OWSM v4 by espnet

OWSM v4 by espnet

Runs the OWSM v4 open speech model on audio you give it, in a Gradio demo you can open in a browser. For people who want to try or compare speech models.

Made by espnet 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: This Space is a browser demo of OWSM v4, an open Whisper-style speech model from the ESPnet team. It runs the model named espnet/owsm_v4_medium_1B on audio you give it, so you can try the model before building on it. The same demo also lives at a newer address, and this older one stays because other pages link to it.

Who it suits: Speech researchers, students and developers who want to see what an open speech model does with their own audio. It also suits anyone who wants to compare two models side by side.

Industry | Who uses it | How they use it | Example
Research | Speech researchers | Try the model on their own recordings before using it in a study | Run one recording here, then run it through the encoder-only sibling and compare
Education | Students of speech technology | Watch an open model work without setting anything up | Open the Space page and give it a short clip
Software | Developers | Check what the model does before wiring it into an app | Run a local copy with python app.py

If you need a finished service for daily work, you can skip it, since this is a demo.

Where to use it: The demo runs in these places:

• A web browser: open the Space page on Hugging Face and use it straight away.
• Windows, macOS or Linux: run a copy on your own computer with Python 3.12.

Use case | Who | What they get back
Try the model online | Anyone curious about open speech models | The model's output for the audio they give it
Compare two models | Researchers | A way to run the same audio through this demo and its encoder-only sibling
Run it locally | Developers | The same Gradio demo on their own computer

How to set it up: You need a few things first:

• Python 3.12
• The huggingface_hub package, which the first command installs
• The packages listed in requirements.txt

The steps are short. First install the Hugging Face tools. Then download the Space into a folder, move into it, install Gradio at the version the Space uses, install the other packages and start the app.

1. Install the Hugging Face package.
2. Download the Space into a folder called OWSM_V4_Demo.
3. Move into that folder.
4. Install Gradio 6.27.0.
5. Install the packages from requirements.txt.
6. Start the app.

pip install -U huggingface_hub
hf download espnet/OWSM_V4_Demo --repo-type space --local-dir OWSM_V4_Demo
cd OWSM_V4_Demo
pip install gradio==6.27.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space page on Hugging Face, or start the local copy with the last command above. Gradio then shows a web page where you give the demo audio to work on.

python app.py

Tool | What it does
hf download | Copies the Space's files into a folder on your computer
python app.py | Starts the demo on your own computer

Example: You open the Space, give it one short recording and read what the OWSM v4 model produces for it. Then you run the same recording through the encoder-only sibling Space to see how the two models differ.

For newcomers and experts: If you are new to speech models, open the Space page and try one short recording first. You do not need to install anything for that. The local copy is only worth the setup once you want to change or study the demo.

If you know the field, the demo serves the model espnet/owsm_v4_medium_1B, and the project's INTERSPEECH paper describes improving open Whisper-style models through data scaling and cleaning. Its sibling serves an encoder-only model with CTC and offers the same tasks. Changes to the demo code are made in the ESPnet repository and published to both Spaces.

You open the Space or run python app.py
A Gradio page takes your audio
The OWSM v4 model works on it
The result shows on the page

Good to know: The card does not list supported languages, audio formats or length limits.

• License: apache-2, as the Space lists it.
• Cost to run: none named. The hosted Space runs on Hugging Face hardware of the zero-a10g type.
• Data it reaches: the audio you give it. A local copy also downloads the Space's files from Hugging Face.
• Not for: editing the demo here. The card says changes belong in the ESPnet repository, not in this Space.

On Hugging Face, 13 November 2026

• Likes: 12
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/espnet/OWSM_V4_Demo
• Files: https://huggingface.co/spaces/espnet/OWSM_V4_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.