Effect of retaining syllable-boundary markers

Determine whether retaining the syllable-boundary markers produced by the Kurdish grapheme-to-phoneme converter improves the performance of the Central Kurdish text-to-speech models.

Background

The released preprocessing pipeline removes syllable-boundary markers from the grapheme-to-phoneme output before the text reaches the F5-TTS model, even though the marker symbol is present in the model vocabulary.

The review identifies the removal as a potentially consequential preprocessing choice but reports that the available evidence does not establish whether preserving the markers would improve synthesis quality. Resolving this would clarify whether the discarded prosodic information is useful for Central Kurdish speech synthesis.

References

Whether retaining it would have improved results is not something this study can determine.

Assessing the Reusability of Public Speech Resources for Low-Resource Languages: A Central Kurdish Case Study  (2609.11246 - Asadpour, 10 Sep 2026) in Section 10, paragraph beginning “One choice can be established exactly”