Project Details

Jazykově nezávislá detekce klíčových slov

Project Period: 1. 1. 2012 - 31. 12. 2014

Project Type: grant

Code: GPP202/12/P567

Agency: Czech Science Foundation

Program: Postdoktorandské granty

English title
Language-independent spoken term detection
Type
grant
Keywords

keyword spotting, query-by-example, language independent, hidden Markov models, artificial neural network

Abstract

This project aims at research and development of language-independent keyword spotter in spoken speech. The keywords will be entered as examples (Query-by-Example). The application of project results is in search in speech where current approaches fail: exotic languages (insufficient or no training data) and recordings where speakers change language within the conversation. The first goal is to define evaluation data for several languages and to evaluate the state-of-the-art Query-by-Example systems in cross-lingual environment. Main goals are: (1) to design and evaluate an approach to language-independent high-level feature extraction from speech. We will use combination of several language-dependent artificial neural network classifiers. (2) To design and evaluate a GMM/HMM approach to Query-by-Example. It will be important to correctly estimate the keyword model on several examples and to investigate training of the universal background model. We will also compare achieved results with standard language-dependent approaches.

Team members
Szőke Igor, Ing., Ph.D. (UPGM FIT VUT) , research leader
Janda Miloš, Ing. (UPGM FIT VUT) , team leader
Veselý Karel, Ing., Ph.D. (UPGM FIT VUT) , team leader
Publications

2015

2014

2013

2012

Back to top