On combining i-vectors and discriminative adaptation methods for unsupervised speaker normalization in DNN acoustic models

Samarakoon, Lahiru; Sim, Khe Chai

doi:10.1109/icassp.2016.7472684

Cited by 20 publications

(10 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There has been considerable recent work exploring the use of i-vectors [27] for this purpose. I-vectors, which can be regarded as basis vectors spanning a subspace of speaker variability, were first used for adaptation in a GMM framework by Karafiat et al [28], and were later successfully employed for DNN adaptation [29]- [34]. Other examples of auxiliary features include the use of speakerspecific bottleneck features obtained from a speaker separation DNN [35], the use of out-of-domain tandem features [24], as well as speaker codes [36]- [38] in which a specific set of units for each speaker is optimised.…”

Section: Dnn Acoustic Modelling and Adaptationmentioning

confidence: 99%

“…Directly adapting all the weights of a large DNN is computationally and data intensive, and results in large speakerdependent parameter sets. Smaller subsets of the DNN weights may be modified, including biases and slopes of hidden units [7], [34], [43], [44]. Another recently developed approach relies on learning hidden unit contributions (LHUC) for testonly adaptation [10], [11] as well as in a SAT framework [45].…”

Section: Dnn Acoustic Modelling and Adaptationmentioning

confidence: 99%

See 1 more Smart Citation

Differentiable Pooling for Unsupervised Acoustic Model Adaptation

Swietojanski

Renals

2016

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

We present a deep neural network (DNN) acoustic model that includes parametrised and differentiable pooling operators. Unsupervised acoustic model adaptation is cast as the problem of updating the decision boundaries implemented by each pooling operator. In particular, we experiment with two types of pooling parametrisations: learned $L_p$-norm pooling and weighted Gaussian pooling, in which the weights of both operators are treated as speaker-dependent. We perform investigations using three different large vocabulary speech recognition corpora: AMI meetings, TED talks and Switchboard conversational telephone speech. We demonstrate that differentiable pooling operators provide a robust and relatively low-dimensional way to adapt acoustic models, with relative word error rates reductions ranging from 5--20% with respect to unadapted systems, which themselves are better than the baseline fully-connected DNN-based acoustic models. We also investigate how the proposed techniques work under various adaptation conditions including the quality of adaptation data and complementarity to other feature- and model-space adaptation methods, as well as providing an analysis of the characteristics of each of the proposed approaches.Comment: 11 pages, 7 Tables, 7 Figures in IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 24, num. 11, 201

show abstract

Section: Dnn Acoustic Modelling and Adaptationmentioning

confidence: 99%

Section: Dnn Acoustic Modelling and Adaptationmentioning

confidence: 99%

Differentiable Pooling for Unsupervised Acoustic Model Adaptation

Swietojanski

Renals

2016

IEEE/ACM Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

show abstract

“…Domain Adaptation (DA) is a particular case of transfer learning that leverages labeled data in the source domain, to learn a classifier for unlabeled data in the target domain [15]. In the recent years, domain adaptation methods have been successfully developed and applied in many practical tasks such as sentiment analysis [28], object recognition in different situations [29,30], facial recognition [31], speech recognition [32], video recognition [33] etc. Generally, it is assumed that the task is the same for different domains, i.e.…”

Section: Introductionmentioning

confidence: 99%

Multi-Layer domain adaptation method for rolling bearing fault diagnosis

et al. 2019

View full text Add to dashboard Cite

In the past years, data-driven approaches such as deep learning have been widely applied on machinery signal processing to develop intelligent fault diagnosis systems. In real-world applications, domain shift problem usually occurs where the distribution of the labeled training data, denoted as source domain, is different from that of the unlabeled testing data, known as target domain. That results in serious diagnosis performance degradation. This paper proposes a novel domain adaptation method for rolling bearing fault diagnosis based on deep learning techniques. A deep convolutional neural network is used as the main architecture. The multi-kernel maximum mean discrepancies (MMD) between the two domains in multiple layers are minimized to adapt the learned representations from supervised learning in the source domain to be applied in the target domain. The domain-invariant features can be efficiently extracted in this way, and the cross-domain testing performance can be significantly improved. Experiments on a popular rolling bearing dataset are carried out to validate the effectiveness of the domain adaptation approach, and the diagnosis performance is extensively evaluated in different scenarios. Comparisons with other approaches and related works on the same dataset demonstrate the superiority of the proposed method. The experimental results of this study suggest the proposed domain adaptation method offers a new and promising tool for intelligent fault diagnosis.

show abstract

“…Even though DNNs are often trained on large databases to learn variabilities in the input feature space, they are still vulnerable due to mismatched conditions between training and testing data and lead to significant performance degradations [3]. Speaker adaptation is commonly used in DNNs to reduce the mismatch conditions occurring due to speaker variabilities among training and testing data [4]. Adaptation to acoustic conditions has also been shown to improve DNN‐based acoustic modelling performance [3].…”

Section: Introductionmentioning

confidence: 99%

“…The utterance-dependent (UD) transformation matrix is also produced in a different way. Furthermore, comparing our approach to most other DNN adaptation methods [4,6], the main differences lie in the structure used to represent the additional information, and in the training procedure. In other methods [9], the speaker information is given as an additional feature to the DNN.…”

mentioning

confidence: 99%

Hidden variability subspace learning for adaptation of deep neural networks

2018

View full text Add to dashboard Cite

This Letter proposes a deep neural network (DNN) adaptation method, herein referred to as the hidden variability subspace (HVS) method, to achieve improved robustness under diverse acoustic environments arising due to differences in conditions, e.g. speaker, channel, duration and environmental noise. In the proposed approach, a set of conditiondependent parameters is estimated to adapt the hidden layer weights of the DNN in the HVS to reduce the condition mismatch. These condition-dependent parameters are then connected to various layers through a new set of adaptively trained weights. The authors evaluate the proposed hidden variability learning method on a language identification task and show that significant performance gains can be obtained by discriminatively estimating a set of adaptation parameters to compensate the mismatch in the trained model.

show abstract

On combining i-vectors and discriminative adaptation methods for unsupervised speaker normalization in DNN acoustic models

Cited by 20 publications

References 22 publications

Differentiable Pooling for Unsupervised Acoustic Model Adaptation

Differentiable Pooling for Unsupervised Acoustic Model Adaptation

Multi-Layer domain adaptation method for rolling bearing fault diagnosis

Hidden variability subspace learning for adaptation of deep neural networks

Contact Info

Product

Resources

About