Presentation | 2002/6/21 Combination of Multiple Recognizer Outputs Based on Maximum Word Acceptance Rate Path Selection Hirofumi YAMAMOTO, Konstantin MARKOV, Kozo OKUDA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this paper, a new method for combining multiple recognition results is proposed. There are three important issues when combining multiple results. (1) What score is associated with each word. (2) How to combine different word sequences, i.e., how to construct a word graph. (3) How to select the best path through the word graph. In our approach, a word a-posteriori probability is used as the word score. The word graph is constructed by the N-dimensional DP matching of multiple word sequences based on the minimum edit distance. As for the best path, we select the word path that gives the maximum expected word acceptance rate. We evaluated this method in two experiments. In the first experiment, we combined the outputs of three systems having different acoustic features. The proposed method resulted in a 2.2-point lower word error rate. In the second experiment, we combined the outputs of the three systems, which use acoustic models with three different frame shifts, and we achieved a 0.6-point lower word error rate. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | ROVER / DP-matching / word acceptance rate |
Paper # | SP2002-49 |
Date of Issue |
Conference Information | |
Committee | SP |
---|---|
Conference Date | 2002/6/21(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Speech (SP) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Combination of Multiple Recognizer Outputs Based on Maximum Word Acceptance Rate Path Selection |
Sub Title (in English) | |
Keyword(1) | ROVER |
Keyword(2) | DP-matching |
Keyword(3) | word acceptance rate |
1st Author's Name | Hirofumi YAMAMOTO |
1st Author's Affiliation | ATR Spoken Language Translation Research Laboratories() |
2nd Author's Name | Konstantin MARKOV |
2nd Author's Affiliation | ATR Spoken Language Translation Research Laboratories |
3rd Author's Name | Kozo OKUDA |
3rd Author's Affiliation | ATR Spoken Language Translation Research Laboratories:(Present address)Sanyo Electric Co., Ltd. Technology R & D Headquarters Digital Systems Development Center Human Interface Department |
Date | 2002/6/21 |
Paper # | SP2002-49 |
Volume (vol) | vol.102 |
Number (no) | 160 |
Page | pp.pp.- |
#Pages | 5 |
Date of Issue |