Earth Species Project

What Earth Species Project does

Earth Species Project (ESP) is a 501(c)(3) nonprofit research organization founded in 2018 and headquartered in Berkeley, California, that develops open-source AI foundation models to decode non-human animal communication. Its core product, NatureLM-audio, is described as the first large audio-language model specifically designed for bioacoustics and is built on a LLaMA 3.1-8B backbone, enabling species classification, individual recognition, life-stage prediction, and call-type detection across thousands of species via natural-language prompts. The broader product portfolio includes BirdAVES and AVES (self-supervised audio encoders), Voxaboxen (vocalization detection/classification), Biodenoising (noise reduction), and PrimateFace (facial landmark analysis across 60+ primate genera), with an explicit strategic intent to extend into multimodal acoustic, behavioral, and visual foundation models spanning the Tree of Life.

The technology stack combines self-supervised pre-training, transfer learning from human speech and music, and novel adaptation methods, and is validated through the BEANS / BEANS-Zero benchmark suite, peer-reviewed publication, and external pilot deployments (FrogID at the Australian Museum, Raincoast Conservation Foundation for killer whales, University of León for carrion crows, Wageningen University for zebra finches, University of Pennsylvania for primates). All models are released under open-source licenses on GitHub and HuggingFace rather than sold, and ESP monetizes exclusively through philanthropic grants and donations, having raised approximately $18.2M cumulatively — most notably $10M from Reid Hoffman and $7M from the Waverley Street Foundation in October 2024. The organization is fully remote, employs 11–50 staff, and operates with a 2025 budget of approximately $8M growing to approximately $13M in 2026 under new CEO Steven VanRoekel and Chief Scientist Olivier Pietquin.

Name

Earth Species Project

Legal name

Earth Species Project

Website

https://earthspecies.org

Company type

Private

Founded year

2018

Operating status

Operating

Headcount range

11–50 employees

Short description

Earth Species Project is a 501(c)(3) nonprofit research organization that builds open-source AI foundation models — most notably NatureLM-audio, the first large audio-language model for bioacoustics — to decode animal communication, serving academic researchers, conservation groups, and philanthropic funders worldwide.

Ownership category

akta.pro rank

Product category

AI Bioacoustics Research Tools

NAICS

Administration of Conservation Programs (92412)

SIC

Services-Commercial Physical & Biological Research (8731)

akta.pro primary industry

Environmental DNA (eDNA) & Advanced Biodiversity Methods (Bioacoustics, Camera Traps) (EUAHAHAM)

akta.pro secondary industries

Wildlife Conservation & Species Protection (BPAGAGAB), Conservation & Human–Wildlife Conflict Mitigation (BPAGAIAI)

Keywords

• Bioacoustic AI models
• Animal communication AI
• Foundation model research
• Open source AI research
• Wildlife conservation technology

Headquarters

HQ city

Berkeley

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Operations, Marketing or Sales, Others

Revenue model

1. Philanthropic Grants and Donations: ESP is 100% philanthropically funded through grants from individual donors, family foundations, and institutional philanthropy. Major funding includes $10M from Reid Hoffman and $7M from Waverley Street Foundation. Additional supporters include Paul G Allen Family Foundation, National Geographic, and others.

Go-to-market motion2 records

Distribution channels3 records

Marketing channels6 records

Core offering

Earth Species Project develops open-source AI foundation models and tools (NatureLM-audio, BirdAVES, AVES, Voxaboxen, Biodenoising, PrimateFace) for decoding animal communication and supporting bioacoustics research. The models are freely distributed through GitHub and HuggingFace to researchers, ethologists, and conservation organizations studying species across the Tree of Life.

Product overview

Earth Species Project develops open-source AI foundation models and tools for decoding animal communication. The flagship product is NatureLM-audio, the first large audio-language model for bioacoustics, complemented by BirdAVES and AVES (self-supervised audio encoders), Voxaboxen (detection/classification tool), Biodenoising (noise reduction), and PrimateFace (facial analysis). All models are freely available on GitHub and HuggingFace. The company conducts research partnerships with universities and conservation organizations and funds external projects through a grants program.

Differentiator

Problem solved

Functional benefit

Products and services

• NatureLM-audio The first large audio-language model specifically designed for bioacoustics. Analyzes animal vocalizations across thousands of species using natural language prompts, achieving state-of-the-art zero-shot performance on species classification, life-stage prediction, call-type detection, and audio captioning.
• BirdAVES Self-supervised audio foundation model specifically trained for bird vocalizations. Achieves over 20% performance improvement over AVES on bird-related datasets, available in biox-base, biox-large, and bioxn-large configurations.
• AVES First-ever self-supervised foundation model for animal vocalizations. Encodes a wide range of animal vocalizations for downstream tasks, used in Voxaboxen, ISPA, and ocean soundscape monitoring.
• Voxaboxen Machine learning tool for detecting and classifying animal vocalizations in recorded audio with high temporal specificity, based on the AVES bioacoustics foundation model. Used to analyze 127,000+ crow vocalizations in the carrion crow study.
• Biodenoising Novel method for denoising animal vocalizations without requiring clean data, leveraging pre-trained speech enhancement models adapted for animal sounds. Accepted to ICASSP 2025.
• PrimateFace Machine learning resource for automated facial analysis across human and non-human primates. Detects faces, tracks 68 facial landmarks, and analyzes facial movements across more than 60 primate genera. Presented at NeurIPS 2025.
• ESP Research Grants Program Grant program in partnership with FootPrint Coalition and Experiment Foundation that funds external research projects in AI and non-human communication, providing up to $10,000 in fast-grant funding per project.

Quantifiable outcome

• Over 127,000 crow vocalizations detected and classified automatically using Voxaboxen
• +4 more outcomes

Named customers5 records

Segments3 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability12 records

Feature8 records

Partnerships

Seven partnerships are on record, tiered core and minor.

• University of LeóncoreStrategic or Co-development Partner · 15 December 2025Research partnership studying carrion crow communication in Northern Spain. Used lightweight biologgers (MiniDTAGs) to record high-resolution acoustic data, and ESP's Voxaboxen ML model to detect over 127,000 vocalizations. Study found minimal impact of biologgers on crow behavior.
• FrogID (Australian Museum)coreStrategic or Co-development Partner · 14 August 2025FrogID, the world's largest frog-focused citizen science project, evaluated NatureLM-audio for processing frog recordings. Testing showed strong performance on frog vs. non-frog classification that could save 300+ hours of manual validation annually. Partnership exploring integration into FrogID's pre-processing pipeline.
• Raincoast Conservation FoundationcoreStrategic or Co-development Partner · 27 June 2025Pilot project to study killer whale communication using AI. Combines Raincoast's decades of field expertise and individual whale identification with ESP's frontier AI tools. Uses synchronized drone footage and underwater hydrophone recordings to build datasets linking killer whale vocalizations to specific behaviors and individuals.
• FootPrint CoalitionminorStrategic or Co-development Partner · 18 August 2023Partnership with Robert Downey Jr's FootPrint Coalition initiative to accelerate technologies for sustainable abundance. Launched a grant program to catalyze new research into interspecies communication through Experiment.com.
• Experiment FoundationminorStrategic or Co-development Partner · 18 August 2023Partnership to fund new research projects in AI and non-human communication. Projects funded through Experiment.com crowdfunding platform with up to $10,000 in fast-grant funding.
• Raincoast Cetacean Conservation Research ProgramcoreStrategic or Co-development PartnerLed by Dr. Valeria Vergara and Dr. Lance Barrett-Lennard, co-directors with over 20 years of experience studying cetaceans. Partnership focuses on understanding killer whale communication dynamics from above and below.
• The Collective Intelligence ProjectminorStrategic or Co-development PartnerPartnership to conduct first global survey on AI-powered interspecies understanding. Surveyed 1,057 participants across 67 countries to understand public sentiment on using AI to decode animal communication.

Scale indicators9 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Rainforest Connection: Rainforest Connection builds AI-powered acoustic monitoring systems to combat illegal logging and protect biodiversity, including the open-source Arbimon platform. Highly comparable to ESP's bioacoustics-for-conservation mission, serving similar research and conservation customers.
• Cornell Lab of Ornithology (BirdNET): Cornell's BirdNET is an open-source AI model for bird sound identification and the de facto standard for avian bioacoustics. Directly comparable to ESP's BirdAVES in target taxon, methodology (self-supervised learning on bird audio), and target researcher audience.
• Wildlife Acoustics: Wildlife Acoustics is the dominant commercial provider of bioacoustic recording hardware and SongScope/Kaleidoscope analysis software. Comparable in serving bioacoustics researchers and conservation professionals, but on the hardware/closed-software side rather than ESP's foundation-model approach.
• Conservation Metrics: Conservation Metrics applies AI and machine learning to large-scale wildlife monitoring data, including bioacoustics and camera traps. Direct peer in applying AI tooling to conservation science workflows for similar end customers (researchers, NGOs, government agencies).
• Wild Me / Wildbook: Wild Me develops open-source AI (Wildbook) for individual animal identification from photos, used in wildlife research and conservation. Comparable as an open-source AI-for-wildlife platform serving research and conservation communities, though focused on visual rather than acoustic modalities.

Emerging players

• Zamba AI: Zamba offers open-source machine learning models for wildlife camera trap analysis. Comparable as an emerging open-source AI-for-wildlife player, but focused on camera trap / image data rather than bioacoustics.

Broad incumbents

• Google DeepMind (bioacoustics work): Google DeepMind has published foundational research on bioacoustics and animal communication analysis as part of its broader AI research portfolio. A broad incumbent with vastly larger resources whose bioacoustics work directly competes for scientific mindshare and could release competing foundation models.
• Microsoft AI for Earth: Microsoft's AI for Earth program historically funded AI applications in environmental and conservation science, including bioacoustics. Broad incumbent with overlapping mission and historical grant-making role comparable to ESP's philanthropic funding model, though scaled within a much larger corporate context.
• OpenAI: OpenAI develops the frontier foundation models that ESP's NatureLM-audio builds upon (LLaMA 3.1-8B). A broad incumbent whose general-purpose audio/multimodal models could compete with ESP's bioacoustic-specific approach if released with similar capabilities.
• Anthropic: Anthropic is a frontier AI lab whose multimodal and language model capabilities could be applied to bioacoustics. Broad incumbent representing the risk that general-purpose foundation models subsume specialized vertical applications.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/earthspecies
• YouTubeyoutube.com/channel/UCQCG3L5CavOgTHks9CK5Aww
• LinkedInlinkedin.com/company/earthspecies
• Xx.com/earthspecies

Revenue estimate

Valuation estimate

Number of profiles

Profiles8 records

Funding overview

Funding rounds2 records

Investors2 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Earth Species Project