Underdog unveils 32MB speech model for on-device audio generation

Traders Union
All News
Business & Wealth
Companies
Underdog 32MB audio model

Highlights

Underdog launched a browser demo using its new 32MB Woof voice model, enabling fully offline, local text-to-speech and audio generation capabilities.
The 31-million-parameter model targets edge AI applications, providing privacy-focused, device-owned audio tools with minimal cloud dependence for personalization and faster response.
Following a recent 167MB speech-to-text release, Underdog also introduced a 27-billion-parameter model for laptops and a waitlisted local alternative to Muse and Instinct.

Browser demo highlights local audio capabilities

As first reported by TechCrunch, Underdog has released a retro-themed browser game built around its new Woof voice model, which the company says can run on most devices entirely offline. The demo lets users type text into a virtual typewriter and hear it spoken back, while a separate music tuner generates different styles of audio within the same experience.

The game is set in a room with a CRT monitor, typewriter and desk lamp, and opens with a warning about “intrusive” cloud providers, a clear jab at cloud-based AI services. Underdog says the model has 31 million parameters and is open for anyone to download, underscoring its push for lightweight, locally run tools.

Edge AI strategy targets privacy and device ownership

Although a model of this size is unlikely to replace much larger systems, it could still support local assistant functions or personalized podcast creation on phones and computers. That positions the product within a growing segment of edge AI focused on lower infrastructure needs, faster local response and reduced dependence on remote servers.

This demo follows Underdog’s launch, weeks earlier, of a 167 MB speech-to-text model for transcription and dictation on edge devices. Founder Sigil Wen told TechCrunch he began building the app because he wanted an alternative to note-taking tools such as Granola, with AI models that run on users’ devices and allow them to keep control of their data.

Beyond the two voice models, Underdog has also released a 27-billion-parameter model that it says can run on laptops, along with a waitlisted app positioned as a locally hosted alternative to Muse and Instinct.

Lenovo’s upcoming Tech World presentation at CES 2027 will spotlight its Hybrid AI strategy, spanning consumer devices and enterprise systems with an emphasis on privacy, security, and data control. Our earlier article noted that Lenovo plans to showcase AI innovations across PCs, smartphones, and wearables alongside enterprise AI infrastructure, positioning hybrid and locally controllable deployments as a key part of its broader AI push.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.