Original title: Large-scale bioacoustic detection using semantic segmentation: a deep learning framework applied to fin whale calls in ocean-bottom seismometer recordings
Jocelyn Japnanto, Alex A. Saoulis, Miriam Romagosa, Rita Leitão +4 · September 2026
Scientists want to track fin whales across vast oceans, but manually listening to endless audio recordings takes too long. Ocean-bottom seismometers, which are sensors placed on the seafloor to study earthquakes, also quietly record months of ocean sounds. The researchers built an artificial intelligence tool to spot fin whale songs in this data. Sound recordings are turned into spectrogramsPictures that show sound frequencies over time, where time is on one axis, frequency on another, and brightness indicates loudness., which are visual pictures of sound frequencies over time. The model inspects each pixel in these images, calculates the probability that the pixel belongs to a 20-Hz whale call, and draws boxes around every detected note. It successfully learned on one sensor type in one region and worked accurately on a completely different sensor type elsewhere without extra training. Scanning 378,912 h of audio across 46 seafloor sites, the system found 6.3 million whale calls. This revealed seasonal singing rhythm changes and large-scale whale activity without needing new ocean hardware.
You are trying to find a specific bird call in a noisy forest. You have a recording from one spot, but you want to know if the same bird is singing in a different forest far away. Normally you would need to listen again from scratch. In this paper, the researchers trained a computer to recognize the whale's 20-Hz call from one ocean region, and it worked in another region without retraining, because the call's pattern is consistent even when the background noise changes.
This work turns thousands of ocean-bottom seismometers, already deployed for geology, into a free whale monitoring network. It lets scientists track fin whale populations across entire ocean basins without deploying new equipment. The resulting catalogue of 6.3 million calls reveals seasonal patterns and could inform shipping lane adjustments and fishing regulations to protect whales.
The abstract doesn't say what the limitations are. It notes that the method detects only 20-Hz fin whale notes and was tested on two deployments, implying that further validation is needed for other species or environments.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Large-scale bioacoustic detection using semantic segmentation: a deep learning framework applied to fin whale calls in ocean-bottom seismometer recordings — Jocelyn Japnanto, Alex A. Saoulis, Miriam Romagosa et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.