Feature Request: Sparse BakedData

[mmechtley]

Currently BakedData stores the weight of every phoneme for every baked frame. However, most phonemes in a frame have weight 0, and for large amounts of phonemes this can lead to a lot of extra data (with BakedData rivaling uncompressed wav audio in file size). BakedData could instead store only the phonemes with non-0 weight each frame, preserving the full list/order of phonemes as a separate top-level field, significantly reducing file size.