Discriminative score normalization for keyword search decision

Author

Van Tung Pham ; Haihua Xu ; Chen, Nancy F. ; Sivadas, Sunil ; Boon Pang Lim ; Eng Siong Chng ; Haizhou Li

Author_Institution

Nanyang Technol. Univ., Singapore, Singapore

fYear

2014

fDate

4-9 May 2014

Firstpage

7078

Lastpage

7082

Abstract

Many keyword search (KWS) systems make “hit/false alarm (FA)” decisions based on the lattice-based posterior probability, which is incomparable across keywords. Therefore, score normalization is essential for a KWS system. In this paper, we investigate the integration of two novel features, ranking-score and relative-to-max, into a discriminative score normalization method. These features are extracted by considering all competing hypotheses of a putative detection. A metric-based normalization method is also applied as a post-processing step to further optimize the term-weighted value (TWV) evaluation metric. We report empirical improvements over standard baselines using the Vietnamese data from IARPA´s Babel program in the NIST OpenKWS13 Evaluation setup.

Keywords

content-based retrieval; feature extraction; search problems; Babel program; IARPA; KWS systems; NIST OpenKWS13 Evaluation setup; TWV evaluation metric; Vietnamese data; discriminative score normalization method; empirical improvements; hit-false alarm decisions; keyword search systems; lattice-based posterior probability; metric-based normalization method; postprocessing step; putative detection; ranking-score; relative-to-max; term-weighted value evaluation metric; Feature extraction; Keyword search; Lattices; Measurement; NIST; Speech; Speech processing; confidence estimation; discriminative modeling; keyword spotting; score normalization; spoken term detection (STD); under-resourced languages;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech and Signal Processing (ICASSP), 2014 IEEE International Conference on

Conference_Location

Florence

Type

conf

DOI

10.1109/ICASSP.2014.6854973

Filename

6854973