Automatic selection of transcribed training material

T.M. Kamm1, G.G.L. Meyer1
1Center for Language and Speech Processing, Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA

Tóm tắt

Conventional wisdom says that incorporating more training data is the surest way to reduce the error rate of a speech recognition system. This, in turn, guarantees that speech recognition systems are expensive to train, because of the high cost of annotating training data. We propose an iterative training algorithm that seeks to improve the error rate of a speech recognizer without incurring additional transcription cost, by selecting a subset of the already available transcribed training data. We apply the proposed algorithm to an alpha-digit recognition problem and reduce the error rate from 10.3% to 9.4% on a particular test set.

Từ khóa

#Speech recognition #Error analysis #Iterative algorithms #Training data #Costs #System testing #Natural languages #Speech processing #Data mining #Automatic speech recognition

Tài liệu tham khảo

noel, 1997, Alphadigits, Center for Spoken Lang Understand Oregon Graduate Inst Sci Technol Portland OR kemp, 1999, Unsupervised Training of a Speech Recognizer: Recent Experiments, Proc EUROSPEECH, 2725 hamaker, 1998, Advances in Alpha Digit Recognition Using Syllables, Proc ICASSP, 421 hamaker, 1997, A proposal for a standard partitioning of the OGI AlphaDigit corpus, Inst Signal Inform Process Mississippi State Univ 10.1006/jcss.1997.1504 young, 1999, The HTK Book Version 2 2 lamel, 2000, Lightly Supervised Acoustic Model Training, presented at ISCA ITRW Workshop on Automatic Speech Recognition Challenges for the New Millennium zavaliagkos, 1998, Utilizing Untranscribed Training Data to Improve Performance, presented at Broadcast News Transcription and Understanding Workshop