→ In this final session - you shall be introduced to some of our team idea (which is not finish - but hope you may continued)

→ thank you for Tu for contributing the idea ☺️

Refined Idea 1 (MDD with MAS + Multi-Loss)

Refined Idea 2: Accent-Specific Phoneme Modeling for MDD

Refined Idea 3: Reference-Pool Rescoring for MDD (RAG-inspired)

Refined Idea 4: Cross-Lingual Voice Conversion as Data Augmentation for MDD

• Idea 3 is about adding a retrieval-style calibration layer post-MDD training.
• Idea 4 is about cross-lingual data augmentation via voice conversion to create synthetic mispronunciations for pretraining/fine-tuning.

Long-term Project (building the Web for data-campaign for collecting Vietnamese-L2 dataset)