Fast Adaptation of Deep Neural Network Based on Discriminant Codes for Speech Recognition

By Shaofei Xue, Ossama Abdel-Hamid, Hui Jiang, Lirong Dai, Qingfeng Liu. 2014

155 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Speech & Audio, 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 Speech & Audio

In Signal Processing

Citations in the corpus

ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context (2020, background): “Previous works on hybrid ASR have successfully introduced the context to acoustic models by either stacking a large number of layers, or having a separately trained global vector to represent the speaker and the environment information [13, 14, 15, 16].”
Neural Voice Cloning with a Few Samples (2018, methodology): “The ﬁrst group is speaker adaptation for the whole-model [Yu et al., 2013], a portion of the model [Miao and Metze, 2015, Cui et al., 2017], or merely to a speaker embedding [Abdel-Hamid and Jiang, 2013, Xue et al., 2014].”
Letter-Based Speech Recognition with Gated ConvNets (2017, background): “Most state of the art systems also perform speaker adaptation; iVectors compute a speaker embedding capturing both speaker and environment information (Xue et al., 2014), while fMMLR is a two-pass decoder technique which computes a speaker transform in the ﬁrst pass (Gales and Woodland, 1996).”
Deep Encoder-Decoder Models for Unsupervised Learning of Controllable Speech Synthesis (2018, methodology): “The idea was introduced for speaker adaptation in neural network ASR in [8], [9] under the name “discriminant condition codes” (DCC), and was independently adapted for multi-speaker speech synthesis several times: ﬁrst by Luong et al. [7] and more recently by Arık et al. [11] (Deep Voice 2) and…”
Speaker adaptive training in deep neural networks using speaker dependent bottleneck features (2016, methodology): “Appending speaker information to the input features, in the form of I-vector [3], D-vectors [4], vectorised CMLLR transforms [5] or using speaker codes [6, 7], to make the DNN aware of speaker specific changes have shown to improve the DNN performance.”
Modeling speaker variability using long short-term memory networks for speech recognition (2015, background): “The speaker information can be derived in many different ways, such as the speaker code [12][13] which is jointly learned with the rest of model parameters for each speaker, and the speaker or utterance level i-vector [14][15] which is a popular technique for speaker veriﬁcation and recognition…”
Rectified linear neural networks with tied-scalar regularization for LVCSR (2015, methodology): “To our knowledge, this is the best reported performance on this task for the speaker-independent training (without speaker-specific adaptation and normalization as in [23, 24]) using the framelevel cross-entropy error criterion.”
Maximum a posteriori adaptation of network parameters in deep models (2015, background): “Recent approaches, such as regularization based [7, 8], subspace based [9, 10], transformation based [6, 11, 12, 13], i-Vector based [14], native neural network based [15, 16, 17], factorization based [18] and fast adaptation schemes based on discriminant speaker codes [19, 16, 20, 21], have been pr”
Speaker-Invariant Training Via Adversarial Learning (2018, background): “In [5], SD speaker codes are transformed by a set of SI matrices and then directly added to the biases of the hidden-layer afﬁne transformations.”
Factorized Hidden Layer Adaptation for Deep Neural Network Based Acoustic Modeling (2016, background): “In speaker code [58], a small number of parameters are estimated for fast DNN adaptation.”
Rapid adaptation for deep neural networks through multi-task learning (2015, background): “In recent years, several adaptation approaches have been proposed to address this mismatch issue, such as regularization based [3, 4], subspace based [5, 6], transformation based [7, 8, 9, 10], i-Vector based [11], native neural network adaptation [12, 13, 14], factorized adaptation [15] and speaker”
JHU ASpIRE system: Robust LVCSR with TDNNS, iVector adaptation and RNN-LMS (2015, background): “…and DARPA BOLT Contract No HR0011-12-C-0015. iVectors which capture both speaker and environment speciﬁc information have been shown to be useful for rapid adaptation of the neural network [4, 5, 6]. iVector based adaptation has also been shown to be effective in reverberant environments [7].”
Hierarchical Bayesian combination of plug-in maximum a posteriori decoders in deep neural networks-based speech recognition and speaker adaptation (2017, background): “…a recognition accuracy equal to 17.4%, which is comparable to what reported in the recent literature (Swietojanski et al., 2016; Seide et al., 2011; Xue et al., 2014) when (i) the same training material is used, (ii) the same testing conditions are adopted, and (iii) a feed-forward DNN-based…”
Adapting and controlling DNN-based speech synthesis using input codes (2017, methodology): “and a data-driven numerical encoding sometimes called “discriminant condition codes” (DCCs) [17].”
Low-rank plus diagonal adaptation for deep neural networks (2016, background): “The speaker-specific auxiliary features can be derived in many different ways, such as i-vector [15, 16, 17] and speaker code [18, 19, 20], and then are concatenated with the standard acoustic feature vectors.”
Speaker Adaptive Training of Deep Neural Network Acoustic Models Using I-Vectors (2015, methodology): “With the speaker-code adaptation scheme, [18] accomplishes SAT of DNNs by jointly learning the SD speaker codes and the SI DNN models.”
A time delay neural network architecture for efficient modeling of long temporal contexts (2015, background): “iVectors which capture both speaker and environment specific information have been shown to be useful for instantaneous and discriminative adaptation of the neural network [5, 6].”
Conditional Teacher-student Learning (2019, unknown): “[30, 31], subspace-based [32, 33] and adversarial learning-based [34, 35] approaches.”