Title: | Sequence-to-Sequence CNN-BiLSTM Based Glottal Closure Instant Detection from Raw Speech |
Authors: | Matoušek, Jindřich Tihelka, Daniel |
Citation: | MATOUŠEK, J. TIHELKA, D. Sequence-to-Sequence CNN-BiLSTM Based Glottal Closure Instant Detection from Raw Speech. In Artificial Neural Networks in Pattern Recognition; 10th IAPR TC3 Workshop, ANNPR 2022; Dubai, United Arab Emirates, November 24-26, 2022; Proceedings. Cham: Springer Nature Switzerland AG, 2022. s. 107-120. ISBN: 978-3-031-20649-8 , ISSN: 0302-9743 |
Issue Date: | 2022 |
Publisher: | Springer Nature Switzerland AG |
Document type: | konferenční příspěvek ConferenceObject |
URI: | 2-s2.0-85142752874 http://hdl.handle.net/11025/51298 |
ISBN: | 978-3-031-20649-8 |
ISSN: | 0302-9743 |
Keywords in different language: | glottal closure instant detection;deep learning;recurrent neural network;convolutional neural network |
Abstract in different language: | In this paper, we propose to frame glottal closure instant (GCI) de- tection from raw speech as a sequence-to-sequence prediction problem and to explore the potential of recurrent neural networks (RNNs) to handle this prob- lem. We compare the RNN architecture to widely used convolutional neural net- works (CNNs) and to some other machine learning-based and traditional non- learning algorithms on several publicly available databases. We show that the RNN architecture improves GCI detection. The best results were achieved for a joint CNN-BiLSTM model in which RNN is composed of bidirectional long short-term memory (BiLSTM) units and CNN layers are used to extract relevant features. |
Rights: | Plný text je přístupný v rámci univerzity přihlášeným uživatelům. © The Author(s), under exclusive licence to Springer Nature B.V. |
Appears in Collections: | Konferenční příspěvky / Conference Papers (KKY) OBD |
Files in This Item:
File | Size | Format | |
---|---|---|---|
Matousek_Tihelka_Sequence-to-Sequence_ANNPR_2022.pdf | 551,79 kB | Adobe PDF | View/Open Request a copy |
Please use this identifier to cite or link to this item:
http://hdl.handle.net/11025/51298
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.