Weakly supervised learning dataset (WSLD) and the associated learning challenge
WSLD dataset is now freely available. It is a compact dataset for evaluation and benchmarking of weakly supervised learning algorithms designed for categorical sequential data.
In the learning challenge, the goal of the learning algorithm is to discover the relationship between data sequences and the associated explanatory variables (data labeling). Algorithm performance is evaluated by its ability to infer correct labels for new (unseen) test sequences after an initial learning stage.
Feel free to beat the baseline results with your own approach!
Please send your new results, comments, and improvement ideas by email.
Dataset description (.pdf).
Data download (.mat and .txt formats).
Research interests
[ Computational modeling of language acquisition ]
[ Cognitive aspects of language and speech processing ]
[ Statistical learning ]
Projects and funding I am involved in
D2I - From Data to Intelligence, 2012- (TEKES/TIVIT SHOK)
PARMEEG - New Pattern Recognition Methods in Detection and Classification of Events in the EEG of Preterm Babies, 2010- (TEKES)
TOMU - Towards Multimodal Pattern Discovery and Recognition, 2010- (with Nokia NRC)
Langnet - the Finnish Graduate School in Language Studies, 2010-2011
ACORNS - Acquisition of Communication and Recognition Skills, 2007-2009 (EU FP6 FET)
Teaching (course assistant)
S-89.3610 - Speech Processing (5 cr) [2007, 2008, 2009, 2010, 2011]
S-89.3640 - Methods of Speech Processing (3 cr) [2008, 2009, 2010, 2011, 2012]
S-89.3680 - Speech Processing Seminar (3 cr) [2008]
S-89.4830 - Postgraduate Course in Speech Processing (8 cr) [2008]
Full-paper peer reviewed publications
Räsänen O.: "Computational modeling of phonetic and lexical learning in early language acquisition: existing models and future directions", accepted for publication, Speech Communication.
Räsänen O. & Rasilo H.: "Acoustic analysis supports the existence of a single distributional learning mechanism in structural rule learning from an artificial language", Accepted to Proc. 34th Annual Conference of the Cognitive Science Society (CogSci2012), Sapporo, Japan, 2012.
Räsänen O.: "Context induced merging of synonymous word models in computational modeling of early language acquisition", Proc. ICASSP'2012, Kyoto, Japan, pp. 5037-5040, 2012
(.pdf).
Räsänen O.: "Hierarchical unsupervised discovery of user context from multivariate sensory data", Proc. ICASSP'2012, Kyoto, Japan, pp. 2105-2108, 2012
(.pdf).
Räsänen O. & Laine U.: "A method for noise-robust context-aware pattern discovery and recognition from categorical sequences", Pattern Recognition, Vol. 45, pp. 606-616, 2012
(link)
(.pdf).
Räsänen O., Leppänen J., Laine U., Saarinen, J.: "Comparison of Classifiers in Audio and Acceleration Based Context Classification in Mobile Phones", In Proc. EUSIPCO'11, Barcelona, Spain, pp. 946-950, 2011 (.pdf).
Rasilo H., Laine U., Räsänen O. & Altosaar T.: "Method for speech inversion with large scale statistical evaluation", In Proc. Interspeech'11, Florence, Italy, pp. 2693-2696 , 2011
(.pdf).
Räsänen O.: "A computational model of word segmentation from continuous speech using transitional probabilities of atomic acoustic events", Cognition, Vol. 120, pp. 149-176, 2011
(link)
(.pdf).
Räsänen O., Laine U. & Altosaar T.: "Blind segmentation of speech using non-linear filtering methods", in Ipsic I. (Ed.): Speech Technologies, InTech Publishing, 2011 (.pdf).
Rasilo H., Laine U. & Räsänen O.: "Estimation studies of vocal tract shape trajectory using a variable length and lossy Kelly-Lochbaum model", In Proc. Interspeech'10, Chiba, Japan, pp. pp. 2414-2417, 2010 (.pdf).
Räsänen O.: "Fully Unsupervised Word Learning from Continuous Speech Using Transitional Probabilities of Atomic Acoustic Events", In Proc. Interspeech'10, Chiba, Japan, pp. pp. 2922-2925, 2010 (.pdf).
ten Bosch L., Räsänen O., Driesen J., Aimetti G., Altosaar T., Boves L.: "Do Multiple Caregivers Speed up Language Acquisition", In Proc. Interspeech'09, Brighton, England, pp. 704-707, 2009 (.pdf).
Aimetti G., Moore R., ten Bosch L., Räsänen O. & Laine U.: "Discovering Keywords from Cross-Modal Input: Ecological vs. Engineering Methods for Enhancing Acoustic Repetitions", In Proc. Interspeech'09, Brighton, England, pp. 1171-1174, 2009 (.pdf).
Räsänen O., Laine U.K. & Altosaar T.: "A noise robust method for pattern discovery in quantized time
series: the concept matrix approach", In Proc. Interspeech'09, Brighton, England, pp. 3035-3038, 2009 (.pdf).
Räsänen O., Laine U.K. & Altosaar T.: "An Improved Speech Segmentation Quality Measure: the R-value", In Proc. Interspeech'09, Brighton, England, pp. 1851-1854, 2009
(.pdf).
Räsänen O., Laine U.K. & Altosaar T.: "Self-learning Vector Quantization for Pattern Discovery from Speech", In Proc. Interspeech'09, Brighton, England, pp. 852-855, 2009 (.pdf).
Räsänen O. & Driesen J.: A comparison and combination of segmental and fixed-frame signal representations in NMF-based word recognition, In Proc. 17th Nordic Conference on Computational Linguistics (NODALIDA'09), Odense, Denmark, 2009 (.pdf).
Räsänen O., Laine U.K. & Altosaar T.: "Computational language acquisition by statistical bottom-up processing", In Proc. Interspeech'08, Brisbane, Australia, pp. 1980-1983, 2008 (.pdf).
Other publications
Laine U. & Räsänen O.: "Indirect estimation of formant frequencies through mean spectral variance with application to automatic gender recognition.",
In Proc. 6th International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA), Firenze, Italy, 2009 (.pdf).
ten Bosch L., Boves L. & Räsänen O.: "Learning meaningful units from multimodal input - the effect of interaction strategies." In Proc. Workshop on Child, Computer and Interaction 2009 (WOCCI), Boston, MA, United States, 2009 (.pdf).
Räsänen O.: "A Review of Missing-Feature Methods in Automatic Speech Recognition", in Palomäki K. J., Remes U., and Kurimo M. (Eds.), Studies on noise robust automatic speech recognition. Technical Report TKK-ICS-R19, Helsinki University of Technology, Department of Information and Computer Science, Espoo, Finland, September 2009.
Räsänen O., Altosaar T. & Laine U.K.: "Comparison of prosodic features in Swedish and Finnish IDS/ADS speech", In Proc. Nordic Prosody X, Helsinki, Finland, 2008 (.pdf).
Räsänen O.: "Speech Segmentation and Clustering Methods for a New Speech Recognition Architecture", M.Sc. Thesis, 2007
(.pdf).
Other stuff (working papers/reports/patents)
Laine U.K., Räsänen O., Fagerlund S., Altosaar T., Aimetti G. & Henter G.: "PD module with self-directed search, derived segmental quality measures, full integration of CMM", ACORNS project deliverable, 2009.
Räsänen O.: "Statistical learning and language acquisition: a review", special assignment in speech processing, TKK, 2008 (.pdf)
Laine U.K., Räsänen O., Altosaar T., Driesen J., Aimetti G. & Henter G.: "Methods for enhanced pattern discovery in speech processing", ACORNS project deliverable, 2008,
(.pdf).
Laine U. & Räsänen O.: "Method for pattern discovery and recognition", Patent application number FIN-20086260, 2008
Laine U. & Räsänen O.: "Audiosignaalin segmentointi automaattisesti ilman opetusta (Automatic, unsupervised segmentation of audio signals)", patent no. 120223 (FI), 2007.
