Meta Learning for End-To-End Low-Resource Speech Recognition

Hsu, Jui-Yang; Chen, Yuan-Jui; Lee, Hung-yi

doi:10.1109/icassp40776.2020.9053112

Cited by 76 publications

(54 citation statements)

References 21 publications

(21 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Few-shot learning: We consider the problem of adversarial image classification as metalearning-based few-shot classification. There are three data sets [12][13][14][15][16]43 : a training set, a support set, and a test set. Note that the support set and testing set share the same label space, but the training set has its own label space that is disjoint with the support/testing set.…”

Section: Problem Setupmentioning

confidence: 99%

“…There are three data sets: a training set, a support set, and a test set, in the setting of metalearning. [12][13][14][15][16]43 On the ImageNet-A data set and the ImageNet-R data set we choose 30 images of the 200 classes to construct a training set, the remaining samples of these 200 classes are used as support set and test set, and the ratio of these two is 1: 1. On the ImageNet-C and Tiny ImageNet-C data sets, we randomly choose 20 images of each class (75 classes in total) for training and testing.…”

Section: Experimental Settingsmentioning

confidence: 99%

“…Following the classical meta‐learning principle, 12 – 16,43 a classifier could be trained to allot a category label

\overset{x}{true}

to each sample in the test set

\overset{y}{true}

, and we would only utilize the support set. But, in most different scenarios, this classifier usually performs poorly as the support set lacks labeled samples.…”

Section: Proposed Modelmentioning

confidence: 99%

“…Perhaps humans have the ability to take advantage of the experiences to learn. 11 Is that not one of the mechanisms of metalearning [12][13][14][15][16] ? Thus, why not utilize the mechanics of meta-learning to construct a network that captures the relation between the subject and its adversarial image?…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Fighting fire with fire: A spatial–frequency ensemble relation network with generative adversarial learning for adversarial image classification

et al. 2021

View full text Add to dashboard Cite

Adversarial images generated by generative adversarial networks are not close to any existing benign images, and contain nonrobust features that have been identified as critical to the robustness of a machine learning model. Since adversarial images have an underlying distribution that differs from normal images, these kinds of images can offer valuable features for training a robust model. To deal with these special features, we focus on a novel machine learning task of adversarial images classification, where adversarial images can be used to investigate the problem of classifying adversarial images themselves. In the setting of this novel task, adversarial images are the ONLY kind of data used in training and testing, rather than not just a set of testing images as usual. To this end, we propose a novel spatial–frequency ensemble relation network with generative adversarial learning. First, we present a spatial–frequency ensemble representation learning to extract the feature of training images. Second, we design a meta‐learning‐based relation model to gain the relationship between images. Third, to achieve a robust model, we utilize generative adversarial learning and transform the relationship into a Jacobian matrix. Finally, we design a discriminator model that determines whether an adversarial image is from the matching category or not. Experimental results demonstrate that our approach achieves significantly higher performance compared with other state‐of‐the‐arts.

show abstract

Section: Problem Setupmentioning

confidence: 99%

Section: Experimental Settingsmentioning

confidence: 99%

“…Following the classical meta‐learning principle, 12 – 16,43 a classifier could be trained to allot a category label

\overset{x}{true}

to each sample in the test set

\overset{y}{true}

, and we would only utilize the support set. But, in most different scenarios, this classifier usually performs poorly as the support set lacks labeled samples.…”

Section: Proposed Modelmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Fighting fire with fire: A spatial–frequency ensemble relation network with generative adversarial learning for adversarial image classification

et al. 2021

View full text Add to dashboard Cite

show abstract

“…speakers was used to train the meta-learner, and finally achieved a lower WER in DNN and TDNN models. Hsu et al [95] proposed MetaASR, which was learned from six source tasks through the Model-Agnostic Meta Learning (MAML) algorithm to obtain a good initialization parameter of the shared encoder that performs quick finetuning for four target tasks. The results showed that MetaASR performed better than MultiASR in only four target languages directly.…”

Section: ) Meta Learningmentioning

confidence: 99%

Acoustic Modeling Based on Deep Learning for Low-Resource Speech Recognition: An Overview

Kang

Chen

et al. 2020

IEEE Access

View full text Add to dashboard Cite

The polarization of world languages is becoming more and more obvious. Many languages, mainly endangered languages, are of low-resource attribute due to lack of information. Both language conservation and cultural heritage face important challenges. Therefore, speech recognition for lowresource scenario has become a hot topic in the field of speech. Based on the complex network structures and huge model parameters, deep learning has become a powerful science in the process of speech recognition, which has a broad and far-reaching significance for the study of low-resource speech recognition. Aiming at the characteristic of low resource, this paper reviews the history and research status of two kinds of acoustic models of deep learning neural networks and acoustic end-to-end structures. We further elaborate on several key techniques for improving performance in the two aspects of data and model training. There are two projects for low-resource languages introduced in this paper. The possible future developments are finally pointed out. These works provide some reference for computer speech and language processing.

show abstract

Learning from the Negativity: Deep Negative Correlation Meta-Learning for Adversarial Image Classification

Zheng

Lan

Wang

et al. 2021

MultiMedia Modeling

View full text Add to dashboard Cite

Meta Learning for End-To-End Low-Resource Speech Recognition

Cited by 76 publications

References 21 publications

Fighting fire with fire: A spatial–frequency ensemble relation network with generative adversarial learning for adversarial image classification

Fighting fire with fire: A spatial–frequency ensemble relation network with generative adversarial learning for adversarial image classification

Acoustic Modeling Based on Deep Learning for Low-Resource Speech Recognition: An Overview

Learning from the Negativity: Deep Negative Correlation Meta-Learning for Adversarial Image Classification

Contact Info

Product

Resources

About