
MultiTask Learning with HighOrder Statistics for Xvector based TextIndependent Speaker Verification
The xvector based deep neural network (DNN) embedding systems have demo...
read it

Deep Neural Network Embedding Learning with HighOrder Statistics for TextIndependent Speaker Verification
The xvector based deep neural network (DNN) embedding systems have demo...
read it

Speaker Embedding Extraction with Phonetic Information
Speaker embeddings achieve promising results on many speaker verificatio...
read it

Multiobjective Optimization Training of PLDA for Speaker Verification
Most current stateoftheart textindependent speaker verification syst...
read it

Improved RawNet with Feature Map Scaling for Textindependent Speaker Verification using Raw Waveforms
Recent advances in deep learning have facilitated the design of speaker ...
read it

Shared latent subspace modelling within GaussianBinary Restricted Boltzmann Machines for NIST iVector Challenge 2014
This paper presents a novel approach to speaker subspace modelling based...
read it

Multitask Metric Learning for Textindependent Speaker Verification
In this work, we introduce metric learning (ML) to enhance the deep embe...
read it
Gaussian speaker embedding learning for textindependent speaker verification
The xvector maps segments of arbitrary duration to vectors of fixed dimension using deep neural network. Combined with the probabilistic linear discriminant analysis (PLDA) backend, the xvector/PLDA has become the dominant framework in textindependent speaker verification. Nevertheless, how to extract the xvector appropriate for the PLDA backend is a key problem. In this paper, we propose a Gaussian noise constrained network (GNCN) to extract xvector, which adopts a multitask learning strategy with the primary task classifying the speakers and the auxiliary task just fitting the Gaussian noises. Experiments are carried out using the SITW database. The results demonstrate the effectiveness of our proposed method
READ FULL TEXT
Comments
There are no comments yet.