Robustness of linear discriminant analysis in automatic speech recognition

Katz, Marcel; Meier, H.-G.; Dolfing, Hans; Klakow, Dietrich

doi:10.1109/icpr.2002.1047921

Cited by 6 publications

(6 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The resulting concatenated LDA input vectors have grown up to 954 components per time frame. For both corpora, the white noise components have not caused any degradation in WER, i.e., increasing the feature vector size by In contrast to [5], the experiments on real data presented here have not led to significant changes in WER. Nevertheless, it should be mentioned that the average number of observations per LDA class observed here (4500) differs strongly from [5], where only 100 observations were presented per LDA class.…”

Section: Combination With White Noise Componentsmentioning

confidence: 74%

“…In [5], feature vectors of an artificial recognition task have been augmented with an increasing number of white noise components. The classification error rate was doubled when augmenting a twodimensional feature vectors with 200 white noise components.…”

Section: Analysis Of Resultsmentioning

confidence: 99%

“…In [4] a decrease in WER was observed when overly increasing the LDA input window length. In [5], addition of random coefficients to the feature vectors for an artificial recognition problem also showed degradations. The latter experiment was repeated for speech recognition on real data in this work and could not be confirmed.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Feature combination using linear discriminant analysis and its pitfalls

Schlüter¹,

Zolnay²,

Ney³

2006

Interspeech 2006

View full text Add to dashboard Cite

Section: Combination With White Noise Componentsmentioning

confidence: 74%

Section: Analysis Of Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Feature combination using linear discriminant analysis and its pitfalls

Schlüter¹,

Zolnay²,

Ney³

2006

Interspeech 2006

View full text Add to dashboard Cite

“…Although high resolution MS brings accurate molecular information, it also brings a huge amount of data, which makes the rapid screening of active ingredients technically difficult and less efficient. How to efficiently process these complex raw data and deeply mine the information on composition and structure of substances has gradually become an urgent need in chemical, medical and biological fields [16,17] …”

Section: Introductionmentioning

confidence: 99%

Pattern Recognition Analysis of Metabolites in Escherichia coli Based on ESI‐Orbitrap Mass Spectrometry

Wang

Yan

et al. 2023

Chemistry & Biodiversity

View full text Add to dashboard Cite

To achieve rapid detection of carbapenem-resistant Escherichia coli strains, a pattern recognition method based on electrospray ionization Orbitrap mass spectrometry (ESI-Orbitrap MS) was used for the analysis of drug-resistant, and sensitive strains of metabolites were analyzed. Results of five clustering methods applied to analytical data of metabolites were evaluated using iso-phenotypic coefficients. The effectiveness of three methods, principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA) and orthogonal partial least squares discriminant analysis (OPLS-DA), was compared. Univariate statistics such as t-test and fold change were also used to examine the screened differential information. Both PLS-DA and OPLS-DA could achieve rapid identification of strain classes, and OPLS-DA was more powerful in screening 96 significantly different ions. This work is expected to be useful for rapid and accurate identification of strains.

show abstract

“…In machine learning and data mining, Fisher Discriminant Analysis (FDA) is one of the widely used discriminant algorithms that seeks to find directions so that data in the same classes are projected near to each other while ones in different classes are projected as far as possible for classification or dimension reduction. This has wide applications in face recognition [78], speech recognition [71], and digit recognition [17].…”

Section: Privacy-preserving Fisher Discriminant Analysismentioning

confidence: 99%

Privacy-preserving data mining via secure multiparty computation

Han¹

View full text Add to dashboard Cite

Throughout my postgraduate study, I feel that PhD life can be condensed to be a process of simply iterating from finding a question and then discovering a systematical solution after having a bee in one's head. However, during the process, various professional skills have been developed, including quickly reading/writing a technical paper, systematically thinking a problem, presentation skills and social communication and so on. In the last four years, I could haven't finished my PhD studies without many people's help. I have been very fortunate to work with all the great people. I would like to express my gratitude to the following organizations and people for their support and contribution: Ministry of Education, Singapore and Nanyang Technological University, Singapore, for offering me MOE undergraduate scholarship and the NTU research scholarship respectively to pursue a higher degree. Associate Professor Ng Wee Keong, my Ph.D. supervisor, for his invaluable guidance, advice, encouragement, and sharing of his knowledge. Only from him, I started to know the world of research and to do the research seriously as a part of my life. Also from him, I have learned how to tackle a problem systematically, deeply, and carefully. Through individual discussions, he inspired me of many good and invaluable questions to my research and of my passion to discover more solutions. I have also learned how to write a paper and present my work properly and clearly from him after his patient face-to-face tutoring. Without his supervision and encouragement, I would never have achieved the outputs in the thesis. I am very thankful to Dr. Vincent C. S. Lee and Dr. Philip S. Yu for their kindness and invaluable discussion during the period they visited NTU. With those i ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library discussions, new interesting and problems have been inspired. Their comments and suggestions on my draft papers are very helpful to make them professional. Additional thanks go to Dr. Sun Aixin, Dr. Vivekanand Gopalkrishnan, and Dr. James Cheng for their helps and encouragement whenever I have problems. I would also thank Dr. Li Xiang for her kind supervision and guide on my Final Year Project during my undergraduate study.

show abstract

Robustness of linear discriminant analysis in automatic speech recognition

Cited by 6 publications

References 3 publications

Feature combination using linear discriminant analysis and its pitfalls

Feature combination using linear discriminant analysis and its pitfalls

Pattern Recognition Analysis of Metabolites in Escherichia coli Based on ESI‐Orbitrap Mass Spectrometry

Privacy-preserving data mining via secure multiparty computation

Contact Info

Product

Resources

About