 2003. [PDF]. [Summary of the results in PPT presentation]

E. Karpov: Real-Time Speaker Identification, M.Sc. thesis, Department of Computer Science, University of Joensuu, 2003. [PDF]

T. Kilpeläinen: Automaattisen puhujanvarmennuksen päätöslogiikka, Pro gradu -tutkielma, Tietojenkäsittelytieteen laitos, Joensuun yliopisto, 2002. ("The decision logic of automatic speaker verification", M.Sc. thesis, Department of Computer Science, University of Joensuu, 2002.) [PDF] (in Finnish).

T. Kilpeläinen: Automaattisen puhujanvarmennuksen päätöslogiikkaa, Kandidaatintutkielma, Tietojenkäsittelytieteen laitos, Joensuun yliopisto, 2001. ("On the decision logic in automatic speaker verification", B.Sc. thesis, Department of Computer Science, University of Joensuu, 2001.) [PDF] (in Finnish).

T. Kinnunen: Automaattinen puhujan tunnistus, Pro gradu -tutkielma, Tietojenkäsittelytieteen laitos, Joensuun yliopisto, 1999. ("Automatic speaker recognition", M.Sc. thesis, Department of Computer Science, University of Joensuu, 1999.) [PDF] (in Finnish).

Journal papers:

T. Kinnunen, E. Karpov, P. Fränti, "Real-Time Speaker Identification and Verification", Accepted for publication in IEEE Transactions on Speech and Audio Processing.

Conference papers:

T. Kinnunen, E. Karpov, P. Fränti, "Efficient Online Cohort Selection Method for Speaker Verification", Proc. 8th Int. Conf. on Spoken Language Processing (ICSLP'2004), Vol. III, pp. 2401-2402, Jeju Island, Korea, Oct. 4-8, 2004. [PDF]

T. Kinnunen, E. Karpov, P. Fränti, "Real-Time Speaker Identification", Proc. 8th Int. Conf. on Spoken Language Processing (ICSLP'2004), Vol. III, pp. 1805-1808, Jeju Island, Korea, Oct. 4-8, 2004. [PDF]

J. Saastamoinen, E. Karpov, V. Hautamäki, P. Fränti, "Automatic Speaker Recognition for Series 60 Mobile Devices", Proc. 9th International Conference Speech and Computer (SPECOM'2004), pp. 353-360, St. Petersburg, Russia, September 20-22, 2004. [PDF]

T. Kinnunen, V. Hautamäki, P. Fränti, "Fusion of Spectral Feature Sets for Accurate Speaker Identification", Proc. 9th International Conference Speech and Computer (SPECOM'2004), pp. 361-365, St. Petersburg, Russia, September 20-22, 2004. [PDF]

E. Karpov, T. Kinnunen, P. Fränti, "Symmetric Distortion Measure for Speaker Recognition", Proc. 9th International Conference Speech and Computer (SPECOM'2004), pp. 366-370, St. Petersburg, Russia, September 20-22, 2004. [PDF]

T. Kinnunen, V. Hautamäki, P. Fränti, "On the Fusion of Dissimilarity-Based Classifiers for Speaker Identification", Proc. 8th European Conference on Speech Communiation and Technology (Eurospeech 2003), pp. 2641-2644, Geneva, Switzerland, September 1-4, 2003. [PDF]

T. Kinnunen, E. Karpov, P. Fränti: "A Speaker Pruning Algorithm for Real-Time Speaker Identification", Proc. 4th International Conference on Audio- and Video-Based Biometric Person Authentication (AVBPA 2003), pp. 639-646, Guilford, UK, June 9-11, 2003. [PDF]

T. Kinnunen: "Designing a Speaker-Discriminative Adaptive Filter Bank for Speaker Recognition", Proc. 7th International Conference on Spoken Language Processing (ICSLP 2002), pp. 2325-2328, Denver, Colorado, USA, September 16-20, 2002. [PDF]

T. Kinnunen, I. Kärkkäinen: "Class-Discriminative Weighted Distortion Measure for VQ-Based Speaker Identification", Proc. Joint IAPR International Workshop on Statistical Pattern Recognition (S+SPR 2002), pp. 681-688, Windsor, Canada, August 6-9, 2002. [PDF]

T. Kinnunen, I. Kärkkäinen, P. Fränti: "Is Speech Data Clustered? - Statistical Analysis of Cepstral Features", Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001)), vol. 4, pp. 2627-2630, Aalborg, Denmark, September 3-7, 2001. [PDF]

T. Kinnunen, P. Fränti: "Speaker Discriminative Weighting Method for VQ-Based Speaker Identification", Proc. 3rd International Conference on audio-and video-based biometric person authentication (AVBPA 2001)), pp. 150-156, Halmstad, Sweden, June 6-8, 2001. [PDF]

T. Kinnunen, T. Kilpeläinen, P. Fränti: "Comparison of Clustering Algorithms in Speaker Identification", Proc. IASTED Int. Conf. Signal Processing and Communications (SPC 2000), pp. 222-227, Marbella, Spain, September 19-22, 2000. [PDF]