The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Dalam pengecaman pertuturan, anotasi keyakinan menggunakan ciri keyakinan tunggal atau gabungan ciri berbeza untuk pengelasan. Ciri keyakinan ini sentiasa diekstrak daripada maklumat penyahkodan. Walau bagaimanapun, terbukti bahawa kira-kira 30% pengetahuan pemahaman pertuturan manusia terutamanya diperoleh daripada maklumat peringkat tinggi. Oleh itu, cara untuk mengekstrak ciri keyakinan tahap tinggi yang bebas dari segi statistik daripada maklumat penyahkodan patut diselidik dalam pengecaman pertuturan. Dalam makalah ini, algoritma pengekstrakan ciri keyakinan baru berdasarkan persamaan topik terpendam dicadangkan. Setiap pengedaran topik perkataan dan pengedaran topik konteks dalam satu hasil pengecaman diperoleh terlebih dahulu menggunakan model topik peruntukan Dirichlet terpendam (LDA), dan kemudian, ciri keyakinan perkataan yang dicadangkan diekstrak dengan menentukan persamaan antara kedua-dua pengedaran topik ini. Eksperimen menunjukkan bahawa ciri yang dicadangkan meningkatkan bilangan sumber maklumat ciri keyakinan dengan kesan pelengkap maklumat yang baik dan boleh meningkatkan prestasi anotasi keyakinan dengan berkesan digabungkan dengan ciri keyakinan daripada maklumat penyahkodan.
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Salinan
Wei CHEN, Gang LIU, Jun GUO, Shinichiro OMACHI, Masako OMACHI, Yujing GUO, "Novel Confidence Feature Extraction Algorithm Based on Latent Topic Similarity" in IEICE TRANSACTIONS on Information,
vol. E93-D, no. 8, pp. 2243-2251, August 2010, doi: 10.1587/transinf.E93.D.2243.
Abstract: In speech recognition, confidence annotation adopts a single confidence feature or a combination of different features for classification. These confidence features are always extracted from decoding information. However, it is proved that about 30% of knowledge of human speech understanding is mainly derived from high-level information. Thus, how to extract a high-level confidence feature statistically independent of decoding information is worth researching in speech recognition. In this paper, a novel confidence feature extraction algorithm based on latent topic similarity is proposed. Each word topic distribution and context topic distribution in one recognition result is firstly obtained using the latent Dirichlet allocation (LDA) topic model, and then, the proposed word confidence feature is extracted by determining the similarities between these two topic distributions. The experiments show that the proposed feature increases the number of information sources of confidence features with a good information complementary effect and can effectively improve the performance of confidence annotation combined with confidence features from decoding information.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.E93.D.2243/_p
Salinan
@ARTICLE{e93-d_8_2243,
author={Wei CHEN, Gang LIU, Jun GUO, Shinichiro OMACHI, Masako OMACHI, Yujing GUO, },
journal={IEICE TRANSACTIONS on Information},
title={Novel Confidence Feature Extraction Algorithm Based on Latent Topic Similarity},
year={2010},
volume={E93-D},
number={8},
pages={2243-2251},
abstract={In speech recognition, confidence annotation adopts a single confidence feature or a combination of different features for classification. These confidence features are always extracted from decoding information. However, it is proved that about 30% of knowledge of human speech understanding is mainly derived from high-level information. Thus, how to extract a high-level confidence feature statistically independent of decoding information is worth researching in speech recognition. In this paper, a novel confidence feature extraction algorithm based on latent topic similarity is proposed. Each word topic distribution and context topic distribution in one recognition result is firstly obtained using the latent Dirichlet allocation (LDA) topic model, and then, the proposed word confidence feature is extracted by determining the similarities between these two topic distributions. The experiments show that the proposed feature increases the number of information sources of confidence features with a good information complementary effect and can effectively improve the performance of confidence annotation combined with confidence features from decoding information.},
keywords={},
doi={10.1587/transinf.E93.D.2243},
ISSN={1745-1361},
month={August},}
Salinan
TY - JOUR
TI - Novel Confidence Feature Extraction Algorithm Based on Latent Topic Similarity
T2 - IEICE TRANSACTIONS on Information
SP - 2243
EP - 2251
AU - Wei CHEN
AU - Gang LIU
AU - Jun GUO
AU - Shinichiro OMACHI
AU - Masako OMACHI
AU - Yujing GUO
PY - 2010
DO - 10.1587/transinf.E93.D.2243
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E93-D
IS - 8
JA - IEICE TRANSACTIONS on Information
Y1 - August 2010
AB - In speech recognition, confidence annotation adopts a single confidence feature or a combination of different features for classification. These confidence features are always extracted from decoding information. However, it is proved that about 30% of knowledge of human speech understanding is mainly derived from high-level information. Thus, how to extract a high-level confidence feature statistically independent of decoding information is worth researching in speech recognition. In this paper, a novel confidence feature extraction algorithm based on latent topic similarity is proposed. Each word topic distribution and context topic distribution in one recognition result is firstly obtained using the latent Dirichlet allocation (LDA) topic model, and then, the proposed word confidence feature is extracted by determining the similarities between these two topic distributions. The experiments show that the proposed feature increases the number of information sources of confidence features with a good information complementary effect and can effectively improve the performance of confidence annotation combined with confidence features from decoding information.
ER -