WFL Model for ASR/Forced Alignment

This release includes a WFL model to provide base labels for the dataset's phoneme system. It is trained with this dataset, along with extra data from my sisters.

This model is licensed under the MIT License.

08-25-2026: Initial release
08-28-2026: Switch to a medium model for better accuracy
08-29-2026: Freeze encoder completely for better generalization (hypothetically). Include best and latest model.

Dataset Download

tiny-tagalog-dataset-unsegmented.7z contains the unsegmented wave files along with the original HTK mono labels.

tiny-tagalog-dataset-segmented.7z contains the segmented wave files along with .ds files. It contains phoneme timing and MIDI note information. The phoneme timing may not be up to par to standard.