Convolutional maxout neural networks for low-resource speech recognition

Cai, Meng; Shi, Yongzhe; Kang, Jian; Liu, Jia; Su, Tengrong

doi:10.1109/iscslp.2014.6936676

Cited by 15 publications

(15 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the acoustic model based on CNN, the input feature vectors are divided into N non-overlapping frequency bands The formula is as follows: [37], [38]. The nonlinear function of the original network is changed from sigmoid to maxout.…”

Section: ) Convolutional Neural Networkmentioning

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

Section: ) Convolutional Neural Networkmentioning

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

show abstract

“…The untranscribed data is then decoded using the trained SGMM acoustic model and semi-supervised training is performed. The final ASR system is based on the convolutional maxout neural network (CMNN) acoustic model [23] with 40dimensional Mel filterbank features and their first-and secondorder derivatives. Two convolutional layers with 256 maxout neurons and five fully-connected layers with 1000 maxout neurons are used.…”

Section: Experiments Setup 41 the Asr System Setupmentioning

confidence: 99%

A Novel Discriminative Score Calibration Method for Keyword Search

Cai

Zhang

et al. 2016

Interspeech 2016

Self Cite

View full text Add to dashboard Cite

The performance of keyword search systems depends heavily on the quality of confidence scores. In this work, a novel discriminative score calibration method has been proposed. By training an MLP classifier employing the word posterior probability and several novel normalized scores, we can obtain a relative improvement of 4.67% for the actual term-weighted value (ATWV) metric on the OpenKWS15 development test dataset. In addition, a LSTM-CTC based keyword verification method has been proposed to supply extra acoustic information. After the information is added, a further improvement of 7.05% over the baseline can be observed.

show abstract

“…Although some of these studies involved experiments with CNNs as well, these applied the maxout activation only in the fully connected layers [40]. To our knowledge, the only exceptions where the maxout activation was extended to the convolutional neurons as well are the studies of Cai et al [41] and Renals and Swietojanski [42]. Below, we will explain how convolutional and maxout neurons are related, and we will present our solution for the swift evaluation of convolutional maxout units.…”

Section: Maxout Neural Networkmentioning

confidence: 99%

Phone recognition with hierarchical convolutional deep maxout networks

Tóth

2015

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

Deep convolutional neural networks (CNNs) have recently been shown to outperform fully connected deep neural networks (DNNs) both on low-resource and on large-scale speech tasks. Experiments indicate that convolutional networks can attain a 10-15 % relative improvement in the word error rate of large vocabulary recognition tasks over fully connected deep networks. Here, we explore some refinements to CNNs that have not been pursued by other authors. First, the CNN papers published up till now used sigmoid or rectified linear (ReLU) neurons. We will experiment with the maxout activation function proposed recently, which has been shown to outperform the rectifier activation function in fully connected DNNs. We will show that the pooling operation of CNNs and the maxout function are closely related, and so the two technologies can be readily combined to build convolutional maxout networks. Second, we propose to turn the CNN into a hierarchical model. The origins of this approach go back to the era of shallow nets, where the idea of stacking two networks on each other was relatively well known. We will extend this method by fusing the two networks into one joint deep model with many hidden layers and a special structure. We will show that with the hierarchical modelling approach, we can reduce the error rate of the network on an expanded context of input. In the experiments on the Texas Instruments Massachusetts Institute of Technology (TIMIT) phone recognition task, we find that a CNN built from maxout units yields a relative phone error rate reduction of about 4.3 % over ReLU CNNs. Applying the hierarchical modelling scheme to this CNN results in a further relative phone error rate reduction of 5.5 %. Using dropout training, the lowest error rate we get on TIMIT is 16.5 %, which is currently the best result. Besides experimenting on TIMIT, we also evaluate our best models on a low-resource large vocabulary task, and we find that all the proposed modelling improvements give consistently better results for this larger database as well.

show abstract

Convolutional maxout neural networks for low-resource speech recognition

Cited by 15 publications

References 24 publications

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

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

A Novel Discriminative Score Calibration Method for Keyword Search

Phone recognition with hierarchical convolutional deep maxout networks

Contact Info

Product

Resources

About