Set this up for future audio discussions, before I forget

Thanks Sara!

Thanks Sara! I am copying here a message that I just wrote in another channel: Hello everybody, my name is Silvia Zuffi, my topic is 3D animal (quadruped) modeling, but I have a question for a new thing that I want to explore not related to quadrupeds. I want to record and analyze/detect the voice of fishes. We have done some initial recordings, and well, it is hard to say what we got. I was wondering if among the people here there is somebody that has experience in capturing underwater sound. Thank you!

Hey @Silvia Zuffi I haven't worked with fish, but have worked with audio recordings of bats. I can't point to someone specifically but there is good marine audio experience at Cornell and St Andrew's (https://synergy.st-andrews.ac.uk/marinebioacoustics/).

ok thanks

Hi @Silvia Zuffi, you can contact Aaron Rice (https://www.birds.cornell.edu/ccb/aaron-rice/). He is the fish guy in my lab and I am sure he would be happy to help (if you still have a need for it)!

http://www.soest.hawaii.edu/ore/dclde/

For thos interested in the automated detection/classification of marine mammal calls 🙂

This is so cool!!

great workshop. i highly recommend it!

That's really cool!

https://www.aicrowd.com/challenges/lifeclef-2020-bird-monophone

aicrowd.com

AIcrowd | Challenges

Crowdsourcing AI to solve real-world problems

Original URL: https://www.aicrowd.com/challenges/lifeclef-2020-bird-monophone

There is an upcoming WildLabs meetup all about audio, you can find out more and register here: https://www.wildlabs.net/resources/events/wildlabs-virtual-meetup-invitation-acoustic-monitoring

WILDLABS.NET

WILDLABS VIrtual Meetup Invitation: Acoustic Monitoring

The fourth and final event in Season 3 of the WILDLABS Virtual Meetup Series will explore Acoustic Monitoring in conservation. The meetup will be held on Tuesday, March 10th from 9:00pm_10:30pm GMT / 5:00pm_6:30pm EDT.

Original URL: https://www.wildlabs.net/resources/events/wildlabs-virtual-meetup-invitation-acoustic-monitoring

Hi, I'm new to slack and just getting into using acoustic monitoring. I'm going to be using machine learning to classify recordings for bats. From my literature search it seems like some good options that are already available for this are bat detective and waveman, but I'm interested in what other folks are using and developing. (Sorry if I'm posting this in the wrong place; I'm still learning my way around here.)

GitHub

chenxing3/Waveman

Waveman is pipeline to identify bat species for audio file - chenxing3/Waveman

Original URL: https://github.com/chenxing3/Waveman

Congrats to Rainforest Connections and Sieve Analytics in recent merge!! Go Topher and Mitch!
-signed, your admiring fans at the Songs of Adaptation project 😊

Our Kaggle competition is live. If you have any questions about the competition, please let me know! https://www.kaggle.com/c/birdsong-recognition/overview

kaggle.com

Cornell Birdcall Identification

Build tools for bird population monitoring

Original URL: https://www.kaggle.com/c/birdsong-recognition/overview

https://www.futureinreview.com/fireside/?utm_source=Inews+Lists&utm_medium=email&utm_campaign=Join+Me+at+FiReSide+Next+Week%3F

Future in Review

SNS FiReSide - Using AI for Whale Communication

SNS FiReSide: Using AI for Whale Communication

Est. reading time

5 minutes

Original URL: https://www.futureinreview.com/fireside/?utm_source=Inews+Lists&utm_medium=email&utm_campaign=Join+Me+at+FiReSide+Next+Week%3F

BirdCLEF 2021 competition is live: https://www.kaggle.com/c/birdclef-2021/overview

https://www.eurekalert.org/pub_releases/2021-06/cu-gte060421.php

EurekAlert!

Gift to establish K. Lisa Yang Center for Conservation Bioacoustics

The Cornell Lab of Ornithology's Center for Conservation Bioacoustics will begin a new era of innovation thanks to a major gift from the philanthropist and Lab Advisory Board member K. Lisa Yang.

Original URL: https://www.eurekalert.org/pub_releases/2021-06/cu-gte060421.php

Hello all! 👋 Is there anyone here working on gibbon 🐵 vocalisations? If so drop me a message I'd love to hear about your work 😄

Slow-Fast Auditory Streams for Audio Recognition

Evangelos Kazakos, Arsha Nagrani, Andrew Zisserman and Dima Damen

"We propose a two-stream convolutional network for audio recognition, that operates on time-frequency spectrogram inputs. Following similar success in visual recognition, we learn Slow-Fast auditory streams with separable convolutions and multi-level lateral connections. The Slow pathway has high channel capacity while the Fast pathway operates at a fine-grained temporal resolution. We showcase the importance of our two-stream proposal on two diverse datasets: VGG-Sound and EPIC-KITCHENS-100, and achieve state-of-the-art results on both."

https://ekazakos.github.io/auditoryslowfast/

just won an outstanding paper award at ICASSP :)

Is anyone here studying the Hainan peacock-pheasant (Polyplectron katsumatae)? 🧐

Any good book recommendations for bioacoustics and/or audio processing?

Did you see @gvanhorn's recent blog posts? Not a book but I found them really helpful 🙂

https://www.macaulaylibrary.org/2021/07/19/from-sound-to-images-part-1-a-deep-dive-on-spectrogram-creation/

https://www.macaulaylibrary.org/2021/08/05/from-sound-to-images-part-2-spectrogram-image-processing/

macaulaylibrary.org

From Sound to Images, Part 1: A deep dive on spectrogram creation.

In our first post, we described the idea of using a computer vision model to identify bird vocalizations. But how does a computer vision model “listen” to a sound? For Sound ID, we use the short-t

Original URL: https://www.macaulaylibrary.org/2021/07/19/from-sound-to-images-part-1-a-deep-dive-on-spectrogram-creation/

macaulaylibrary.org

From Sound to Images, Part 2: Spectrogram Image Processing.

Picking up on where we left off in the previous post, we will now look at the various ways one can transform the spectrogram image prior to analysis by a convolutional neural network (CNN) and how the

Original URL: https://www.macaulaylibrary.org/2021/08/05/from-sound-to-images-part-2-spectrogram-image-processing/

I haven't, but I'll look into them. I ask as my department has some funds they haven't used so I want some books, because, books!

Good morning everyone. There is a fellow grad student in one of the labs that I am associated with who has run into a bit of a bind due to supply chain issues. She was hoping to deploy AudioMoths this summer as part of a Masters project, but they are currently out of stock everywhere she's looked. With fingers crossed, I thought I would check if anyone knows of any ARUs collecting dust this summer that would be available for rent or loan?

Hello all! Anyone here actively working in ML and bioacoustics and has time to chat and bounce ideas on an on/off basis?

Hi all, we have published recently a report on some work we have been doing on fish sounds, I would be interested in some feedback and suggestions! https://arxiv.org/abs/2201.05013v2

arXiv.org

Fish sounds: towards the evaluation of marine acoustic...

The marine ecosystem is changing at an alarming rate, exhibiting biodiversity loss and the migration of tropical species to temperate basins. Monitoring the underwater environments and their...

Original URL: https://arxiv.org/abs/2201.05013v2

https://www.kaggle.com/c/birdclef-2022/overview

kaggle.com

BirdCLEF 2022

Identify bird calls in soundscapes

Original URL: https://www.kaggle.com/c/birdclef-2022/overview

https://peerj.com/articles/13152/

PeerJ

Computational bioacoustics with deep learning: a review and roadmap

Animal vocalisations and natural soundscapes are fascinating objects of study, and contain valuable evidence about animal behaviours, populations and ecosystems. They are studied in bioacoustics and ecoacoustics, with signal processing and analysis an important component. Computational bioacoustics has accelerated in recent decades due to the growth of affordable digital sound recording devices, and to huge progress in informatics such as big data, signal processing and machine learning. Methods are inherited from the wider field of deep learning, including speech and image processing. However, the tasks, demands and data characteristics are often different from those addressed in speech or music analysis. There remain unsolved problems, and tasks for which evidence is surely present in many acoustic signals, but not yet realised. In this paper I perform a review of the state of the art in deep learning for computational bioacoustics, aiming to clarify key concepts and identify and analyse knowledge gaps. Based on this, I offer a subjective but principled roadmap for computational bioacoustics with deep learning: topics that the community should aim to address, in order to make the most of future developments in AI and informatics, and to use audio data in answering zoological and ecological questions.

Original URL: https://peerj.com/articles/13152/

We are finally able to share a GUI-enabled version of BirdNET. See: https://github.com/kahst/BirdNET-Analyzer In addition to 2000+ bird species, this version also includes a bunch of North American frog and mammal species. This is a beta, if you run into any issues, please let us know!

Hi all — excited to be here! I’m Masato, a Senior AI Researcher working at Earth Species Project. We’ve been working on an exciting project where we build a standardized benchmark for bioacoustics (“ESP Bioacoustics Benchmark”) and I’m wondering if anyone here is interested in providing early feedback for this project.

Here I copied the project description:

Machine learning (ML) and deep learning (DL) have been increasingly used for solving bioacoustics tasks, but typical research studies in bioacoustics focus on different types of species and ML/DL models, making it difficult to compare across different approaches in a standardized way. Inspired by the benchmark datasets in other ML domains such as ImageNet and GLUE, we propose the ESP Bioacoustic Benchmark, a collection of bioacoustics tasks and datasets specifically designed to measure and compare the performance of ML algorithms in a standardized way. We specifically focus on the two unique aspects of bioacoustics—diversity of species and the small data size, and design the benchmark so that it rewards ML models that address these two issues successfully. We also publish the preprocessed datasets, the library code, and non-DL and DL baseline implementations to make it easier to reproduce and build new models on top of the benchmark infrastructure.

Here are some key questions we would like to see answered:

• Would you be interested in using this benchmark? Why and why not?


• Does the current selection of datasets make sense? What would you add and/or remove?


• Would you be interested in contributing a dataset for this?

earthspecies.org

Earth Species Project

An open-source collaborative and nonprofit dedicated to decoding animal communication. We are motivated by the recent monumental milestone in machine learning: the invention of techniques that can translate languages without dictionaries. We believe that an understanding of non-human languages will change our ecological impact on this planet. Like the telescope, these new tools will forever change our perspective.

Original URL: https://www.earthspecies.org/

Earth Species Project on Notion

ESP Bioacoustic Benchmark

Machine learning (ML) and deep learning (DL) have been increasingly used for solving bioacoustics tasks, but typical research studies in bioacoustics focus on different types of species and ML/DL models, making it difficult to compare across different approaches in a standardized way. Inspired by the benchmark datasets in other ML domains such as ImageNet and GLUE, we propose the ESP Bioacoustic Benchmark, a collection of bioacoustics tasks and datasets specifically designed to measure and compare the performance of ML algorithms in a standardized way. We specifically focus on the two unique aspects of bioacoustics—diversity of species and the small data size, and design the benchmark so that it rewards ML models that address these two issues successfully. We also publish the preprocessed datasets, the library code, and non-DL and DL baseline implementations to make it easier to reproduce and build new models on top of the benchmark infrastructure.

Original URL: https://earthspecies.notion.site/ESP-Bioacoustic-Benchmark-b6d5ef9327ae4f3983c1923bb7e33ab2

Good afternoon everyone. It seems like many paper I read laud bioacoustic monitoring as a reasonable large-scale monitoring approach; however, I'm not aware of that many programs which have deployed recorders and intend to maintain them indefinitely. I'd like to compile a list of applied long-term acoustic monitoring programs (i.e. programs that are managed like they will exist for years) and I was hoping I could tap into the wealth of knowledge here.

Anyone else going to AudioXD? https://kitzeslab.github.io/audioxd/

This channel is - ironically - very quiet. And it's Friday, so I'm going to post something that doesn't really offer anything of value, just a neat bioacoustics experience I had today... this morning I was walking my dog in a quiet park, and we heard a sound that first seemed like another (upset) dog, but then I noticed it was sort of doing a call-and-response with what was clearly an owl, so I thought maybe the unusual sound was a juvenile owl. It also made my dog completely lose her mind, so I didn't quite manage to get a clean recording of the unknown sound, but Merlin (a) did an amazing job ignoring my barking dog and (b) told me that the adult owl was a great horned owl, so when I got home, I searched for "great horned owl juvenile sounds", and voila, this is exactly what we heard:

https://www.youtube.com/watch?v=uQDnPDaOXyI

It's wild how good a job Merlin did ignoring my dog in this recording and identifying the owl, which must be around 1% as loud as the dog sounds:

https://drive.google.com/file/d/1MiwW5JiW-lFTTiyN7A9osMpHNbylWd9v/view?usp=sharing

Huge props to the Merlin Sound ID team, that's amazing!

YouTube

Scott Ramos (https://www.youtube.com/c/ScottRamos)

Great Horned Owl juvenile begging calls

Original URL: https://www.youtube.com/watch?v=uQDnPDaOXyI

Dear audio channel, I would like to ask a suggestion. We have done some work on sound separation for fish vocalizations, but I am not really in the field, and I was wondering what could be a venue (small conference/workshop) to present the work. Any suggestions? https://arxiv.org/abs/2201.05013v2

arXiv.org

Fish sounds: towards the evaluation of marine acoustic...

The marine ecosystem is changing at an alarming rate, exhibiting biodiversity loss and the migration of tropical species to temperate basins. Monitoring the underwater environments and their...

Original URL: https://arxiv.org/abs/2201.05013v2

Hello Everyone, Happy New Year !
I am Félix Michaud, PhD student at Sorbonne Université and the National Museum of Natural History in Paris under the supervision of Jérôme Sueur and Sylvain Haupert. We are very happy to share our work dedicated to the reduction of label noise in bird sound datasets detailed in the new paper Unsupervised classification to improve the quality of a bird song recording dataset.

You will find the code available here : https://github.com/ear-team/bambird

Hi everyone. I’ve just had an email off IBAC saying their conference is now open for abstract submissions and thought it might be of interest to people here 🙂 https://2023.ibac.info/index.html

https://academicjobsonline.org/ajo/jobs/24693

I've been flipping through a number of published avian bioacoustic datasets today and I am not seeing many with intraspecies call type labels. Is anyone aware of any published datasets with call and song type labels?
NIPS4b has call and songs
Picidea has some intraspecies labels

Hey! I have been using the BirdNET-Analyzer model v2.4 to compute embeddings for a research project I am working on. If any of the authors are here, thanks for making this model available! It works very well.

Does anyone know if there exists a detailed description of the model architecture used in BirdNET v2.4? From the version history the architecture changed to using EfficientNet V2 blocks in v1.2, and I assume that there has been a lot of changes as well until v2.4. I believe that the paper ("BirdNET: A deep learning solution for avian diversity monitoring") describes the initial Wide-ResNet model, correct me if I am wrong.

Any pointers to where I can find a description of the current architecture (if this is available) would be very welcome!

Hello!

I am thinking about the best hardware to record some audio at a site in the Baltic Sea where we are studying Guillemots. It is a rather nice site in the sense that both networking and power infrastructure is already in place, and currently cameras are set up to take images of the birds, but no audio is being recorded. The cameras seem to encode audio at a G.711 standard meaning that it is sampled at 8kHz, which is not sufficient, otherwise simply connecting microphones to the cameras would have been a great alternative.

The easiest type of audio recording device to integrate into this site would be something that just needs a power supply and with a file system that can be accessed via wifi. I have not really found anything like this and may be looking in the wrong places. The AudioMoth may be an option if we install the USB Microphone firmware and connect it to a Raspberry Pi or something for networking capabilities, but I am wondering if there may be anything else out there that is doing these things out of the box?

Question: Does anyone know of an acoustic recording device which runs on power supplied by a cable (no need to replace batteries) and with a file system that is accessible over wifi?

Yes there is! I would go with a Barix Instreamer (https://www.barix.com/product/instreamers/) and stream the data from the site to your lab. There you can set up a server to store the data and analyze it (you can also listen in, which is nice).

Hey Folks, bottom line up front: ❓ Has anyone heard of a tool for classification of the black grouse from audio recording?

Context below:
A friend of a friend of mine, leads a research project piloting the monitoring of black grouse in the Izera Mountains using acoustic recorders. Story as old as time -- manual listening and identifying birbs takes a tremendous amount of time for them. They tried a few tools, namely CornelLab - Merlin, but nothing they tried was identifying black grouse correctly.

Hi all, this week we released our pretrained model 'SurfPerch' which is a highly effective transfer learner for new datasets from coral reefs and similar habitats (available on Kaggle). It can be easily fine tuned to find any reef or fish like sound you want with far less training data and compute cost. We have a tutorial demonstrating how this can can be performed using example data on gdrive with google colab alongside an active learning process for added labelling speed: https://github.com/BenUCL/surfperch/blob/surfperch/SurfPerch_Demo_with_Calling_in_Our_Corals.ipynb

Our research article detailing the models creation is on arvix: https://arxiv.org/abs/2404.16436. As part of this we assembled and open source ReefSet, an annotated library of 57k reef sounds across 38 classes from 16 unrelated datasets: https://paperswithcode.com/dataset/reefset

There's also a blog post on Google's keyword blog: https://blog.google/outreach-initiatives/arts-culture/a-new-ai-tool-to-help-monitor-coral-reef-health/

arXiv.org

Leveraging tropical reef, bird and unrelated sounds for superior transfer learning in marine bioacoustics

Machine learning has the potential to revolutionize passive acoustic monitoring (PAM) for ecological assessments. However, high annotation and compute costs limit the field's efficacy. Generalizable pretrained networks can overcome these costs, but high-quality pretraining requires vast annotated libraries, limiting its current applicability primarily to bird taxa. Here, we identify the optimum pretraining strategy for a data-deficient domain using coral reef bioacoustics. We assemble ReefSet, a large annotated library of reef sounds, though modest compared to bird libraries at 2% of the sample count. Through testing few-shot transfer learning performance, we observe that pretraining on bird audio provides notably superior generalizability compared to pretraining on ReefSet or unrelated audio alone. However, our key findings show that cross-domain mixing which leverages bird, reef and unrelated audio during pretraining maximizes reef generalizability. SurfPerch, our pretrained network, provides a strong foundation for automated analysis of marine PAM data with minimal annotation and compute costs.

Original URL: https://arxiv.org/abs/2404.16436

paperswithcode.com

Papers with Code - ReefSet Dataset

The full version of ReefSet used in Williams et al. (2024). This dataset contains strongly labeled audio clips from coral reef habitats, taken across 16 unique datasets from 11 countries. This dataset can be used to test transfer learning performance of audio embedding models.<br><br>This folder includes:<br>- 57,084 WAV files that make up ReefSet_v1.0. Each file is 1.88 seconds in length, sampled at 16 kHz. All files have an associated label.<br>- reefset_annotations.json, which contains the associated label, file ID, filename, data sharer, dataset and recorder type used for each sample in ReefSet_v1.0. This information is also indicated in the filename of each file.<br>- 'reefset_labels_by_dataset.csv' provides a table containing the counts of each label class within each dataset.

Original URL: https://paperswithcode.com/dataset/reefset

Google

A new AI tool to help monitor coral reef health

We’re releasing the AI model SurfPerch, that is being deployed thanks to contributions from people around the world to our Calling in Our Corals platform

Original URL: https://blog.google/outreach-initiatives/arts-culture/a-new-ai-tool-to-help-monitor-coral-reef-health/

<https://github.com/BenUCL/surfperch/blob/surfperch/SurfPerch_Demo_with_Calling_in_Our_Corals.ipynb | SurfPerch_Demo_with_Calling_in_Our_Corals.ipynb>

<p>```<br>{<br> "nbformat": 4,<br> "nbformat_minor": 0,<br> "metadata": {<br> "colab": {<br> "provenance": [],<br> "gpuType": "T4",<br> "authorship_tag": "ABX9TyNYxOeuKWO+buUsZvszg7c7",<br> "include<em>colab</em>link": true<br> },<br> "kernelspec": {<br> "name": "python3",<br> "display_name": "Python 3"<br> },<br> "language_info": {<br> "name": "python"<br> },<br> "accelerator": "GPU"<br> },<br> "cells": [<br> {<br> "cell_type": "markdown",<br> "metadata": {<br> "id": "view-in-github",<br> "colab_type": "text"<br> },<br> "source": [<br> "<a href=\"https://colab.research.google.com/github/BenUCL/surfperch/blob/surfperch/SurfPerch<em>Demo</em>with<em>Calling</em>in<em>Our</em>Corals.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"<br> ]<br> },<br> {<br> "cell_type": "markdown",<br> "source": [<br> "# <strong><strong>SurfPerch Tutorial with Calling in Our Corals</strong></strong> 🐠🐦\n",<br> "\n",<br> "This notebook was adapted from the original NeurIPS workshop tutorial 'CCAI Agile Modeling for Bioacoustic Monitoring Tutorial', authored with equal contribution by:\n",<br> "** Jenny Hamer, Google DeepMind\n",<br> "** Rob Laber, Google Cloud\n",<br> "** Tom Denton, Google Research\n",<br> "\n",<br> "The adaptation presented here was authored by:\n",<br> "** Ben Williams, UCL (Google DeepMind Student Researcher)"<br> ],<br> "metadata": {<br> "id": "c-iKkbImw<em>I</em>"<br> }<br> },<br> {<br> "cell_type": "markdown",<br> "source": [<br> "# Table of Contents\n",<br> "** <a href="#introduction">Introduction</a>\n",<br> " - <a href="#background">Background</a>\n",<br> " - <a href="#cioc">Calling in Our Corals</a>\n",<br> " - <a href="#SurfPerch">SurfPerch</a>\n",<br> " - <a href="#glossary">Glossary</a>\n",<br> " - <a href="#agile_modeling">Agile Modeling for Bioacoustics</a>\n",<br> " - <a href="#software-requirements">Software Requirements</a>\n",<br> " - <a href="#adapting">Adapting to New Use Cases</a>\n",<br> "** <a href="#set-up">Set Up</a>\n",<br> "** <a href="#methodology">Methodology</a>\n",<br> " - <a href="#embed_data">Embed the Search Dataset</a>\n",<br> " - <a href="#similarity-search">Audio Similarity Search</a>\n",<br> " - <a href="#active-learning">Active Learning</a>\n",<br> "** <a href="#results">Results</a>"<br> ],<br> "metadata": {<br> "id": "5Vk0HPvDDIvz"<br> }<br> },<br> {<br> "cell_type": "markdown",<br> "source": [<br> "<a name=introduction></a>\n",<br> "# Introduction\n",<br> "\n",<br> "> (8 min read)\n",<br> "\n"<br> ],<br> "metadata": {<br> "id": "81Y68NsAHe6X"<br> }<br> },<br> {<br> "cell_type": "markdown",<br> "source": [<br> "<a name=background></a>\n",<br> "## Background\n",<br> "\n",<br> "Passive acoustic monitoring (PAM) can be used to collect valuable information on the state of marine habitats. However, this data can easily range from dozens to thousands of hours in size, becoming too large for marine scientists to manually analyse. Machine learning offers a powerful solution to this challenge.\n",<br> "\n",<br> "The goals of this notebook are:\n",<br> "1. Provide marine scientists with a workflow that enables them to rapidly anaylse PAM data using machine learning.\n",<br> "2. Guide users through how to build a highly accurate machine learning detector known as a 'classifier' for any given sound of interest in their dataset, starting from a single example.\n",<br> "3. Demonstrate how this can be run entirely from this notebook using data uploaded to Google Drive.\n",<br> "\n",<br> "We demonstrate this using data gathered for the Calling in Our Corals project.\n"<br> ],<br> "metadata": {<br> "id": "vHZcexFduQ9X"<br> }<br> },<br> {<br> "cell_type": "markdown",<br> "source": [<br> "<a name=cioc></a>\n",<br> "## Calling in Our Corals\n",<br> "\n",<br> "Coral reef habitats have rich and diverse soundscapes. However, a key challenge bioacousticians working on these habitats must overcome is identifying which sounds are of interest. For any given dataset, the source of most of sounds are typically unknown and undocumented making it challenging for marine bioacousticians to know where to begin.\n",<br> "\n",<br> "<a href="https://artsandculture.google.com/experiment/calling-in-our-corals/zgFx1tMqeIZyTw?hl=en">Calling in Our Corals</a> (CIOC) is a collaboration started in 2023 between Google Arts and Culture and marine biologists hoping to overcome this challenge. The project is centred around a platform used to crowd source annotations for marine acoustic data. Over 24'000 unique users logged on to the platform and listened to a combined 400 hours of audio recorded from marine habitats in 10 different countries, annotating any periods they identified as containing sounds of interest. By taking the audio clips with the highest number of annotations for each dataset a diverse set of biological sounds were revealed from each.\n",<br> "\n",<br> "Here, we use some of these sounds from three datasets as an entry point into our machine learning pipeline. The results shared at the end only represents a preliminary analysis using a subset of the data as scientists continue to work on the outputs. However, we find this reveals interesting patterns between the soundscape and state of the habitats."<br> ],<br> "metadata": {<br> "id": "PK9mjDu3C6DE"<br> }<br> },<br> {<br> "cell_type": "code",<br> "source": [<br> "#@title Learn more about coral reef sounds here\n",<br> "from IPython.display import YouTubeVideo\n",<br> "\n",<br> "# YouTube video ID\n",<br> "video_id = \"POITH02VVrw\"\n",<br> "\n",<br> "# Display the YouTube video in the notebook\n",<br> "YouTubeVideo(video_id)"<br> ],<br> "metadata": {<br> "colab": {<br> "base_uri": "<a href="https://localhost:8080/">https://localhost:8080/</a>",<br> "height": 321<br> },<br> "cellView": "form",<br> "id": "-863qtX_Uj2Z",<br> "outputId": "ecfd15f7-2289-414e-ca0b-728ab206c47f"<br> },<br> "execution_count": null,<br> "outputs": [<br> {<br> "output<em>type": "execute</em>result",<br> "data": {<br> "text/plain": [<br> "<IPython.lib.display.YouTubeVideo at 0x7ade04d88d60>"<br> ],<br> "text/html": [<br> "\n",<br> " <iframe\n",<br> " width=\"400\"\n",<br> " height=\"300\"\n",<br> " src=\"https://www.youtube.com/embed/POITH02VVrw\"\n",<br> " frameborder=\"0\"\n",<br> " allowfullscreen\n",<br> " \n",<br> " &gt;</iframe>\n",<br> " "<br> ],<br> "image/jpeg": "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…</p>

<a href="https://github.com/BenUCL/surfperch">BenUCL/surfperch</a>

Has anyone managed to get BirdNET working. ?

Hello, all! My name is Clare, and I am at UBC’s Institute for Resources, Environment, and Sustainability working on my Master’s. I previously worked with the late Dr. Karen Bakker on her book about interspecies communication and acoustics (hence my introduction into this space).

My thesis is looking into the relationship between anthropogenic noise and birdsong in urban soundscapes, probing at perceptions of nature and spaces (as being “good” or “bad” for recreation, spending time in nature, etc) as a function of the soundscape. I was wondering if any of you were aware of a program, or had trained one yourself, that can classify presence/absence (ideally relative abundance as well) of what I am thinking of as sound “guilds” (attributes like construction, road sound, birds, geophonic sound, human beings, etc) in a soundscape recording. The alternative is to design soundscapes with controlled attribute (“guild”) ratios; if anyone has experience doing that, I would also love to connect. I am reachable here, or at my email theclareprice@gmail.com. Thanks so much, looking forward to chatting 🙂

I am happy to share that our latest work "From Weak to Strong Sound Event Labels using Adaptive Change-Point Detection and Active Learning" has been accepted at EUSIPCO 2024 and nominated to the best student paper category.

• code: https://github.com/johnmartinsson/adaptive-change-point-detection


• paper: https://eurasip.org/Proceedings/Eusipco/Eusipco2024/pdfs/0000902.pdf


• data: https://zenodo.org/records/10811797

Just reposting this here - https://aiforconservation.slack.com/archives/CLWGQ4BJ6/p1724704711146579

Vanesa Reyes (https://aiforconservation.slack.com/team/U075U3GF4FQ)

<p>Hey there! At WILDLABS we have launched a <a href="https://wildlabs.net/article/listen-future-bioacoustics-horizon-scan">bioacoustics horizon scan</a>, and since AI has become a fundamental part of the field, we'd love to hear from you what innovations do you predict will revolutionise bioacoustics in the next two decades? These will be considered for inclusion in the horizon scan prioritisation process, involving diverse experts from across the globe. You can submit your ideas using this <a href="https://forms.gle/dKg8QPM9SMR3bFCeA">Google form</a>.</p>

Original URL: https://aiforconservation.slack.com/archives/CLWGQ4BJ6/p1724704711146579

Slack Conversation

Hi everyone,

I'm preparing a manuscript, and we have also published a bioacoustic dataset (focused on bird sounds and soundscapes). The editor is satisfied with the structure and metadata we've created, but they are requiring us to specify the ISO standard used for the metadata description. We based our work on other datasets, and while the editor is OK with that, we’re unsure which ISO standard to reference, as we don’t know what standards those datasets were following.

Does anyone know which ISO standard(s) might apply to bioacoustic datasets, or how we could align our metadata description with ISO recommendations?

Thanks in advance for any guidance!

Hello all! Does anyone have knowledge of or experience with any off-the-shelf, wireless, ARUs? I.e. like an AudioMoth that has cellular or satellite connectivity? I'm starting to explore what's out there and thus far I am aware of Rainforest Connection's Guardian system, these Automated Surveying Units by Instinct Environmental, and these more DIY units designed by Oekofor called ecoPi

Edge-detection and the ability to load custom BirdNet weights would also be hugely valuable

Hi Nathaniel, if you have questions about the ecoPIs, send me a direct message. This is a project we are running right now using them: https://birdnet.cornell.edu/swamp/

Haikubox and Birdweather PUCs do real-time detections (sends alerts to an app on your phone)

Haikubox

Haikubox | Bird Call Identifier &amp; Bird Song Recognition

Haikubox is the smart bird song identification device that brings you real-time alerts, bird call sound recordings, and loads of information about your birds. Identify bird calls by sound, listen to and download your birds' recordings, track activity, and share your birds with family and friends using the Haikubox app.

Original URL: https://haikubox.com/

BirdWeather

BirdWeather - a living library of bird vocalizations.

Join a growing community of citizen scientists with PUC, our AI powered bioacoustics platform.

Original URL: https://www.birdweather.com/

Hi all. This whole time I didn't know we had an "audio" channel here. So 👋 !

Hi all, our self-supervised audio encoder for animal sounds, AVES, is now pip-installable! This means that you can install the package and get the embeddings from audio files with a single command without writing any code. More details below ⬇️ https://x.com/earthspecies/status/1912205602257772649

X (formerly Twitter)

Earth Species Project (@earthspecies) on X

AVES is now pip-installable 🎉<br><br>This self-supervised, transformer-based model is pretrained on large-scale animal vocalization datasets &amp; thanks to our incredible engineering team is now more accessible than ever–ready to run with just a single command.<br><br>🔗<https://t.co/oiecCLTH6h>

Original URL: https://x.com/earthspecies/status/1912205602257772649

👋 Hello from Earth Species Project!

We’re thrilled to share that we’ve open-sourced NatureLM-audio, the world’s first audio-language foundation model for bioacoustics. This release includes:

• the model weights


• inference code


• the newly curated BEANS-Zero benchmark


• and our evaluation toolkit

earthspecies.github.io

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

The world's first audio-language model designed for animal sound understanding and conservation research..

Original URL: https://earthspecies.github.io/naturelm-audio-demo/

Had a new paper out where we show embeddings from pretrained NN's can be used to train classifiers or unsupervised learning algs to explore marine soundscapes pretty much as well as NN's trained from scratch, and do much better than 'hand crafted features'. The findings seem pretty obvious now but when the work was first done in 2021 this was still a new topic! It was hard to find good data so it spent a lot of back and forth time in review hence the delay.
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1013029

journals.plos.org

Unlocking the soundscape of coral reefs with artificial intelligence: pretrained networks and unsupervised learning win out

Artificial intelligence has the potential to revolutionise bioacoustic monitoring of coral reefs. So far, only a limited set of work has used machine learning to train detectors for specific sounds such as individual fish species. However, building detectors is a time-consuming process that involves manually annotating large amounts of audio, followed by complicated model training. This process must then be repeated for any new dataset. Instead, we explore machine learning techniques for whole soundscape analysis, comparing the acoustic properties of raw recordings from the entire habitat. We identify multiple machine learning methods for whole soundscape analysis and rigorously test these using datasets from Indonesia, Australia and French Polynesia. Our findings reveal that use of a neural network pretrained on 5.2 million hours of unrelated YouTube audio offers a powerful tool to produce compressed representations of reef audio data, conserving the data’s key properties whilst being executable on a standard personal laptop. These representations can then be used to explore patterns in reef soundscapes using unsupervised machine learning, which is effective at grouping similar recordings together. We show these groupings hold relationships with ground truth ecological data, including coral cover, the fish community and depth.

Original URL: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1013029

I want to share a webpage about a recent bioacoustic monitoring setup that we have established at the Auk Lab at Stora Karlsö in the Baltic Sea. The Auk Lab has been monitoring and collecting data about Common Guillemots at Stora Karlsö for over a decade, and this year I was fortunate enough to be part of a bioacoustic monitoring installation on the island. These seabirds are important indicators of the overall health of the ecosystems in the Baltic Sea. Hopefully we will learn more about the Guillemots from their vocalizations.

There will be many interesting audio-visual and other multimodal modelling challenges here (the curent weather conditions are logged every second et cetera), which will make for fun audio modelling problems. Can we jointly identify audio-visual behavior cues to learn more about what a certain vocalization might mean? Can wind and rain noise reduction methods be informed by the logged weather data? It is going to be fun trying to figure it out!

auklab-audio.github.io

Auklab Audio Research

News and updates about the continuous acoustic monitoring at the Auk Lab

Original URL: https://auklab-audio.github.io/

I'm proud to say that the results of the first BioDCASE Challenge are now online! Click "Results" here: https://biodcase.github.io/challenge2025/

We have a new preprint out using AI accelerated passive acoustic monitoring of coral reefs to measure ecological outcomes from restoration that I hope is of interest to some: https://doi.org/10.1101/2025.09.24.678197 🪸

1. We gathered data from 45 sites across five countries totaling >12 months of raw audio. We used SurfPerch and the Google Perch teams incredible 'agile modelling' workflow to build 34 classifiers for common fish and invertebrate sounds, generating 912k high confidence detections of these.


2. We used these detections as indicators of four important coral reef ecosystem functions (fish recruitment, fish diversity, parrotfish grazing, shrimp bioturbation). This revealed recovery in 3/4 functions on restored reefs around 3yrs of age, with some bouncing back within just 3mo at some locations.


3. We've also published the full raw audio dataset and all 912k detections open-source on figshare for anyone to use (about 800GB once unzipped): https://doi.org/10.5522/04/29958062.v2 - would love to hear if anyone finds uses for this!

kaggle.com

Google | SurfPerch | Kaggle

SurfPerch is a domain-adapted model for classification of sounds in coral reefs.

Original URL: https://www.kaggle.com/models/google/surfperch

figshare

Coral Reef Soundscapes from a Global Restoration Programme

OverviewSee the associated study (Williams et al., 2025) for full details on the dataset and analysis.This repository contains a large dataset of raw soundscape recordings and human-in-the-loop AI generated sonotypes detections (fish and invertebrate sounds) from these soundscapes. This data was used to conduct a global study investigating the recovery of coral reef ecosystem functions following active restoration. Data were collected from 45 tropical reef sites across Australia, Indonesia, Kenya, Mexico, and the Maldives, totalling over 1year of raw audio.Each site is one of four habitat types: degraded, healthy, early-stage restored (

Original URL: https://doi.org/10.5522/04/29958062.v2

Hi all, we’re Synature (https://synature.ch/en/), building tools to make bioacoustics more accessible. We offer a smart microphone that auto-uploads recordings to reduce fieldwork overhead, and an intuitive data management platform.
We’re looking for users (individuals or organizations) to pilot our data management platform (we have species detection and bioacoustic indices on the platform). The only requirement: you already have audio data ready to upload.
Please complete this short form if you're interested: https://forms.gle/Vh8btMSvthXoMfKf7

Google Docs

Synapp Platform Waitlist

Many thanks for your interest on testing and using our platform! We are enabling access to early adopters on a rolling basis. It is only possible to use our platform if you already have audio data available.<br><br>We can detect around 10'000 species globally with a focus on birds - however we are continuously expanding our scope - in the last weeks we added bat species, frogs, toads, grasshoppers and crickets!<br><br>Please answer the questions below in order for us to understand your dataset and determine suitability.

Original URL: https://forms.gle/Vh8btMSvthXoMfKf7

Hello! I just joined this Slack channel, happy to be here. I'm a machine learning specialist working at Ocean Networks Canada (ONC), primarily focusing on passive acoustic monitoring and classification of "anomalies" (e.g., instrumentation failures, environmental sounds, marine mammal sounds) in ocean recordings. I am wondering if anyone knows of any citizen science based projects for crowd-sourced annotations in this domain. Especially any that might allow ONC to upload its data.

I am doing some prototyping around supply chain mapping, and I want to have a few publicly-available schematics/BOMs for electronic devices handy whose supply chains I can analyze (I can do this for proprietary Google things, but that doesn't help me talk about the work to anyone outside Google). It doesn't matter if the actual products have anything to do with conservation or sustainability, but I figure, why not choose a sustainability-related product for demos... to that end, AudioMoth came to mind, and I can find a zillion places on the Internet telling me that for all versions older than the current version, the schematic+BOM should be available, but I can't actually find the schematic/BOM. Does anyone have a link? If it matters, I don't actually care about the schematic, only the parts list / BOM, but I can determine the latter from the former if the schematic is the only thing that is easily available.

I built a bioacoustics webapp for personal use that runs model inference in the browser similar to similar to the BirdNET webapp but with more acoustic analysis features. I'd appreciate any feedback from the community. Disclaimer: not a bioacoustics expert, just trying to build cool tools and learn in the process

https://isaac.earth/bioacoustics

Our colleague Stuart Newson asked me to post this survey! 📝

How are automated species identifications from passive acoustic monitoring verified and quality-assured in practice?

Automated acoustic species identification is now widely used for birds, bats, amphibians, and other terrestrial taxa — but approaches to verification, uncertainty, and quality assurance vary widely between tools, projects, and sectors.

British Trust for Ornithology (BTO), working with Natural England and funded by Department for Environment, Food and Rural Affairs, is running a short anonymous survey to understand current practice around automated species identification in Passive Acoustic Monitoring, and to identify challenges and guidance needs.

We’d love to hear from:
▪️ Developers and researchers building or validating automated acoustic species classifiers.
▪️ Ecological consultants, regulators, NGOs, and monitoring leads using automated species identification outputs.

⏱️ ~20 minutes | 🔒 Anonymous | 📝 Informing future best practice and guidance

Take part here ➡️ https://docs.google.com/forms/d/e/1FAIpQLSdaHB8fJuSptUoMHDbfOSTU4pOBVCp7D0zLOfCzz76SYEZDtw/viewform

Please share with relevant colleagues or networks — broad input across taxa and application areas is essential.

Hi!

My research has a localization component and I am running into an issue with signal alignment. The internal clocks in the ARUs I used (Song Meter Micro) drift, so recordings that were programmed to be synchronized are out of phase. Does anyone have experience with/recommendations for synchronization of audio recordings?

Thanks!

At Bat Conservation Trust, we are gathering evidence to support research into storage strategies for passive acoustic monitoring (PAM) project data. Would you be able to spare 10 minutes of your time to fill in our anonymous survey and contribute to our research?

Survey on storage of passive acoustic monitoring data – Fill out form

The work is part of the National Woodland Bat Survey, a collaboration between Bat Conservation Trust and Forest Research, funded by the UK government through Defra’s Natural Capital and Ecosystem Assessment programme (NCEA) with support from Bat Conservation Trust. We aim to use the results from the survey in a peer reviewed journal article.

Any questions at all please let us know. We thank you very much for taking the time to give your valuable input into this work.

We would be very grateful if you could share within your networks. If you do not work with passive acoustic monitoring data, please forward on to a colleague or contact who might be able to contribute.

Hi, I recently annotated a couple of MP3 files downloaded from Xeno Canto, with bounding boxes in Raven Lite (v.2.0.5). When loading that MP3 and its annotations into Python (via soundfile, torchaudio, or librosa), the annotations didn’t align properly with the audio. It seems to be due to the encoder delay on MP3 files, which Raven Lite is treating differently vs. other softwares.

Checking in with my colleagues, no one knew about this particular behavior but there was general agreement that it’s best to work in WAV across softwares. It got me wondering: is this behavior in Raven widely known? Are there similar situations others have encountered, regarding interoperability/file types/etc? Any best practices that you follow when creating datasets? Thanks!

[ MESSAGE TEXT EMPTY ]

Dan Stowell (https://aiforconservation.slack.com/team/U04LE13E35J)

<p>Delighted to officially publish the results of BioDCASE 2026! Six different tasks in animal sound, each with methods proposed by multiple teams. Dive in to the latest state of the art here: <a href="https://biodcase.github.io/challenge2026/">biodcase.github.io/challenge2026</a></p>

Slack Conversation

Dipper Desktop PAM software
Hi friends! I’ve been working with @Matt Weldy and the Kitzes Lab on a desktop GUI for efficiently running bioacoustics foundation models on PAM datasets, selecting detections for review, and rapidly annotating detected clips. It’s called Dipper and is in active development, so I would love to hear feedback if you have feature requests, suggestions, or bug reports (as github issues). The app supports running all major foundation models (BirdNET, Perch 2, HawkEars, etc), runs on Windows/Mac/Linux, uses GPU to accelerate inference when available, and can be deployed in Server mode for remote web-app access. I see it as complimentary to the existing (and awesome) BirdNET GUI and Chirpity tools, with Dipper’s focus being on large-scale data workflows and efficient keyboard-based annotation when you need to review thousands of clips (e.g.: I need verified detection histories from heaps of PAM data). Looking forward to hearing from you if you try it out!
sammlapp.github.io/dipper

We had a paper using a tiny CNN to detect bomb fishing come out this week. Copying my click baity linked in post below for anyone interested!

Bomb dropped on global coral reef hotspot once every 62 minutes 👇 🪸 💣

Last year the public were shocked to see how destructive bottom trawling is after watching David Attenborough's Ocean documentary. Yet did you know people literally throw bombs onto coral reefs to catch fish?

🌏 - Bomb fishing has been reported in 34 countries, and is a devastatingly effective way to cause ecological collapse. Repeat bomb fishing turns thriving reefs into moonlike rubble fields that are unable to recover, pictured below.

🎙️ - We deployed a single Open Acoustic Devices audio recorder on a rolling basis in an Indo-pacific coral reef and trained a tiny AI model to listen for bombs. Across 3650 hrs of audio we detected and verified 3570 blasts. Thats almost 8500 annually, enough to destroy three football pitches of reef every year - heard by just this one recorder!

⚠️ - We show the technology barrier to monitoring this practice is no longer the bottleneck. Baselines can easily be gethered from any impacted area and used to measure the progress of interventions. Targeted interventions are now the challenge, every new location requires carefully balanced community work from the ground up, alternative livlihood campaigns and enforcement.

Our free to read open access paper: zslpublications.onlinelibrary.wiley.com/doi/10.1002/…

image.png

image.png

image.png