Time delay deep neural network-based universal background models for speaker recognition

By David Snyder, Daniel Garcia-Romero, Daniel Povey. 2015

141 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Exploring the Encoding Layer and Loss Function in End-to-End Speaker and Language Recognition System (2018, background): “Typical examples are the GMM Supervector/i-vector [1, 3] or the recently proposed DNN i-vector [15, 16].”
X-Vectors: Robust DNN Embeddings for Speaker Recognition (2018, methodology): “A traditional i-vector system based on the GMM-UBM recipe described in [11] serves as our acoustic-feature baseline system.”
Spoken Language Recognition using X-vectors (2018, methodology): “Besides the bottleneck layer, it uses the same architecture and training recipe as the system described in Section 2.2 of [15].”
Deep Neural Network Embeddings for Text-Independent Speaker Verification (2017, methodology): “The baseline is a traditional i-vector system that is based on the GMM-UBM Kaldi recipe described in [11].”
Self-Attentive Speaker Embeddings for Text-Independent Speaker Verification (2018, background): “Also, hybrid approaches that incorporate deep neural networks (DNNs) trained for automatic speech recognition (ASR) into the i-vector system have proved to be beneficial in some conditions [6, 7, 8, 9, 10].”
Speaker recognition based on deep learning: An overview (2020, methodology): “However, this comes at the cost of greatly increased computational complexity over the traditional GMM-UBM/i-vector systems [38, 51], since that a DNN usually has more parameters than GMM.”
ABX-Discriminability Measures and Applications. (Mesures de discriminabilité ABX et applications) (2016, background): “Indeed, they are in line with recent development in speech features design associated with the replacement of GMM-HMM-based technology by NN-based technology: state-of-the-art ASR systems are now based on Melspectrograms (see for example [92, 93]), while state-of-the-art speaker recognition systems ”
The IBM 2016 Speaker Recognition System (2016, methodology): “More recently, a supervised GMMUBM (with full covariance matrices) based on DNN posteriors was also successfully evaluated for telephony speaker recognition [8].”
Deep neural network-based speaker embeddings for end-to-end speaker verification (2016, methodology): “Recently, this paradigm has been improved by incorporating deep neural network (DNN) acoustic models [8, 9, 10, 11, 12].”
A Novel Learnable Dictionary Encoding Layer for End-to-End Language Identification (2018, background): “Each component in the dictionary represents a phoneme (or senone) physically, and the statistics is accumulated through senone posteriors, as is done in recently popular DNN i-vector approach [8, 9, 10].”