HSF+IML Joint Seminar - WhAM (Whale Acoustics Model)

Wednesday 14 Oct 2026, 16:00 → 17:00 Europe/Zurich

40/S2-D01 - Salle Dirac (CERN)

 

The HEP Software Foundation

The HEP Software Foundation facilitates cooperation and common efforts in High Energy Physics software and computing internationally. 
There are several focus activity areas: Data analysis, detector simulation, physics generators, pyHEP, Julia in HEP, reconstruction and software triggers, software developer tools and packaging, and training.
The HSF hosts workshops, provides strategic input and generally raises awareness on the importance of HEP software.

Sign up to HSF mailing lists here.

HSF Seminars

HSF Seminars are nominally hosted on the last Wednesday of each month. Presentations are recorded (given consent) and put on YouTube.
See the full series on indico here!

Sign up to announcements for future HSF seminars as well as other HSF events by joining hsf-forum@googlegroups.com (sign up instructions).

Have a seminar topic idea? Send an email to hsf-seminar-conveners@googlegroups.com (current organisers are Nicole Skidmore, Michel Jouvin and Claire Antel)

 

IML - Inter-Experimental LHC Machine Learning Working Group

IML organizes monthly meetings on a variety of subjects. These meetings are often topic-oriented (focusing on a certain ML technique) and may include external experts. Each year IML organizes an annual workshop typically comprised of roughly 300 participants, which includes invited data scientist’s talks, submitted talks, and tutorials. This year's IML workshop is happening 9-13th Nov (next month!).

IML also serves as entry point to find LHC specific machine learning resources, such as software solutions for machine learning starting from the common ROOT file format. We build a forum for community driven summaries of software solutions, announce LHC tailored trainings/school, and list relevant papers and people involved. We can help finding temporary hardware resources (GPUs) for tests. We are currently building up a database with benchmarks datasets and challenges in order to better enable testing new methods in our domain against previous ones.

 

16:00 → 16:05 Introduction

16:05 → 16:50 WhAM (Whale Acoustics Model) and the CETI Project

Abstract: Sperm whales communicate in short sequences of clicks known as codas. In this lecture, Orr Paradise will present WhAM (Whale Acoustics Model), the first transformer-based model capable of generating synthetic sperm whale codas from any audio prompt. WhAM is built by finetuning VampNet, a masked acoustic token model pretrained on musical audio, using 10k coda recordings collected over the past two decades. Through iterative masked token prediction, WhAM generates high-fidelity synthetic codas that preserve key acoustic features of the source recordings. We evaluate WhAM's synthetic codas using Fréchet Audio Distance and through perceptual studies with expert marine biologists. On downstream classification tasks including rhythm, social unit, and vowel classification, WhAM's learned representations achieve strong performance, despite being trained for generation rather than classification.

Speaker will be giving the seminar in person.

Unfortunately, the talk wasn't fully recorded, however for those interested, our speaker gave a similar talk back in April, see YouTube link.

Speaker: Orr Paradise (EPFL)