DuplexJev Speech to Decisions

DuplexJev Speech to Decisions

Reads one speech clip and returns ten decisions about it, such as turn state, barge-in and emotion, in a single pass with no transcription. For voice agent developers and speech researchers.

Made by adventists-ai 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: DuplexJev Speech to Decisions is a browser demo of the DuplexJev model. It takes one speech clip and returns ten decisions about it, such as turn state, barge-in, emotion and non-verbal sounds. It does this in one forward pass, with no speech recognition and no decoding. The page calls the free DuplexJev trial API.

Who it suits: Developers building voice agents or call tools who want to see what a speech-first model can tell from raw audio, and researchers who study conversational audio.

Industry | Who uses it | How they use it | Example
Voice AI | Voice agent developers | Check how a model reads turn-taking and interruptions | Trying a clip where one speaker cuts in
Customer service | Call tool builders | See emotion and non-verbal cues without transcribing | Testing a recorded call clip
Research | Speech researchers | Look at several decisions from a single pass | Comparing the ten outputs for one clip

You can skip it if you need a transcript of what was said, since the source says it uses no speech recognition.

Where to use it:

• A web browser, on the Space's own page at Hugging Face
• A browser on your own computer, after downloading the files with Python on Windows, macOS or Linux

Use case | Who | What they get back
Read a speech clip | Voice developers | Ten decisions about the clip
Try the examples | Anyone curious | Results for the sample clips in the examples folder

How to set it up:

• A browser, if you only use the Space online
• Python and pip, only if you want the files on your own computer
• A connection, since the page calls the DuplexJev trial API

1. Open the Space's page on Hugging Face to use it online. Nothing needs installing.
2. To keep the files, install the huggingface_hub package and download the Space's files.
3. Move into the new folder and open index.html in a web browser.

pip install -U huggingface_hub
hf download adventists-ai/DuplexJev-demo --repo-type space --local-dir DuplexJev-demo
cd DuplexJev-demo

How to use it: Open the page and give it a speech clip, or pick one of the examples if the page offers them. The page sends the clip to the trial API and shows the ten decisions that come back. After a local download, open index.html in your browser instead.

Example: You open the page and choose an example clip where one speaker interrupts another. The page shows the ten decisions for it, including what the turn state is and whether it detects barge-in.

For newcomers and experts: If you are new to this, try an example clip first. Each decision tells you something about the moment in the audio, such as who is speaking or whether someone cut in.

For experts, a single forward pass over the raw audio yields all ten outputs, so nothing is transcribed or decoded first. The trade-off is that the page depends on the hosted trial API, so the model itself does not run in your browser.

Open the page
Pick or record a speech clip
The clip goes to the trial API
One forward pass runs on the raw speech
Ten decisions show on the page

Good to know:

• Built as static HTML, with index.html and style.css
• License: cc-by-nc-4
• Cost to run: the page calls a free trial API, and no key is named
• Data it reaches: your speech clip, sent to the DuplexJev trial API
• Not for: commercial use, since the license is cc-by-nc-4

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: static
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/adventists-ai/DuplexJev-demo
• Files: https://huggingface.co/spaces/adventists-ai/DuplexJev-demo/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.