Conference paper

MATĚJKA Pavel, ZHANG Le, NG Tim, MALLIDI Sri Harish, GLEMBEK Ondřej, MA Jeff and ZHANG Bing. Neural Network Bottleneck Features for Language Identification. In: Proceedings of Odyssey 2014. Joensuu: International Speech Communication Association, 2014, pp. 299-304. ISSN 2312-2846.
Publication language:english
Original title:Neural Network Bottleneck Features for Language Identification
Title (cs):Příznaky z neuronové sítě s úzkým hrdlem pro identifikaci jazyka
Pages:299-304
Proceedings:Proceedings of Odyssey 2014
Conference:Odyssey 2014: The Speaker and Language Recognition Workshop
Place:Joensuu, FI
Year:2014
Journal:Proceedings of Odyssey: The Speaker and Language Recognition Workshop, Vol. 2014, No. 6, 4 Rue des Fauvettes - Lous Tourils, F-66390 BAIXAS, FR
ISSN:2312-2846
Publisher:International Speech Communication Association
URL:http://www.fit.vutbr.cz/research/groups/speech/publi/2014/matejka_odyssey2014_299-304-35.pdf [PDF]
Keywords
language identification, noisy speech, robust feature extraction
Annotation
We have presented the bottleneck features in the context of Language identification. It combines benefits of both phonotactic and acoustic system. Usually, the phonotactic system is favorable for the long duration files, while acoustic for the short ones. This approach takes the advantage of both. In addition, we can also use modeling of context dependent phonemes in bottleneck features. This brings very nice improvement over the context independent phonemes.
Abstract
This paper presents the application of Neural Network Bottleneck (BN) features in Language Identification (LID). BN features are generally used for Large Vocabulary Speech Recognition in conjunction with conventional acoustic features, such as MFCC or PLP.We compare the BN features to several common types of acoustic features used in the state-of-the-art LID systems. The test set is from DARPA RATS (Robust Automatic Transcription of Speech) program, which seeks to advance state-of-the-art detection capabilities on audio from highly degraded radio communication channels. On this type of noisy data, we show that in average, the BN features provide a 45% relative improvement in the Cavgor Equal Error Rate (EER) metrics across several test duration conditions, with respect to our single best acoustic features.
BibTeX:
@INPROCEEDINGS{
   author = {Pavel Mat{\v{e}}jka and Le Zhang and Tim Ng and Harish Sri
	Mallidi and Ond{\v{r}}ej Glembek and Jeff Ma and Bing Zhang},
   title = {Neural Network Bottleneck Features for Language
	Identification},
   pages = {299--304},
   booktitle = {Proceedings of Odyssey 2014},
   journal = {Proceedings of Odyssey: The Speaker and Language Recognition
	Workshop},
   volume = {2014},
   number = {6},
   year = {2014},
   location = {Joensuu, FI},
   publisher = {International Speech Communication Association},
   ISSN = {2312-2846},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php?id=10686}
}

Your IPv4 address: 54.167.44.32
Switch to IPv6 connection

DNSSEC [dnssec]