voice-call-wasm

voice-call-wasm

Holds a spoken conversation inside your browser using three small models on your own CPU, with no audio sent anywhere once the models have downloaded.

Made by fusorlabs on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: voice-call-wasm is a static web page that lets you talk with an AI by voice, with all the computing done on your own CPU. Three separate models share the work instead of one speech-to-speech model. Each is compiled to WebAssembly with fusor-ai, and after one download they run with no network, so the source says nothing you say leaves the page.

Who it suits: People who want to try a private, on-device voice assistant, and developers curious about running speech models in a browser.

Industry | Who uses it | How they use it | Example
Research | Speech technology researchers | Study delay in a spoken exchange | Read the timing of each turn
Software | Web developers | Look at how models are bundled for WebAssembly | Browse the engine and model folders
Privacy | Privacy-minded users | Try a voice call that stays on the device | Talk to it after the models have downloaded
Education | Students of speech systems | See how listening, thinking and speaking are split | Follow the three stages in order

People who want a polished assistant with many skills can skip it, since this is a focused demo.

Where to use it:

• A web browser, by visiting the Space on Hugging Face
• A copy of the files on Windows, macOS or Linux, opened in a web browser
• Offline, after the models have been downloaded once

Use case | Who | What they get back
Have a voice call | Anyone | Spoken replies to what you say
Measure response time | Researchers | A delay figure for each turn, measured until the reply becomes audible

How to set it up:

• A web browser
• Python and pip, only to fetch a copy of the files
• A one-time download of the model bundle from the fusorlabs model repository

1. Install the Hugging Face hub package.
2. Download the Space with the hf command.
3. Move into the new folder.
4. Open index.html in a web browser.

pip install -U huggingface_hub
hf download fusorlabs/voice-call-wasm --repo-type space --local-dir voice-call-wasm
cd voice-call-wasm

How to use it: Online, visit the Space on Hugging Face and start a call from the page. With a local copy, open index.html in your browser and begin the same way.

Tool | What it does
Moondream Parakeet Redux | Writes down what you say while you are still speaking
LFM3-1.7B | Writes the answer to what you said
Phonon | Speaks each clause of the answer while the rest is still being written

Example: You say a short question into the microphone. The page transcribes it as you talk, drafts a reply, and starts speaking the first clause before the whole answer exists.

For newcomers and experts: If you are new, visit the Space and say one sentence. Waiting for the first download to finish is the main thing to expect, and after that the call needs no network.

If you know the field, the design is a pipeline of three models instead of a single speech-to-speech model. Streaming speech recognition feeds a 1.7B language model, and a 31M text-to-speech model speaks clause by clause. Every exchange gets a delay measurement that starts when you stop talking and ends when the reply first becomes audible.

You speak into the microphone
Parakeet Redux transcribes as you talk
LFM3-1.7B writes the reply
Phonon speaks it clause by clause

Good to know:

• Privacy: after the first download, the models run with no network access
• Model licenses: Parakeet Redux carries CC-BY-4.0, Phonon carries Apache-2.0, and LFM3-1.7B has its own LFM Open License v1.0
• License: mit for the page's own code
• Cost to run: none named
• Data it reaches: your microphone audio stays on the page, and the model files come from a Hugging Face repository

On Hugging Face, 13 November 2026

• Likes: 2
• Built with: static
• License: mit

Links

• Space: https://huggingface.co/spaces/fusorlabs/voice-call-wasm
• Files: https://huggingface.co/spaces/fusorlabs/voice-call-wasm/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.