Analysis of DNN-based Embeddings for Language Recognition on the NIST LRE 2017

Analysis of DNN-based Embeddings for Language Recognition on the NIST LRE 2017

Paper in proceedings (conference paper)

In this work, we analyze different designs of a language identification(LID) system based on embeddings. In our case, anembedding represents a whole utterance (or a speech segmentof variable duration) as a fixed-length vector (similar to the ivector).Moreover, this embedding aims to capture informationrelevant to the target task (LID), and it is obtained by training adeep neural network (DNN) to classify languages. In particular,we trained a DNN based on bidirectional long short-term memory(BLSTM) recurrent neural network (RNN) layers, whoseframe-by-frame outputs are summarized into mean and standarddeviation statistics for each utterance. After this pooling layer,we add two fully connected layers whose outputs are used asembeddings, which are afterwards modeled by a Gaussian linearclassifier (GLC). For training, we add a softmax output layerand train the whole network with multi-class cross-entropy objectiveto discriminate between languages. We analyze the effectof using data augmentation in the DNN training, as well asdifferent input features and architecture hyper-parameters, obtainingconfigurations that gradually improved the performanceof the embedding system. We report our results on the NISTLRE 2017 evaluation dataset and compare the performance ofembeddings with a reference i-vector system. We show thatthe best configuration of our embedding system outperforms thestrong reference i-vector system by 3% relative, and this is furtherpushed up to 10% relative improvement via a simple scorelevel fusion.

In this work, we analyze different designs of a language identification(LID) system based on embeddings. In our case, anembedding represents a whole utterance (or a speech segmentof variable duration) as a fixed-length vector (similar to the ivector).Moreover, this embedding aims to capture informationrelevant to the target task (LID), and it is obtained by training adeep neural network (DNN) to classify languages. In particular,we trained a DNN based on bidirectional long short-term memory(BLSTM) recurrent neural network (RNN) layers, whoseframe-by-frame outputs are summarized into mean and standarddeviation statistics for each utterance. After this pooling layer,we add two fully connected layers whose outputs are used asembeddings, which are afterwards modeled by a Gaussian linearclassifier (GLC). For training, we add a softmax output layerand train the whole network with multi-class cross-entropy objectiveto discriminate between languages. We analyze the effectof using data augmentation in the DNN training, as well asdifferent input features and architecture hyper-parameters, obtainingconfigurations that gradually improved the performanceof the embedding system. We report our results on the NISTLRE 2017 evaluation dataset and compare the performance ofembeddings with a reference i-vector system. We show thatthe best configuration of our embedding system outperforms thestrong reference i-vector system by 3% relative, and this is furtherpushed up to 10% relative improvement via a simple scorelevel fusion.

language recognition

language recognition

LOZANO DÍEZ, A.; PLCHOT, O.; MATĚJKA, P.; NOVOTNÝ, O.; GONZALEZ-RODRIGUEZ, J.

2019

26.06.2018

International Speech Communication Association

Les Sables d'Olonne

Proceedings of Odyssey 2018 The Speaker and Language Recognition Workshop

2312-2846

Proceedings of Odyssey: The Speaker and Language Recognition Workshop Odyssey 2014, Joensuu, Finland

2018

6

Republic of Finland

39

46

8

https://www.isca-speech.org/archive/Odyssey_2018/pdfs/42.pdf

lozano_odyssey2018_42