fil-lip: Filipino Lipreading Research Revamp

Problem

The baseline model accuracy was limited and the dataset/model assumptions didn’t align well with Filipino speech patterns.

Constraints

• Data quality and labeling effort are major bottlenecks
• Need measurable improvements over baseline accuracy
• Must keep experiments reproducible and well-tracked

Solution

Planned an iterative research upgrade: dataset cleanup, augmentation, model architecture experiments, and evaluation improvements focused on Filipino language compatibility.

Impact

• Improved experimental rigor and reproducibility
• Created a clearer roadmap for accuracy gains
• Positioned the project for future publication-quality results