Learning Discriminative Features in Sequence Training without Requiring Framewise Labelled Data
Lists
Only you see this
Abstract
In this work, we try to answer two questions: Can deeply learned features with discriminative power benefit an ASR system’s robustness to acoustic variability? And how to learn them without requiring framewise labelled sequence training data? As existing methods usually require knowing where the labels occur in the input sequence, they have so far been limited to many real-world sequence learning tasks. We propose a novel method which simultaneously models both the sequence discriminative training and the feature discriminative learning within a single network architecture, so that it can learn discriminative deep features in sequence training that obviates the need for presegmented training data. Our experiment in a realistic industrial ASR task shows that, without requiring any specific fine-tuning or additional complexity, our proposed models have consistently outperformed state-of-the-art models and significantly reduced Word Error Rate (WER) under all test conditions, and especially with highest improvements under unseen noise conditions, by relative 12.94%, 8.66% and 5.80%, showing our proposed models can generalize better to acoustic variability.
BibTeX
@inproceedings{wang2019learning,
author = {Jun Wang and Dan Su and Jie Chen and Shulin Feng and Dongpeng Ma and Na Li and Dong Yu},
title = {{Learning Discriminative Features in Sequence Training without Requiring Framewise Labelled Data}},
booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
year = {2019},
doi = {10.1109/icassp.2019.8683088},
}| Boosting Noise Robustness of Acoustic Model via Deep Adversarial Training | arXiv · 2018 |
| A Discriminative Decoder for the Recognition of Phoneme Sequences | Infoscience · 2005 |
| Performance of Discriminative HMM Training in Noise | Int. J. Comput. Linguistics Chin. Lang. Process · 2007 |
| SAR-Net: A End-to-End Deep Speech Accent Recognition Network. | arXiv · 2020 |
| Frustratingly Easy Noise-aware Training of Acoustic Models | arXiv · 2020 |
| Additive Phoneme-Aware Margin Softmax Loss for Language Recognition | arXiv · 2021 |