notesum.ai

Published at November 11

NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics

cs.SD

cs.AI

cs.LG

eess.AS

Released Date: November 11, 2024

Authors: David Robinson¹, Marius Miron¹, Masato Hagiwara¹, Olivier Pietquin¹

Aff.: ¹Earth Species Project

Arxiv: http://arxiv.org/abs/2411.07186v1

Model	esc50	watkins	cbi	humbugdb	dcase	enabirds	hiceas	rfcx	gibbons
LLM w/o audio	0.020	0.041	0.005	0.073	0.000	0.001	0.210	0.000	\ul0.013
SALMONN	0.320	0.041	0.004	0.090	0.005	0.004	0.097	0.002	0.005
Qwen2-audio	0.307	0.041	0.004	0.070	0.005	0.004	0.097	0.002	0.005
BioLingual	\ul0.600	\ul0.257	\ul0.705	\ul0.085	\ul0.036	\ul0.109	0.429	\ul0.004	0.018
NatureLM-audio	0.635	0.646	0.755	0.073	0.052	0.279	\ul0.390	0.039	0.003