Updated

Underdog demonstrated a new 32MB speech model that can generate both speech and music, TechCrunch reported. The company emphasized that the model is small despite its range of outputs. The demo is an early look at compact generative audio models.