Filipino g2p model predicts phonemes and stress using sentence context
Towards Stress-Aware Sentence-Level Filipino G2P With Weakly-Supervised ByT5 Fine-Tuning
Computation and Language
Summary
Turning written words into sounds is usually simple for Filipino because the spelling matches the sounds closely. However, figuring out the right stress in sentences is harder and needs context beyond single words. The authors created a method that uses a special language model fine-tuned on sentence-level data, even though this data doesn’t usually have detailed sound information. Their approach improved the accuracy of predicting how Filipino words sound, especially with stress patterns, and it can help tell apart words that look the same but sound different.
What this means in practice
- •For speech synthesis developers: Improve Filipino text-to-speech systems by accurately predicting stress patterns and phonemes using sentence-level models.
- •For language technology engineers: Enhance Filipino speech recognition and pronunciation tools by incorporating better phoneme and stress predictions from sentence-level context.
Authors
Lorenz Bernard Marqueses, Paulo Grane Gabriel Silva, Chastine Cabatay, Ericson Adler Tan, Ann Franchesca Laguna
Abstract
Grapheme-to-phoneme conversion (G2P) refers to the task of converting a sequence of graphemes to a corresponding sequence of phonemes. While Filipino G2P is fairly straightforward due to its shallow orthography, the inclusion of prosodic features such as stress adds a layer of complexity that requires sentence-level context instead of single-word inputs. However, sentence-level data for Filipino typically do not include phoneme transcriptions, posing a challenge for training G2P models. As such, we investigate how to obtain sentence-level phoneme data for Filipino using available data and compare the resulting models with multilingual word-level G2P as well as measure how accurately they predict stress marker position for Filipino. We propose fine-tuning a ByT5-based model, pre-trained on multilingual word-level G2P data, on three sentence-level G2P datasets annotated with an LLM-assisted pipeline guided by data from Wiktionary. This approach produces models that perform well on the G2P task, achieving at best around 0.54% PER and 2.50% CER, a significant decrease compared to base model PER at around 19.74%, on a manually-corrected test set. The model is able to correctly classify most of the main stress classes in Filipino, but struggles particularly with malumi words. We show that a ByT5-based model performs well at sentence-level Filipino G2P and offers strong potential for Filipino homograph disambiguation.