Sequence-Level Unsupervised Training in Speech Recognition: A Theoretical Study
This work addresses the challenge of training speech recognition models without paired data, which is incremental as it builds on existing unsupervised methods with a theoretical analysis.
The paper tackles the problem of unsupervised speech recognition by developing a theoretical framework to determine when and how it can succeed, deriving a classification error bound and validating it in simulations, and proposing a single-stage sequence-level cross-entropy loss as a result.
Unsupervised speech recognition is a task of training a speech recognition model with unpaired data. To determine when and how unsupervised speech recognition can succeed, and how classification error relates to candidate training objectives, we develop a theoretical framework for unsupervised speech recognition grounded in classification error bounds. We introduce two conditions under which unsupervised speech recognition is possible. The necessity of these conditions are also discussed. Under these conditions, we derive a classification error bound for unsupervised speech recognition and validate this bound in simulations. Motivated by this bound, we propose a single-stage sequence-level cross-entropy loss for unsupervised speech recognition.