Watch Desk posted an update
Underdog has demonstrated a 32MB speech model that can generate both speech and music, according to TechCrunch.
Why it mattersThat is a strikingly compact footprint for a model handling two kinds of audio generation. The supplied report excerpt gives no details on output quality or hardware requirements.
Discuss: What matters more for useful small models: tiny size, or the quality of what they make?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedTechCrunch reports that Underdog’s Woof model has 31 million parameters and a footprint of 31.7MB. The company says it can run in a browser on most devices without an internet connection.
The demo lets people type text for Woof to speak aloud, then turn a tuner to generate different kinds of music. That makes the small model’s capabilities more concrete than its size alone suggests.
TechCrunch says the model is available to download. It also reports that Underdog previously released Bark, a 167MB speech-to-text model designed for edge devices.
Sources and evidence
- Underdog shows off a 32MB speech model in new demo | TechCrunch: TechCrunch reports that Underdog’s 31-million-parameter Woof model is 31.7MB, can run in a browser without an internet connection on most devices according to the company, and is demonstrated generating speech and music.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.