End­-to-­End Keyword Search From Speech without ASR

OUYANG, Patrick (2024) End­-to-­End Keyword Search From Speech without ASR. Master thesis, Voice Technology (VT).

PDF MScVT-S5476410.pdf Restricted to Repository staff only Download (866kB)

Abstract

The thesis focuses on developing an end­-to-­end, ASR­-free keyword search from speech system based on pre­-trained wav2vec 2.0 and BERT models. Our experimental results show significant improvements in detecting keywords directly from raw audio inputs com­ pared to previous work. We also present considerations for develop­ ing such a system as well as future directions.

Item Type:	Thesis (Master)
Name supervisor:	Schauble, J.K. and Nayak, S.
Date Deposited:	19 Sep 2025 13:30
Last Modified:	19 Sep 2025 13:30
URI:	https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/555

Actions (login required)

View Item