Boosting systems for large vocabulary continuous speech recognition

• Published in: Speech communication Vol. 54; no. 2; pp. 212 - 218
• Main Authors: Saon, George, Soltau, Hagen
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-02-2012; Elsevier
• Subjects: Acoustic modeling; Acoustics; Algorithms; Applied sciences; Boosting; Decision trees; Exact sciences and technology; Information, signal and communications theory; Iterative methods; Recognition; Signal processing; Speech processing; Speech recognition; Statistics; Telecommunications and information theory; Training; Trains; Boosting; Speech recognition; Acoustic modeling; Performance evaluation; Discriminant analysis; Probabilistic approach; Iterative method; Cable television; Acoustic method; Decision tree; Learning; English; Multiple models approach; Arabic; Gaussian process; Linear combination; Audiovisual document; Hidden Markov models; News; Maximum likelihood; Speech processing; Learning algorithm
• ISSN: 0167-6393; 1872-7182
• DOI: 10.1016/j.specom.2011.07.011

► We apply the Adaboost algorithm to large vocabulary continuous speech recognition. ► Acoustic models are trained sequentially on re-weighted data. ► Phonetic decision trees are also included in the boosting procedure. ► We study the impact of boosting for ML and discriminatively trained models. ► Accuracy gains on English and Arabic broadcast news transcription are obtained.
We employ a variant of the popular Adaboost algorithm to train multiple acoustic models such that the aggregate system exhibits improved performance over the individual recognizers. Each model is trained sequentially on re-weighted versions of the training data. At each iteration, the weights are decreased for the frames that are correctly decoded by the current system. These weights are then multiplied with the frame-level statistics for the decision trees and Gaussian mixture components of the next iteration system. The composite system uses a log-linear combination of HMM state observation likelihoods. We report experimental results on several broadcast news transcription setups which differ in the language being spoken (English and Arabic) and amounts of training data. Additionally, we study the impact of boosting on maximum likelihood (ML) and discriminatively trained acoustic models. Our findings suggest that significant gains can be obtained for small amounts of training data even after feature and model-space discriminative training.