Underdog demos a tiny 32MB speech model that generates both speech and music
TechCrunch AI · rss · 2026-11-06
Startup Underdog showed off a new demo of a 32MB speech model that, despite its tiny footprint, can generate both speech and music. The size implies on-device/edge deployment potential, a notable compression of the barriers to running audio generation locally.
More from Multimodal
- SenseTime Open-Sources SenseNova 6.8 Code2Video: Prompt-to-Video With Editable Code — FellMentKE · 2026-11-06
- Korean team's AI-made K-pop MV "Higher" looks industry-grade — Beginning_Victory597 · 2026-11-06
- RelayVSR is the new king of video upscalers, 20x faster than SeedVR2 — CeFurkan · 2026-11-06
- There is no faithful AI upscale: hallucinated details risk your integrity — round · 2026-11-06
- AI-generated 'naked mole-rat driving a convertible, 1960' stuns with photoreal texture — Megalania34 · 2026-11-06
- Stunning pure H3-generated fantasy short shows how far open video pipelines have come — AdmirableObjective76 · 2026-11-06