Rapid Discriminative Learning

2004 ◽  
Vol 8 (2) ◽  
pp. 108-114
Author(s):  
Jun Rokui ◽  
Keyword(s):  
Neural Network ◽  
Speech Recognition ◽  
Hierarchical Model ◽  
Variable Length ◽  
Discriminative Learning ◽  
Learning Method ◽  
Recognition Method ◽  
Highly Effective ◽  
Hierarchical Neural Network

This paper presents MCE/GPD using GPD that is known as a highly effective discriminative learning method. MCE/GPD is an excellent recognition method that is applicable especially to speech recognition, since it excels in recognizing performance and can be used to deal with variable-length vectors. MCE/GPD involves a problem of calculation resulting from c omplicated algorithms making it impractical. In this paper, we propose a learning method to increase speed at learning based on a hierarchical model. We used a hierarchical neural network to evaluate the method’s performance.

scholarly journals Domain-Adversarial Based Model with Phonological Knowledge for Cross-Lingual Speech Recognition

Electronics ◽  
2021 ◽  
Vol 10 (24) ◽  
pp. 3172
Author(s):  
Qingran Zhan ◽  
Xiang Xie ◽  
Chenguang Hu ◽  
Juan Zuluaga-Gomez ◽  
Jing Wang ◽  
...  
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Speech Recognition ◽  
Training Data ◽  
Target Language ◽  
Learning Method ◽  
Acoustic Features ◽  
Adversarial Learning ◽  
Phonological Knowledge ◽  
Cross Lingual

Phonological-based features (articulatory features, AFs) describe the movements of the vocal organ which are shared across languages. This paper investigates a domain-adversarial neural network (DANN) to extract reliable AFs, and different multi-stream techniques are used for cross-lingual speech recognition. First, a novel universal phonological attributes definition is proposed for Mandarin, English, German and French. Then a DANN-based AFs detector is trained using source languages (English, German and French). When doing the cross-lingual speech recognition, the AFs detectors are used to transfer the phonological knowledge from source languages (English, German and French) to the target language (Mandarin). Two multi-stream approaches are introduced to fuse the acoustic features and cross-lingual AFs. In addition, the monolingual AFs system (i.e., the AFs are directly extracted from the target language) is also investigated. Experiments show that the performance of the AFs detector can be improved by using convolutional neural networks (CNN) with a domain-adversarial learning method. The multi-head attention (MHA) based multi-stream can reach the best performance compared to the baseline, cross-lingual adaptation approach, and other approaches. More specifically, the MHA-mode with cross-lingual AFs yields significant improvements over monolingual AFs with the restriction of training data size and, which can be easily extended to other low-resource languages.

Sign in / Sign up

Export Citation Format

Share Document