Skip to main content
4h ago

Underdog showcases lightweight 32MB speech model in interactive demo

While major AI companies race to develop increasingly massive models that require immense processing power, researchers and startups are pivoting towa

2 min read9 views5 tags
Originally reported bytechcrunch

While major AI companies race to develop increasingly massive models that require immense processing power, researchers and startups are pivoting toward creating smaller models designed to run efficiently on edge devices, such as personal computers and mobile phones.

A startup named Underdog has recently released an interactive game featuring its new voice model, "Woof," which boasts a remarkably small size of just 32MB.

The game immerses users in a retro-tech environment complete with a typewriter and a CRT monitor. Users can interact with a desk lamp that produces a sound when switched, while the introduction warns visitors about "intrusive" cloud providers within the town—a pointed critique of cloud-based AI models.

Players can utilize a typewriter to type messages that the game will speak aloud and manipulate a music tuner to generate various audio tracks.

In a post on X, the company explained, "We made a game for our new frontier text-to-speech model, Woof. 31.7 MB, so small that it runs in your browser... all voices are Woof." This demonstration allows users to defeat "Darkcloud" to receive an invite.

The entire experience functions directly within a web browser using Underdog's speech generation model, which the company claims operates smoothly on most devices without requiring an internet connection. With 31 million parameters, the model is open-source and available for download.

Although unlikely to replace larger models, this lightweight solution is well-suited for power-efficient local assistant applications or the creation of personalized podcasts.

This demo arrives just weeks after Underdog launched "Bark," a 167MB speech-to-text model also optimized for edge devices and transcription tasks.

Sigil Wen, the founder of Underdog, explained his motivation to TechCrunch. Instead of relying on third-party notetakers like Granola, he sought to build a system where AI models run directly on users' devices, ensuring data ownership.

"I think we will be able to run most capable models of today on the devices you already own within months," Wen stated. "We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI."

Beyond Woof and Bark, Underdog has also released a 27-billion-parameter model capable of running on laptops and a waitlisted application positioned as a local alternative to Muse and Instinct.

#AI News#Underdog#Woof#Text-to-speech#Edge devices
ES
Editorial StaffEditor

The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.

View all posts
Reader feedback

What did you think of this story?

User Comments

Filter:
No comments yet. Be the first to comment!
Continue reading
View all news