fufuchiu / speechloom

End-to-end speech model research: waveform encoding, causal decoding, codec tokens, joint losses and streaming.