Faculty of Information Technology, BUT

Publication Details

Exploiting Hidden-Layer Responses of Deep Neural Networks for Language Recognition

LI Ruizhi, MALLIDI Sri Harish, PLCHOT Oldřich, BURGET Lukáš and DEHAK Najim. Exploiting Hidden-Layer Responses of Deep Neural Networks for Language Recognition. In: Proceedings of Interspeech 2016. San Francisco: International Speech Communication Association, 2016, pp. 2365-2369. ISBN 978-1-5108-3313-5. Available from: https://www.researchgate.net/publication/307889648_Exploiting_Hidden-Layer_Responses_of_Deep_Neural_Networks_for_Language_Recognition
Czech title
Využití odezev ze skryté vrstvy hlubokých neuronových sítí pro rozpoznávání jazyka
Type
conference paper
Language
english
Authors
Li Ruizhi (JHU)
Mallidi Sri Harish (JHU)
Plchot Oldřich, Ing., Ph.D. (DCGM FIT BUT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
Dehak Najim (MIT)
URL
Keywords
LID, I-vector, DNN, hidden layers
Abstract
The most popular way to apply Deep Neural Network (DNN) for Language IDentification (LID) involves the extraction of bottleneck features from a network that was trained on automatic speech recognition task. These features are modeled using a classical I-vector system. Recently, a more direct DNN approach was proposed, it consists of estimating the language posteriors directly from a stacked frames input. The final decision score is based on averaging the scores for all the frames for a given speech segment. In this paper, we extended the direct DNN approach by modeling all hidden-layer activations rather than just averaging the output scores. One super-vector per utterance is formed by concatenating all hidden-layer responses. The dimensionality of this vector is then reduced using a Principal Component Analysis (PCA). The obtained reduce vector summarizes the most discriminative features for language recognition based on the trained DNNs. We evaluated this approach in NIST 2015 language recognition evaluation. The performances achieved by the proposed approach are very competitive to the classical I-vector baseline.
Annotation
The most popular way to apply Deep Neural Network (DNN) for Language IDentification (LID) involves the extraction of bottleneck features from a network that was trained on automatic speech recognition task. These features are modeled using a classical I-vector system. Recently, a more direct DNN approach was proposed, it consists of estimating the language posteriors directly from a stacked frames input. The final decision score is based on averaging the scores for all the frames for a given speech segment. In this paper, we extended the direct DNN approach by modeling all hidden-layer activations rather than just averaging the output scores. One super-vector per utterance is formed by concatenating all hidden-layer responses. The dimensionality of this vector is then reduced using a Principal Component Analysis (PCA). The obtained reduce vector summarizes the most discriminative features for language recognition based on the trained DNNs. We evaluated this approach in NIST 2015 language recognition evaluation. The performances achieved by the proposed approach are very competitive to the classical I-vector baseline.
Published
2016
Pages
2365-2369
Proceedings
Proceedings of Interspeech 2016
Conference
Interspeech 2016, San Francisco, US
ISBN
978-1-5108-3313-5
Publisher
International Speech Communication Association
Place
San Francisco, US
DOI
BibTeX
@INPROCEEDINGS{FITPUB11272,
   author = "Ruizhi Li and Harish Sri Mallidi and Old\v{r}ich Plchot and Luk\'{a}\v{s} Burget and Najim Dehak",
   title = "Exploiting Hidden-Layer Responses of Deep Neural Networks for Language Recognition",
   pages = "2365--2369",
   booktitle = "Proceedings of Interspeech 2016",
   year = 2016,
   location = "San Francisco, US",
   publisher = "International Speech Communication Association",
   ISBN = "978-1-5108-3313-5",
   doi = "10.21437/Interspeech.2016-1584",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/11272"
}
Files
Back to top