A generative speech model for daily dialogue.

Identify speakers with stable voice timbre.

Infer speaker embeddings from given audio.