TUKE System for MediaEval 2014 QUESST
- 1. TECHNICAL UNIVERSITY OF KOŠICE Laboratory of Speech Technologies in Telecommunications TUKE system for MediaEval 2014 QUESST Jozef VAVREK, Peter VISZLAY, Martin LOJKA, Matúš PLEVA, and Jozef JUHÁR Department of Electronics and Multimedia Communications Technical University of Košice, Slovak Republic {Jozef.Vavrek, Peter.Viszlay, Martin.Lojka, Matus.Pleva, Jozef.Juhar}@tuke.sk Zero-resource approaches Timit ParDat1 SpeechDat CZ SpeechDat SK Tab.1 Evaluation of primary low-resource (p-low) and general zero resource (g-zero) systems (* indicates late submission) dev Cnxe TWV (act/min) (act/min) 0.161/0.162 0.091/0.091 0.191/0.191 0.106/0.107 Acknowledgments eval (act/min) (act/min) 0.959/0.891 0.973/0.934 0.947/0.853 0.970/0.921 0.154/0.154 0.075/0.077 0.168/0.169 0.102/0.103 0.960/0.892 0.974/0.934 0.948/0.854 0.971/0.922 Tab.2 Processing resources measures p-low* g-zero* Searching Algorithm (Weighted Fast Sequential - DTW): 1) one step forward moving strategy, when each DTW search is carried out sequentially, block by block, with size equal to the length of query; 2) linear time-aligned accumulated distance for speeding up sequential DTW without considerable loss in retrieving performance; 3) optimization of global minimum for set of alignment paths by implementing weighted cumulative distance (WCD) parameter. 500 detected candidates This publication is the result of the Project implementation: University Science Park TECHNICOM for Innovation Applications Supported by Knowledge Technology, ITMS: 26220220182, supported by the Research & Development Operational Programme funded by the ERDF (100%). MediaEval 2014: Query by Example Search on Speech Task, 16-17 October 2014, Barcelona, Spain VAD PCA-based Posteriorgrams GMM-based Query Abstract Two approaches to QbE (Query-by-Example) retrieving system, proposed by the Technical University of Košice (TUKE) for the query by example search on speech task (QUESST), are presented in this paper. Our main interest was focused on building such QbE system, which is able to retrieve all given queries with and without using any external speech resources.Therefore we developed posteriorgram-based keyword matching system, which utilizes a novel weighted fast sequential variant of DTW (WFS-DTW) algorithm in order to detect occurrences of each query within the particular utterance file, using two GMM-based acoustic units modeling approaches. The first one, referred as low-resource approach, employs language-dependent phonetic decoders to convert queries and utterances into posteriorgrams. The second one, defined as zero-resource approach, implements combination of unsupervised segmentation and clustering techniques by using only provided utterance files. Results System Overview system Cnxe TWV p-low g-zero system ISF SSF PMUI PMUS PL p-low (dev) 0.61 0.0034 0.05 2.46 0.010 g-zero (dev) 1.50 0.0042 1.40 3.92 0.225 Conclusions and Future Work Phonetic Decoders Utterances Type 1 - 13 MFCC - PCA-based feature selection - K-means clustering (K=75) Type 2 - 39 MFCC - Type 1 + Viterbi seg. & new GMM training Type 3 - 39 MFCC - flat start training (GMM-based) - phone sequences from Type 1 Type 4 - 39 MFCC - GMM-based seg. (64 GM) - EHHM (64 states / 256 GM) Low-resource approaches WFS-DTW Score normalization & Fusion - scaling 0-1 - max-score merging fusion - z-normalization - Still big differences in performance between p-low and g-zero approaches, even if the score fusion technique was applied. - There is also considerable gap between act and min Cnxe despite the fact that the act and max TWV are perfectly calibrated. - An improved calibration/fusion models based on affine transformation and linear-regression will be investigated in the future. The indexing was done using 2xIBM x3650 (Intel E5530 @ 2.4 GHz, 8 cores), 28 GB RAM, under Debian OS. Searching algorithm was running on 52xIBM dx360 M3 cluster (Intel E5645 @ 2.4GHz, 624 cores), 48 GB RAM per node, running on Scientific Linux 6 and Torque 2.5.13. time-aligned & labelled segments new GMM training