Text to speech

Underdog Woof

A tiny English speech model with two voices, Juno and Mira, only 40.6 MB. It reads phone numbers, dates, prices and codes the way people say them. It runs on an ordinary laptop CPU and needs only numpy, with no graphics card, server or key: by the authors measurements 13.8 times faster than real time on one thread and about 402 MB of memory. On their hard-sentences test the word error rate is 1.42%; the authors compare it with Fish Audio Drama 3 and claim a size 1,022 times smaller and 25% fewer errors. English only. The license is non-commercial, so it cannot be used in paid products without separate permission.

Developer
ConwayResearch (Underdog), not stated
First release
Nov 2026
Latest release
Nov 2026
Sizes
40.6 MB (the whole file with two voices)
License
Non-commercial onlyUnderdog Non-Commercial 1.0: non-commercial use only
Russian
Not supported
Running
On your own serverAlso runs without a GPU
Industries
Media and production
Deploy on your server Hugging Face

What it does

  • offline voicing of English texts on a weak computer
  • prototypes of voice interfaces without a cloud
  • personal and educational projects without commerce

Where it is used

educationresearch projectsprototypes of voice assistants

Hardware requirements

LaptopLaptop or regular PC, up to 8 GB of VRAM — smaller versions
fits
1 GPUOne GPU with 16–80 GB — mid-size versions
no versions
ClusterServer with several GPUs — flagship versions
no versions

Versions

  1. Underdog Woof (Juno и Mira)

How to run it

On your own serverWeights are downloaded from Hugging Face and served with vLLM (load and API) or llama.cpp (modest hardware). The model then runs in a closed network with no per-request fees.Hugging Face
How much hardware you needCalculate the VRAM for the model size, context length and number of concurrent requests.Open the hardware calculator

I can set this up end to end: pick the model size, deploy it on your server and connect it to your systems.

Frequently asked questions

Can Underdog Woof be used in a commercial project?

No. License: Underdog Non-Commercial 1.0: non-commercial use only. A commercial product needs a different model or a separate agreement with the rights holder.

What hardware does Underdog Woof need?

At minimum: Laptop or regular PC, up to 8 GB of VRAM — smaller versions. Some versions also run on an ordinary CPU, without a GPU. You can calculate the exact VRAM for your model size and context in the hardware calculator.

Does Underdog Woof support Russian?

No. The developer documentation lists the supported languages and Russian is not among them.

Where can I download Underdog Woof and what does it cost?

The Underdog Woof weights are open and free to download. You only pay for the hardware it runs on and for the setup. Source links are at the bottom of this page.

How I deploy it for clients

  1. SelectionI pick the model size for your task and hardware and test it on your examples.
  2. DeploymentI deploy it on your server or in a closed network and provide an API.
  3. Fine-tuningI fine-tune it on your data (LoRA) or connect a knowledge base — whichever is cheaper for the task.
  4. IntegrationI connect it to your CRM, ERP, bot, website or team chat and set up monitoring.

Similar models

Text to speechKokorohexgrad (independent developer) · not disclosedCommercial use allowed

A tiny speech synthesis model (82M) that sounds on par with large ones. Runs on a regular CPU; English and a few other languages, no Russian.

DetailsText to speechFish Speech / OpenAudioFish Audio · USA / ChinaNon-commercial only

Speech synthesis with voice cloning and emotion control in 80+ languages, including Russian. Quality is close to paid services, but the weights are for research only. The new Drama 3 (35B, 3B active) voices scenes with several characters (up to 16 voices in one text), understands tags such as a sigh or a laugh, over 80 languages, Russian is in the list.

DetailsText to speechZonosZyphra · USACommercial use allowed

Speech synthesis with voice cloning and fine control over emotion, speed and pitch. In November the compact ZONOS3-Mini came out (Apache 2.0): only 100M active parameters, runs in real time on a phone and laptop, including right in the browser through WebGPU; it clones a voice from a sample or creates a character by age, accent and timbre, and controls emotion and pace down to individual words.

Details

Source: huggingface.co/ConwayResearch/Underdog-Woof. Data checked against the model card on 22 Sep 2026. Have a lawyer review the license before commercial launch.