Alignment Entropy Regularization
This work addresses alignment uncertainty in ASR, offering a method to simplify decoding and enhance alignment, but it is incremental as it builds on existing training criteria.
The paper tackles the problem of multiple possible time alignments in automatic speech recognition by using entropy regularization to reduce the model's uncertainty over alignments, resulting in simpler decoding without sacrificing word error rate and improved alignment quality.
Existing training criteria in automatic speech recognition(ASR) permit the model to freely explore more than one time alignments between the feature and label sequences. In this paper, we use entropy to measure a model's uncertainty, i.e. how it chooses to distribute the probability mass over the set of allowed alignments. Furthermore, we evaluate the effect of entropy regularization in encouraging the model to distribute the probability mass only on a smaller subset of allowed alignments. Experiments show that entropy regularization enables a much simpler decoding method without sacrificing word error rate, and provides better time alignment quality.