Reduce component dependence in Thai pause-placement scoring
Reduce the dependence of Thai pause-placement scoring on Thai word segmentation, CTC forced alignment, and the handling of code-switched and verbalized spans, thereby improving the accuracy and generalizability of the pause detector.
References
The scorer is a composite of several components -- Thai word segmentation, CTC forced alignment, and the handling of code-switched and verbalized spans -- and its accuracy is bounded by all of them. Segmentation decides what counts as word-internal, and the two tokenizers we combine still agree on some compounds that a listener would treat as two words; the character aligner contributes timing jitter, and it has no vocabulary for digits or Latin spans, whose spoken Thai form it therefore cannot place. The guards described below remove the failure modes we identified, but the detector is not exact. Its rates are most reliable as comparisons between systems measured with the same instrument, and reducing these component dependencies is left to future work.