Startup Underdog releases small voice models Woof and Bark for local offline operation
Startup Underdog released the open model Woof (32 MB, 31 million parameters) for speech generation and earlier the model Bark (167 MB) for speech transcription. Both run locally on the device without an internet connection.
Startup Underdog has released the voice model Woof, 32 MB in size with 31 million parameters, designed for generating speech directly on the user's device without an internet connection. The company presented the model through an interactive browser-based game with a retro theme — users can type any text on a virtual typewriter keyboard and the model will read it aloud, or turn a tuning dial to generate different types of music. The Woof model is open for download.
A few weeks earlier, Underdog released the model Bark, 167 MB in size, designed for speech-to-text transcription and dictation on edge devices, again without requiring a cloud connection.
According to Underdog founder Sigil Wen, the motivation was the effort to create their own system of AI models that run on users' devices and allow them to own their data, instead of using tools like Granola. According to the manifesto, Sigil Wen wants to train open, US-made models that are capable and enable people to use AI privately.
According to the article, Underdog has also released a model with 27 billion parameters that runs on laptops, and is preparing an app with a waitlist, which the company positions as a locally hosted alternative to Muse and Instinct.
Why it matters
Small open voice models allow assistants, dictation, or speech transcription to run directly on a phone or computer without sending data to the cloud, which reduces both operating costs and privacy-related risks. For developers, this opens up the possibility of building offline voice features without dependence on cloud providers, although according to the source, these small models cannot yet match the performance of large cloud models.
Release card
Woof
Underdog
- Specifications
- 31 million parameters
- Inputs
- text as input, generated speech and music as output
- Availability
- The model is open for download and runs locally on the user's device without an internet connection.
- generating speech from any text directly on the device without internet
- generating different types of music using tuning
- according to the source, it may be sufficient for local assistant experiences or creating personalized podcasts
- according to the source, such a small model will likely not replace larger models
According to the source, the model positions itself against cloud AI providers, which the article describes as "intrusive," and instead runs locally on the device without an internet connection.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
Two audiences, two different impacts
What this means
For individuals
Users can try speech generation directly in the browser without an internet connection and without sending data to a third party, or download the model and use it in their own local projects.
For a business
Companies developing voice assistants or transcription tools can deploy small open models directly on the user's device, avoiding both the costs and risks associated with sending data to the cloud; Underdog is also preparing its own app as a locally hosted alternative to Muse and Instinct.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.