Owls 4B 180K

🚀 OWLS: Open Whisper-style Large-scale neural model Suite

OWLS is a collection of Whisper-style models. It aims to assist researchers in understanding the scaling properties of speech models. These models, with parameter counts ranging from 0.25B to 18B, are trained on up to 360K hours of data.

🚀 Quick Start

Prerequisites

• Ensure espnet is installed: pip install espnet

Usage

You can use this model in your projects with the following code:

# make sure espnet is installed: pip install espnet
from espnet2.bin.s2t_inference import Speech2Text

model = Speech2Text.from_pretrained(
  "espnet/owls_4B_180K"
)

speech, rate = soundfile.read("speech.wav")
speech = librosa.resample(speech, orig_sr=rate, target_sr=16000) # make sure 16k sampling rate
text, *_ = model(speech)[0]

✨ Features

• Developed using ESPnet, supporting multilingual Speech Recognition and Translation.
• Part of the OWSM project, aiming to create fully open speech foundation models with public data and open - source toolkits.
• The model in this repo has 4.66B parameters in total and is trained on 180k hours of public speech data.
• Supports the following speech - to - text tasks:
  • Speech recognition
  • Any - to - any - language speech translation
  • Utterance - level alignment
  • Long - form transcription
  • Language identification

📦 Installation

Ensure espnet is installed in your environment:

pip install espnet

💻 Usage Examples

Basic Usage

# make sure espnet is installed: pip install espnet
from espnet2.bin.s2t_inference import Speech2Text

model = Speech2Text.from_pretrained(
  "espnet/owls_4B_180K"
)

speech, rate = soundfile.read("speech.wav")
speech = librosa.resample(speech, orig_sr=rate, target_sr=16000) # make sure 16k sampling rate
text, *_ = model(speech)[0]

📚 Documentation

OWLS models

Paper

OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models

📄 License

This project is licensed under the CC - BY - 4.0 license.

📄 Citations

@article{chen2025owls,
  title={OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models},
  author={Chen, William and Tian, Jinchuan and Peng, Yifan and Yan, Brian and Yang, Chao - Han Huck and Watanabe, Shinji},
  journal={arXiv preprint arXiv:2502.10373},
  year={2025}
}