Semi-supervised Learning with Sparse Autoencoders in Phone Classification
This work addresses the challenge of improving speech recognition accuracy with limited labeled data, though it appears incremental as it builds on existing semi-supervised approaches.
The paper tackled the problem of acoustic modeling for automatic speech recognition by proposing a semi-supervised learning method that uses both labeled and unlabeled data simultaneously, achieving competitive error rates compared to state-of-the-art techniques in phoneme classification on the TIMIT database.
We propose the application of a semi-supervised learning method to improve the performance of acoustic modelling for automatic speech recognition based on deep neural net- works. As opposed to unsupervised initialisation followed by supervised fine tuning, our method takes advantage of both unlabelled and labelled data simultaneously through mini- batch stochastic gradient descent. We tested the method with varying proportions of labelled vs unlabelled observations in frame-based phoneme classification on the TIMIT database. Our experiments show that the method outperforms standard supervised training for an equal amount of labelled data and provides competitive error rates compared to state-of-the-art graph-based semi-supervised learning techniques.