End-to-end text-dependent speaker verification (2015-09-27T00:00:00.000000Z)

TL;DR

A data-driven, integrated approach to speaker verification, which maps a test utterance and a few reference utterances directly to a single score for verification and jointly optimizes the system's components using the same evaluation protocol and metric as at test time.

Abstract

In this paper we present a data-driven, integrated approach to speaker verification, which maps a test utterance and a few reference utterances directly to a single score for verification and jointly optimizes the system's components using the same evaluation protocol and metric as at test time. Such an approach will result in simple and efficient systems, requiring little domain-specific knowledge and making few model assumptions. We implement the idea by formulating the problem as a single neural network architecture, including the estimation of a speaker model on only a few utterances, and evaluate it on our internal "Ok Google" benchmark for text-dependent speaker verification. The proposed approach appears to be very effective for big data applications Like ours that require highly accurate, easy-to-maintain systems with a small footprint.

Authors

Noam Shazeer

7 papers

Samy Bengio

13 papers

G. Heigold

6 papers

TL;DR

Abstract

Authors

References29 items

Front-End Factor Analysis For Speaker Verification

Automatic gain control and multi-style training for robust small-footprint keyword spotting with deep neural networks

Deep Neural Network Approaches to Speaker and Language Recognition

On Using Very Large Target Vocabulary for Neural Machine Translation

Improving speaker recognition performance in the domain adaptation challenge using deep neural networks

Deep Unfolding: Model-Based Inspiration of Novel Deep Architectures

The NIST 2014 Speaker Recognition i-vector Machine Learning Challenge

Deep neural networks for small footprint text-dependent speaker verification

A novel scheme for speaker recognition using a phonetically-aware deep neural network

Phonetically-constrained PLDA modeling for text-dependent speaker verification with multiple short utterances

Large Scale Distributed Deep Networks

Improving neural networks by preventing co-adaptation of feature detectors

Front-End Factor Analysis for Speaker Verification

Joint Factor Analysis Versus Eigenchannels in Speaker Recognition

Speaker recognition with recurrent neural networks

Long Short-Term Memory

Locally-connected and convolutional neural networks for small footprint speaker recognition

TensorFlow: Large-scale machine learning on heterogeneous systems

Long short-term memory recurrent neural network architectures for large scale acoustic modeling

Automatic language identification using long short-term memory recurrent neural networks

Text-dependent speaker recognition using PLDA with uncertainty propagation

Text dependent speaker verification using a small development set

New Developments in Voice Biometrics for User Authentication

Analysis of i-vector Length Normalization in Speaker Recognition Systems

Bayesian Speaker Verification with Heavy-Tailed Priors

“Your Word is my Command”: Google Search by Voice: A Case Study

Speaker Verification Using Adapted Gaussian Mixture Models

Score Normalization for Text-Independent Speaker Verification Systems

References

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names