Detail výsledku
Out-of-Vocabulary Words Detection with Attention and CTC Alignments in an End-to-End ASR System
Vydana Hari Krishna
Burget Lukáš, doc. Ing., Ph.D., UPGM (FIT)
Černocký Jan, prof. Dr. Ing., UPGM (FIT)
This work explores the effectiveness of detecting positions ofout-of-vocabulary words (OOVs) in a decoded utterance usingattention weights and CTC per-frame outputs of an end-to-endsystem predicting word sequences. We show that the end-to-endapproach can be effective for the task of OOV detection. CTCalignments are shown to provide better temporal informationabout the positions of OOV words than attention, and thereforeare more suitable for the task. The detected positions of OOVoccurrences are utilized for the recurrent OOV recovery task inwhich probabilistic representations of the pronunciations of thedetected OOVs are clustered in order to find repeating words.Improved detection results are shown to correlate with betterperformance of the recovery of recurrent OOVs.
Speech recognition, Out-of-vocabulary, OOV,Attention, CTC, End-to-end
@inproceedings{BUT175843,
author="Ekaterina {Egorova} and Hari Krishna {Vydana} and Lukáš {Burget} and Jan {Černocký}",
title="Out-of-Vocabulary Words Detection with Attention and CTC Alignments in an End-to-End ASR System",
booktitle="Proceedings Interspeech 2021",
year="2021",
journal="Proceedings of Interspeech",
volume="2021",
number="8",
pages="2901--2905",
publisher="International Speech Communication Association",
address="Brno",
doi="10.21437/Interspeech.2021-1756",
issn="1990-9772",
url="https://www.isca-speech.org/archive/interspeech_2021/egorova21_interspeech.html"
}Vícenásobné služby inteligentního konverzačního agenta pro přijetí, řízení a integraci občanů třetích zemí v EU, EU, Horizon 2020, zahájení: 2020-02-01, ukončení: 2023-04-30, ukončen