Documenting the complete process of building generative audio models from scratch. Every experiment, success, and failure.
Experiment #001
Audio Compression — Building a neural audio codec to represent sound as discrete tokens.
Active — 40% complete
Experiment #002
Perceptual Loss Functions — Exploring loss functions that optimize for human perception.
Planned
Experiment #003
Vector Quantization — Discrete representation of audio for generative modeling.
Planned
Experiment #004
Hierarchical Representation — Multi-scale audio encoding for richer generation.
Planned