Key points
- Underdog Labs released Woof, a 32MB speech and music generation model with 31 million parameters.
- The model operates entirely within the browser without requiring an internet connection or cloud processing.
- This release follows the launch of Bark, a 167MB speech-to-text model designed for edge device transcription.
Underdog Labs has demonstrated that a 32MB model can generate both speech and music locally, according to a report by TechCrunch.reported The startup released an interactive browser game built with its new voice model, Woof, which features a retro tech theme.Underdog Labs shows a tiny computer program can make voices and music on your device without needing the internet.Underdog Labs demonstrates a 31M-parameter voice model capable of offline text-to-speech and music synthesis via a browser-based demo. Users can type text into a virtual typewriter, and the model speaks it back. Rotating a virtual tuner generates different kinds of music. All audio generation occurs within the browser without any internet connection.
The claim
The company states that Woof, which contains 31 million parameters, runs on most devices without cloud reliance. The demo includes a CRT monitor and desk lamp that makes sound when clicked. The intro warns users about intrusive cloud providers, serving as a jab at cloud-based models. Underdog says the model is open for everyone to download.
The response
Sigil Wen, founder of Underdog, told TechCrunch he built the system to let users own their data. He stated he wants to train open, American-made models that are capable and allow private AI. Wen predicted that most capable models of today could run on existing devices within months.
The bottom line for teams
While such a small model likely won’t replace bigger models, it may drive local assistant experiences or create personalised podcasts. Underdog also released a 27-billion-parameter model for laptops and a waitlisted app positioned as a locally-hosted alternative to Muse and Instinct.reported
What the paper shows
The research tests a 31-million-parameter model’s ability to generate speech and music offline.reported Reported results show text-to-speech and music generation in a browser game.reported Stated limitations include the model’s small size, which likely prevents replacement of larger models. The material does not provide independent benchmark metrics for audio quality or latency.
Why it matters
- Evaluators can test local inference capabilities using a freely downloadable 32MB model without cloud costs.
- The 31-million-parameter count offers a baseline for assessing trade-offs between size and audio fidelity.
- Offline operation proves feasible for basic voice tasks, informing edge-device deployment strategies.
- The comparison to cloud providers highlights a growing market segment for privacy-focused, local AI tools.
Topics: models, edge AI, speech synthesis, Underdog Labs, open source
Updated
Then / now / next
AI companies focused on building bigger models requiring massive processing power.
Underdog demonstrates a 32MB model generating speech and music offline in a browser.
The company has a waitlisted app positioned as a locally-hosted alternative to other services.
Story timeline
- Then AI companies focused on building bigger models requiring massive processing power.
- 06 Nov 2026 Underdog shows off a 32MB speech model in new demo
- Now Underdog demonstrates a 32MB model generating speech and music offline in a browser.
- Next The company has a waitlisted app positioned as a locally-hosted alternative to other services.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
What is the size of Underdog's new speech model?
The Woof model is 32MB large and contains 31 million parameters.
Can the model run without an internet connection?
Yes, the model runs within the browser on most devices without any internet connection.
What other models has Underdog released recently?
Underdog launched Bark, a 167MB speech-to-text model, and a 27-billion-parameter model for laptops.
Connected stories
Reading receipt
- Underdog Labs released Woof, a 32MB speech and music generation model with 31 million parameters.
- The model operates entirely within the browser without requiring an internet connection or cloud processing.
- This release follows the launch of Bark, a 167MB speech-to-text model designed for edge device transcription.