This model converts sequences of ARPABET phonemes into readable English text.
It fine-tunes the T5-small model using a phoneme–text dataset derived from BookCorpus.
Metric
Score
Character Error Rate (CER)
0.1377
BLEU
0.6299
chrF
77.72
Example:
Input: SH IY K AE N AO L W EY Z D UW B EH T ER
Output: she can always do better
See full details in the paper: Phoneme-to-Text Transformation using a Sequence-to-Sequence T5 Model.
License: CC BY-NC 4.0
Author: Zane Graper, 2025
