---
title: OWSM v4
emoji: 🌍
colorFrom: blue
colorTo: pink
sdk: gradio
python_version: "3.12"
sdk_version: 6.27.0
app_file: app.py
pinned: false
license: apache-2.0
short_description: The same demo as espnet/owsm-v4, kept at this older address
tags:
  - espnet
  - owsm
  - automatic-speech-recognition
  - speech-translation
  - language-identification
models:
  - espnet/owsm_v4_medium_1B
---

# OWSM v4

The same demo as [`espnet/owsm-v4`](https://huggingface.co/spaces/espnet/owsm-v4):
the same `app.py` and `requirements.txt`, maintained in the ESPnet repository at
[`egs2/owsm_v4/s2t1/demo`](https://github.com/espnet/espnet/tree/master/egs2/owsm_v4/s2t1/demo).

This address is older and has been linked from elsewhere, so it is kept rather
than retired. **Changes belong in the repository**, not in either Space; both
are published from that directory with

```sh
hf upload espnet/owsm-v4        egs2/owsm_v4/s2t1/demo . --repo-type space
hf upload espnet/OWSM_V4_Demo   egs2/owsm_v4/s2t1/demo . --repo-type space
```

(this card being the one file that differs, since it says what you are reading.)

The encoder-only sibling is [`espnet/owsm-ctc-v4`](https://huggingface.co/spaces/espnet/owsm-ctc-v4),
from [`egs2/owsm_ctc_v4/s2t1/demo`](https://github.com/espnet/espnet/tree/master/egs2/owsm_ctc_v4/s2t1/demo).
It offers the same tasks with `owsm_ctc_v4_1B`, so the two models can be
compared on the same audio.

## Citation

```bibtex
@inproceedings{owsm-v4,
  title={{OWSM} v4: Improving Open Whisper-Style Speech Models via Data Scaling and Cleaning},
  author={Yifan Peng and Shakeel Muhammad and Yui Sudo and William Chen and Jinchuan Tian and Chyi-Jiunn Lin and Shinji Watanabe},
  booktitle={Proc. INTERSPEECH},
  year={2025},
}
```