YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

T5-small IPA Phoneme-to-Text Transformer

Model Overview

This model converts sequences of International Phonetic Alphabet (IPA) phonemes into English text. It fine-tunes the T5-small architecture using a custom corpus of IPA–text pairs. The model extends prior work on phoneme-to-text translation by focusing on IPA rather than ARPABET, enabling broader multilingual phonetic compatibility and potential applications in child speech recognition and phonological analysis.

Metric Score
Character Error Rate (CER) 0.1505
BLEU 0.6065
chrF 76.29

Example

Input (IPA): ʃi kæn ˈɔlwɛɪz du ˈbɛtər
Output: she can always do better

Model Details

  • Base Model: T5-small
  • Task: Sequence-to-sequence phoneme → text translation
  • Tokenizer: SentencePiece (same as T5-small)
  • Training Data: Custom corpus of IPA–text pairs derived from transcribed English speech
  • Intended Use: Research on phonological modeling, ASR error correction, and speech pathology

Citation

If you use this model, please cite:

Graper, Z. (2025). Phoneme-to-Text Transformation using a Sequence-to-Sequence T5 Model. University of the Cumberlands, MSAI 699 Capstone Project.

License: CC BY-NC 4.0
Author: Zane Graper, 2025

Downloads last month
15
Safetensors
Model size
60.5M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support