Title:
LEARNING DEVICE FOR ACOUSTIC MODEL AND COMPUTER PROGRAM FOR SAME
Document Type and Number:
WIPO Patent Application WO/2018/066436
Kind Code:
A1
Abstract:
[Problem] To provide a learning device for an acoustic model wherein speech recognition accuracy can be increased in an acoustic model making use of the characteristics of a neural network (NN). [Solution] A learning device 350 includes: a learning processing unit 362 for optimizing a connectionist temporal classification acoustic model (CTC-AM) 364 such that the sum across all learning data for posterior probability of correct subword sequences for learning data is maximized when an observation sequence in learning data stored in a learning data storage unit 360 is presented; and an MBR learning processing unit 366, an accuracy evaluation unit 374 and a learning/evaluation control unit 378 for further optimizing the CTC-AM 364 such that expected values for evaluation values representing the accuracy of word sequence hypotheses estimated using the CTC-AM 364 and language models 368, 370 is maximized when an observation sequence of data for evaluation that is stored in an evaluation data storage unit 376 is presented.
Inventors:
KANDA NAOYUKI (JP)
Application Number:
PCT/JP2017/035018
Publication Date:
April 12, 2018
Filing Date:
September 27, 2017
Export Citation:
Assignee:
NAT INST INF & COMM TECH (JP)
International Classes:
G10L15/06; G10L15/16
Other References:
SAK, HASIM ET AL.: "Learning Acoustic Frame Labeling for Speech Recognition with Recurrent Neural Networks", PROC. ICASSP 2015, 19 April 2015 (2015-04-19), pages 4280 - 4284, XP033064506
KINGSBURY, BRIAN: "Lattice-based Optimization of Sequence Classification Criteria for Neural-network Acoustic Modeling", PROC. ICASSP 2009, 19 April 2009 (2009-04-19), pages 3761 - 3764, XP031460091
KANDA, NAOYUKI ET AL.: "Maximum A Posteriori based Decoding for CTC Acoustic Models", PROC. INTERSPEECH, 8 September 2016 (2016-09-08), pages 1868 - 1872, XP055498597
KINGSBURY, BRIAN: "Lattice-based Optimization of Sequence Classification Criteria for Neural-network Acoustic Modeling", PROC. ICASSP 2009, 19 April 2009 (2009-04-19), pages 3761 - 3764, XP031460091
KANDA, NAOYUKI ET AL.: "Maximum A Posteriori based Decoding for CTC Acoustic Models", PROC. INTERSPEECH, 8 September 2016 (2016-09-08), pages 1868 - 1872, XP055498597
Attorney, Agent or Firm:
SHIMIZU, Satoshi (JP)
Download PDF: