---
license: apache-2.0
datasets:
- EarthSpeciesProject/NatureLM-audio-training
base_model:
- Qwen/Qwen3-Omni-30B-A3B-Instruct
pipeline_tag: audio-classification
tags:
- animal
- biology
- interspecies-communication
- nature
- species
---

# AniMUL-v1
AniMUL is a model for interspecies communication.

# Use
AniMUL can do species classification from audio files.
It can generally describe the sounds of an audio file.

It does *not* generate audio of species.

# Upstream
AniMUL is a fine tune of Alibaba's Qwen3-Omni model.

* https://huggingface.co/Qwen/Qwen3-Omni-30B-A3B-Instruct

AniMUL uses the Earth Species Project's NatureLM dataset
of 26 million audio-text pairs.

* https://www.earthspecies.org/
* https://huggingface.co/datasets/EarthSpeciesProject/NatureLM-audio-training

This project is unofficial and not related to the upstream projects.

# Source Code
Source code for the fine tuning, test inference, and
a web interface is available:

* https://spacecruft.org/deepcrayon/AniMUL
* https://spacecruft.org/deepcrayon/AniMUL-server
* https://spacecruft.org/deepcrayon/AniMUL-stream

Since the model is a fine tune of Qwen3-Omni, you can refer to that
model for other deployment details.

# License
The AniMUL model itself is under the Apache 2.0 license.

The upstream Qwen3-Omni model is under the Apache 2.0 license.

Data in the dataset from the Earth Species Project that was used for training
is under a variety of licenses, including non-open source licenses such
as Creative Commons non-commercial (NC) licenses.
The Earth Species Project dataset page has this note:

"Due to its composite nature, NatureLM-audio-training is subject to multiple licenses.
Individual samples have the 'license' field indicating the specific license for that sample.
The dataset is not intended for commercial use,
and users should adhere to the licenses of the individual datasets."

# Developer
Jeff Moe <moe@spacecruft.org>

Loveland, Colorado, USA