Publication detail

End-to-End Open Vocabulary Keyword Search With Multilingual Neural Representations

YUSUF, B. ČERNOCKÝ, J. SARAÇLAR, M.

Original Title

End-to-End Open Vocabulary Keyword Search With Multilingual Neural Representations

Type

journal article in Web of Science

Language

English

Original Abstract

Conventional keyword search systems operate on automatic speech recognition (ASR) outputs, which causes them to have a complex indexing and search pipeline. This has led to interest in ASR-free approaches to simplify the search procedure. We recently proposed a neural ASR-free keyword search model which achieves competitive performance while maintaining an efficient and simplified pipeline, where queries and documents are encoded with a pair of recurrent neural network encoders and the encodings are combined with a dot-product. In this article, we extend this work with multilingual pretraining and detailed analysis of the model. Our experiments show that the proposed multilingual training significantly improves the model performance and that despite not matching a strong ASR-based conventional keyword search system for short queries and queries comprising in-vocabulary words, the proposed model outperforms the ASR-based system for long queries and queries that do not appear in the training data.

Keywords

Keyword search, spoken term detection, end-to-end keyword search, asr-free keyword search, keyword spotting.

Authors

YUSUF, B.; ČERNOCKÝ, J.; SARAÇLAR, M.

Released

2. 8. 2023

Publisher

IEEE

Location

PISCATAWAY, NJ

ISBN

2329-9290

Periodical

IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH AND LANGUAGE PROCESSING

Year of study

31

Number

08

State

United States of America

Pages from

3070

Pages to

3080

Pages count

11

URL

BibTex

@article{BUT185202,
  author="YUSUF, B. and ČERNOCKÝ, J. and SARAÇLAR, M.",
  title="End-to-End Open Vocabulary Keyword Search With Multilingual Neural Representations",
  journal="IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH AND LANGUAGE PROCESSING",
  year="2023",
  volume="31",
  number="08",
  pages="3070--3080",
  doi="10.1109/TASLP.2023.3301239",
  issn="2329-9290",
  url="https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10201906"
}