Inference of SunoAI's bark model in pure C/C++ using ggml.
The main goal of bark.cpp is to synthesize audio from a textual input with the Bark model.
Bark has essentially 4 components:
- Semantic model to encode the text input
- Coarse model
- Fine model
- Encoder (quantizer + decoder) to generate the waveform from the tokens
- Quantization
- FP16
- Swift package for iOS devices