Beyond task performance: Decoding bioacoustic embeddings with speech features Ines Nolasco⋆ , Jules Cauzinille, Marius Miron, Gagan Narula, Milad Alizadeh, Emmanuel Fernandez, Matthieu Geist, Ellen Gilsenan-McMahon, Olivier Pietquin, Emmanuel Chemla, Sara Keen‡
arXiv:2606.14662v1 [cs.LG] 12 Jun 2026
Earth Species Project
Abstract Pretrained audio embeddings are standard in bioacoustics, yet little is known about which acoustic features these models encode, nor which are useful for a given task. This hinders transparency and limits extension to rare species or data-scarce domains. Here we reveal which speech-like features are encoded in bioacoustic representations. Using the 88 eGeMAPS features across six taxonomic groups, we apply linear and nonlinear regression probes to quantify which acoustic properties each model captures. Results confirm a “no free lunch” pattern: no single model captures the full feature space. A concatenated embedding achieves the highest performance, suggesting complementary acoustic space coverage across models. Loudness features are best encoded (R2 = 0.76) while F0 is hardest to recover (R2 = 0.33). By cross-referencing recoverability with per-species feature salience (NMI), we derive data-driven model selection guidance for bioacoustics.
Figure 1. Starting from a variety of domains (left), we ask which acoustic features (extracted with openSMILE, top) are represented by which bioacoustic models (bottom).
⋆ Corresponding author: [email protected] ‡ Senior author
1
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
1
Introduction
Computational bioacoustics has been transformed by deep learning, enabling large-scale acoustic monitoring and finer-grained ecological questions (Parker, 2025). Despite strong task performance in species classification and detection, it remains largely unclear which acoustic characteristics drive model decisions—interpretability is limited, and the field is only beginning to address this systematically. A common practice in current pipelines is the use of pretrained embeddings. Models are trained on large, diverse corpora to learn generalpurpose representations that are then reused for downstream tasks (Stowell, 2022). While this transfer learning paradigm has shown strong results—especially in low-data regimes—the internal representations are opaque. Unlike handcrafted acoustic features, which have direct physical interpretations, it is unknown which properties of sound are preserved, amplified, or discarded within embedding spaces, or how they align with interpretable acoustic characteristics. Several pretrained models now exist for non-human vocal analysis, spanning supervised and self-supervised paradigms and trained on speech, general audio, and bioacoustic data. Whether these models learn similar or complementary representations remains an open question. Recent benchmarking initiatives compare models on downstream classification performance (Miron, Robinson, Alizadeh, et al., 2025), but do not directly reveal what information is encoded. Consequently, principled guidelines for model selection are still lacking. In this work, we address the following research question: Which acoustic features are encoded in the embedding representations of deep learning models used in bioacoustics? We extract 88 interpretable acoustic descriptors from the eGeMAPS feature set—covering spectral, temporal, cepstral, and modulation-based properties—from six bioacoustic domains: dogs, mosquitoes, bats, marine mammals, birds, and human speech. We then apply linear and nonlinear regression probes to measure how well these features can be recovered from embeddings produced by pretrained models (see (Belinkov, 2022)). The performance gap between the two serves as an indicator of representational entanglement. To connect recoverability to practical utility, we further compute the Normalised Mutual Information (NMI) between each feature and labels of each dataset. Cross-referencing NMI with probe R2 reveals whether models encode the features that are acoustically salient for a given task, providing a principled basis for model selection beyond benchmark scores. Our contributions are threefold: • A regression probing framework to measure feature recoverability from pretrained embeddings, enabling comparison of what models encode rather than task performance alone. • An analysis of 6 pretrained models across 88 eGeMAPS features and six bioacoustic datasets, revealing substantial cross-model variation and complementarity. • A cross-referencing of feature recoverability with per-species feature salience (NMI), providing practical insights into model fitness for specific bioacoustic tasks.
2
Background
Early machine listening and speech research focused on identifying measurable signal properties such as frequency, amplitude and temporal structure, that provide compact, interpretable representations of complex sounds (Sharma, Umapathy, and Krishnan, 2020). To standardise extraction, the openSMILE toolkit (Eyben, Wöllmer, and B. Schuller, 2010) was developed, enabling reproducible feature computation across benchmarks such as the INTERSPEECH Computational Paralinguistics Challenge (B. Schuller et al., 2013). Within this framework, the Geneva Minimalistic Acoustic Parameter Set (GeMAPS) and its extended version (eGeMAPS) were curated as a principled, minimal feature set for speech-related tasks, particularly emotion recognition. Although developed for speech, these features transfer well to bioacoustics: there is substantial overlap in informative acoustic properties across human and non-human vocalisations (Torrisi et al., 2026; Fagerlund, 2007), and Mel-frequency Cepstral Coefficients (MFCCs) achieved early successes in bioacoustic classification (Deecke and Janik, 2006; Somervuo, Harma, and Fagerlund, 2006). Nevertheless, bioacoustic signals can differ markedly from speech. For instance, bats vocalise in the ultrasonic range (Jones and Holderied, 2007), elephants use infrasonic rumbles (Payne, Langbauer Jr, and Thomas, 1986), and insects produce sound through stridulation rather than vocal tracts (Alexander, 1957). Furthermore, bioacoustic recordings are often made in natural environments, with overlapping species and varying background noise conditions. Finding representations that generalise across taxa and deployment conditions therefore remains challenging (Stowell, 2018).
2
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
With the rise of deep learning, the field moved away from handcrafted features toward representations learned directly from data. Pretrained embeddings have become the dominant paradigm: a model is trained on large corpora and its frozen encoder representations are reused for downstream tasks with minimal additional training (Stowell, 2022). Success cases (Ghani et al., 2023; Best, Paris, et al., 2023) have motivated broad adoption, but results are mixed and difficult to predict, particularly under domain mismatch between pretraining and target data (Miron, Robinson, Alizadeh, et al., 2025). Benchmarking efforts to guide model selection have been established in both speech SUPERB (Yang et al., 2021), HEAR (Turian et al., 2022) and, more recently, bioacoustics (Miron, Robinson, Alizadeh, et al., 2025; Kather, Ghani, and Stowell, 2025; Schwinger et al., 2025). These evaluations typically extract embeddings from frozen encoders and assess them with a simple classification head. The choice of classification head is key: shallow multilayer perceptrons (MLPs) allow some non-linear adaptation of the embedding space (Turian et al., 2022), while linear probes (Belinkov, 2022) impose stricter constraints. Clustering or retrieval methods require no training at all, directly reflecting the geometry of the embedding space (Miron, Robinson, Alizadeh, et al., 2025; Kather, Ghani, and Stowell, 2025). This shift from handcrafted to learned features has come at the cost of interpretability. Understanding what acoustic patterns models exploit is important both for transparency in applied conservation contexts (Parker, 2025) and for scientific discovery, since learnt representations can reveal biologically meaningful acoustic properties relevant to animal communication and evolution studies. In speech, probing studies have established that self-supervised models such as wav2vec 2.0 encode information hierarchically across layers: shallow layers capture acoustic properties such as fundamental frequency and formants, middle layers encode phonetic content, and deeper layers represent word identity and semantics (Pasad, Chou, and Livescu, 2021; Pasad, B. Shi, and Livescu, 2023; Choi and Yeo, 2022). A clear layer-wise correspondence with MFCCs has also been demonstrated (Raymondaud, Rouvier, and Dufour, 2024). In bioacoustics, this research direction is still underdeveloped. An important insight from Kather, Ghani, and Stowell (2025) links training paradigm to representation content, finding that supervised models tend to produce better-structured clusters, supporting species classification, while self-supervised models encode more general characteristics of the sounds and thus allow better generalization across domains. Beyond this, it remains unclear what different models actually encode, whether their representations are complementary or redundant, and, critically, whether they capture the features that are most discriminative for particular taxonomic groups. On the whole, our current understanding of bioacoustic embeddings lags far behind that of speech encoding. Addressing this gap is the central aim of the present work.
3
Materials and Methods
Fig. 1 depicts our methodological framework. We use six bioacoustic audio datasets spanning diverse taxa (§3.1). For each recording, we extract a set of interpretable acoustic features (§3.1), as well as embeddings from six pretrained models (§3.2). We then conduct three complementary experiments: 1) apply regression probes to determine how recoverable features are from each embedding space (Emb2Feat); 2) apply cross-model probes to determine overlap in what information is encoded by each model (Emb2Emb); and 3) calculate importance of interpretable features and compare this against feature recoverability, giving insight into how well relevant features have been encoded in each embedding (FeatImportance). 3.1
Dataset and feature extraction with openSMILE
Our data is extracted from the training split of the BEANS benchmark dataset described and available in Hagiwara et al. (2023). The selected sets support a range of tasks from species to individual classification (see Table 1). Interpretable acoustic features were extracted with openSMILE (Eyben, Wöllmer, and B. Schuller, 2010) (§2). We used the eGeMAPS feature set which defines 88 global acoustic features that are further categorised in seven broader types (described in Table 2). 3.2
Models
We extract pre-trained embeddings from the last layer of six different models that are commonly used in bioacoustics research and have been the subject of recent evaluation in Miron, Robinson, Alizadeh, et al. (2025). Models were pretrained on various domains and with different learning paradigms (see Table 3). Extraction
3
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
Table 1. Dataset summary
Dataset
# Audio
Task
# Class
Dogs
414
IndividualID
10
Bats
5987
IndividualID
10
CBI – birds
14206
Species
264
MM – marine mammals
1004
Species
31
Mosquitoes
5407
Species
14
Speech
7036
Spoken words
35
Total
34054
-
-
Table 2. eGeMAPSv02 feature groups used in this study (88 features total). Group
Description
# Feat.
F0
Fundamental frequency level, range and slope
10
Loudness
Perceptual loudness level, range and long-term energy
11
Harmonicity
Jitter, shimmer, harmonics-to-noise ratio, harmonic amplitude ratios
10
Spectral Shape
Spectral tilt (alpha ratio, Hammarberg index), spectral slopes, spectral flux
17
Formants
F1, F2, F3 frequency, bandwidth, and amplitude (voiced frames)
18
MFCC
Mel-frequency cepstral coefficients 1–4 (all frames and voiced frames)
16
Temporal
Voiced/unvoiced segment rate and duration, loudness peak rate
6
and access to the models’ checkpoints is done with the AVEX API1 . Embeddings are averaged on the time dimension for the whole clip resulting in a single vector per sample. Table 3. Summary of Audio encoder models. Model
Archit.
Learn.
Training Domain
Dim
BEATS_base (Chen et al., 2022)
Transf.
SSL
general audio
768
NatureLM (Robinson et al., 2024)
Transf.
SSL + LLM
Bio + Speech + Music
768
BirdMAE (Rauch et al., 2025)
Transf.
SSL
Bio
1280
BirdNET (Kahl et al., 2021)
CNN
Sup
Bio(Birds)
1024
EffNet_all (Miron, Robinson, Alizadeh, et al., 2025)
CNN
Sup
Bio + general audio
1280
Perch (Ghani et al., 2023)
CNN
Sup
Bio(Birds)
1280
3.3
Linear and non-linear regression probes
We define two regression probes. For the linear probe, we fit a ridge regression model using scikit-learn2 that learns a linear mapping from the embeddings vector to each feature value. The non-linear probe is implemented as a shallow multi-layer perceptron consisting of a single hidden layer with 256 units, ReLU activation, and dropout (p = 0.2), followed by a linear output unit. The network is optimised with Adam (lr = 0.001) using MSE loss, batch size 128 and early stopping with patience 20 (maximum 100 epochs). For both probes, input embeddings and target features are Z-score normalised. Predictions are inverse-transformed to the original scale before computing the evaluation metrics. We evaluate regression performance under 1 https://github.com/earthspecies/avex 2 https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.Ridge.html
4
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
5-fold cross-validation stratified by dataset and report per-feature coefficients of determination R2 averaged across folds. 3.4
Determining feature relevance for tasks with NMI
To measure relevance of interpretable acoustic features, we computed normalized mutual information (NMI) for each dataset. This quantifies how well samples with different labels could be distinguished using each ) feature (higher NMI indicates higher feature importance for the task). NMI(X; Y ) = I(X;Y H(Y ) where I(X; Y ) is the mutual P information between a continuous acoustic feature X and the discrete class label Y , and H(Y ) = − c p(c) log p(c) is the Shannon entropy of the label distribution (in nats). MI is estimated with a k-nearest neighbour entropy estimator (k = 3).
4
Results
4.1
Emb2Feat: Recoverability of features
Figure 2. Mean R² scores for linear decoding of eGeMAPS feature categories (dots). Vertical stems show non-linear gain and the stem cap marks the MLP R² value. Each line corresponds to one of six pretrained audio embedding models and all embeddings concatenated. X-axis shows feature categories. Higher R² values indicate greater decodability of feature category from the embedding space.
Results of recoverability (with linear and non-linear probes) of feature types for different models are presented in Fig. 2. Overall, we observe that BirdMAE and BEATS_base embeddings are the best encoders across all feature types. This tendency might be explained by the ability of SSL models to encode general acoustic properties as suggested in (Kather, Ghani, and Stowell, 2025). However, due to the differences in training domain and architectures this warrants further investigation. For most of the feature types, concatenation of all models (dashed line) provides the best representation; this suggests that models may be complementary in what they encode. The general small improvement in R2 (max +0.08) when applying a nonlinear setup might indicate that the non-linearity applied is still very shallow for the level of entanglement that these embeddings present. Given these results, we proceed considering only the linear case. Regarding feature types, loudness- and spectral shape-related features are recovered best across all models while F0-related features are worst. Zooming in on individual features, Fig. 3 presents the R2 values across models and datasets for the union of the top 5 linearly recoverable features across all models. From here we can first observe the high variation in feature recoverability across models, but also across datasets within the same model (e.g., BirdNET and
5
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
Figure 3. Union of the top5 linearly decodable eGeMAPS features for each model, shown across datasets (rows) and models (columns). Colour represents mean R² for each feature-dataset-model combination, illustrating which acoustic characteristics are most recoverable from each embedding.
6
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
Perch). It is also of note that several MFCC-related features are present, but not a single F0. The results on the concatenated model and the overall heterogeneity observed leads us to analyse how embeddings relate to each other and in particular whether some models might be able to directly predict other models’ embeddings. 4.2
Emb2Emb
Figure 4. Pairwise embedding predictability across all models. Cells show mean R² for predicting one embedding (target, columns) from another (source, rows), using a linear probe trained on combined bioacoustic and speech datasets. Higher values indicate greater overlap in encoded information between model pairs.
Fig. 4 shows that no single model is best suited to predict others, although BirdMAE and EffNet_all seem to be partially predictable by most. In contrast, BirdNET is the least predictable model. These results indicate that each pre-training dataset, objective, and architecture generate rather unique feature extraction capabilities, warranting further investigation, as we explore in the following section. 4.3
FeatImportance
In Fig. 5, we cross-reference the most acoustically salient features per task (measured by NMI) with each model’s ability to encode them (measured by R2 ). Salient features vary considerably across taxonomic groups: birds and dogs rely more heavily on F0-related features; mosquitoes and bats on loudness; and marine mammals and speech commands on MFCCs. Consistent with earlier results, loudness features are generally well-encoded while F0 and several MFCC features remain hard to recover linearly from the embeddings of the last layer. This supports the observation that half of the top-10 salient features for speech commands are not reliably encoded by any individual model. Notably, concatenated embeddings do not always result in the highest recoverability of task-relevant features (e.g., Birds, Bats and Speech); these results might in part be explained by the high dimension of concatenated embeddings, which can easily overfit and degrade regression performance.
5
Final remarks
We propose an evaluation framework of embedding representations based on acoustic feature content. Our results confirmed that no single model captures the full breadth of the eGeMAPS features, and that models encode complementary rather than redundant representations; however, concatenation of models is not always the solution to augment representation power. Cross-referencing recoverability with per-species feature salience (NMI) further revealed that task-relevant features are not always well-encoded. These results suggest
7
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
Figure 5. Task relevance versus embedding decodability of the top10 eGeMAPS features per dataset. The x-axis shows NMI; the y-axis shows mean R2 from linear probing. Grey bars indicate features that are both task-relevant and linearly decodable; red bars indicate task-relevant features with low linearly decodability, highlighting gaps in embedding representations.
8
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
that strategies compensating for weak representation points, such as selecting models known to encode tasksalient features, or concatenating embeddings that cover complementary acoustic properties, could directly improve downstream classification performance in bioacoustics. Our study has several limitations. First, eGeMAPS features may be poorly extracted from signals, biasing ground truth labels. In this scenario, low R2 values could indicate poor feature extraction, rather than poor feature encoding. Relatedly, the eGeMAPS feature set is optimized for human speech and may not capture the most relevant properties of non-human signals. The systematically low F0 recoverability may partly reflect unreliable F0 extractors rather than absent encoding (Best, Araya-Salas, et al., 2025). This hypothesis motivates the need for bioacoustic-specific feature sets that can serve as ground truths in future evaluation studies. Second, pooling embeddings across time can result in a loss of temporal information, especially with non-transformer based models and it may skew recoverability estimates of time-varying features. Finally, our study did not include layer-wise analyses potentially overlooking layer-wise encoding dynamics described in speech-based models (Cauzinille et al., 2025; Pasad, B. Shi, and Livescu, 2023). We recommend future studies further investigate the relationship between feature decodability, feature importance, and also include model performance on downstream tasks. As it stands, our framework demonstrates how recoverable features are, but does not yet indicate which features models rely on to perform classification. This would be the next step towards complete interpretable systems. These axes of evaluation have been formalised in the evaluation framework proposed in Plachouras et al. (2025). Probing randomly initialized models would further help disentangle the role of architectural inductive biases from learned representations (Ulyanov, Vedaldi, and Lempitsky, 2018), and combining our proposed framework with ablation or attention-based analyses could further clarify the link between embedding content and model behaviour (Miron, Robinson, Hagiwara, et al., 2026). Taken together, our findings contribute to a more interpretable and principled use of bioacoustic embeddings, moving beyond benchmarks to provide content-based guidance for model selection and expanding our understanding of what models have learned about the acoustic structure of non-human communication.
9
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
References Alexander, Richard D (1957). “Sound production and associated behavior in insects”. In. Belinkov, Yonatan (2022). “Probing classifiers: Promises, shortcomings, and advances”. In: Computational Linguistics 48.1, pp. 207–219. Best, Paul, Marcelo Araya-Salas, Axel G Ekström, Bárbara Freitas, Frants H Jensen, Arik Kershenbaum, Adriano R Lameira, Kenna DS Lehmann, Pavel Linhart, Robert C Liu, et al. (2025). “Bioacoustic fundamental frequency estimation: a cross-species dataset and deep learning baseline”. In: Bioacoustics 34.4, pp. 419– 446. Best, Paul, Sébastien Paris, Hervé Glotin, and Ricard Marxer (2023). “Deep audio embeddings for vocalisation clustering”. In: Plos one 18.7, e0283396. Cauzinille, Jules, Marius Miron, Olivier Pietquin, Masato Hagiwara, Ricard Marxer, Arnaud Rey, and Benoit Favre (2025). “Crossing the Species Divide: Transfer Learning from Speech to Animal Sounds”. In: arXiv preprint arXiv:2509.04166. Chen, Sanyuan, Yu Wu, Chengyi Wang, Shujie Liu, Daniel Tompkins, Zhuo Chen, and Furu Wei (2022). “Beats: Audio pre-training with acoustic tokenizers”. In: arXiv preprint arXiv:2212.09058. Choi, Kwanghee and Eun Jung Yeo (2022). “Opening the black box of wav2vec feature encoder”. In: arXiv preprint arXiv:2210.15386. Deecke, Volker B and Vincent M Janik (2006). “Automated categorization of bioacoustic signals: avoiding perceptual pitfalls”. In: The Journal of the Acoustical Society of America 119.1, pp. 645–653. Eyben, Florian, Martin Wöllmer, and Björn Schuller (2010). “Opensmile: the munich versatile and fast opensource audio feature extractor”. In: Proceedings of the 18th ACM international conference on Multimedia, pp. 1459–1462. Fagerlund, Seppo (2007). “Bird species recognition using support vector machines”. In: EURASIP Journal on Advances in Signal Processing 2007.1, p. 038637. Ghani, Burooj, Tom Denton, Stefan Kahl, and Holger Klinck (2023). “Global birdsong embeddings enable superior transfer learning for bioacoustic classification”. In: Scientific Reports 13.1, p. 22876. Hagiwara, Masato, Benjamin Hoffman, Jen-Yu Liu, Maddie Cusimano, Felix Effenberger, and Katie Zacarian (2023). “Beans: The benchmark of animal sounds”. In: ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, pp. 1–5. Jones, Gareth and Marc W Holderied (2007). “Bat echolocation calls: adaptation and convergent evolution”. In: Proceedings of the Royal Society B: Biological Sciences 274.1612, pp. 905–912. Kahl, Stefan, Connor M Wood, Maximilian Eibl, and Holger Klinck (2021). “BirdNET: A deep learning solution for avian diversity monitoring”. In: Ecological Informatics 61, p. 101236. Kather, Vincent S, Burooj Ghani, and Dan Stowell (2025). “Clustering and novel class recognition: evaluating bioacoustic deep learning feature extractors”. In: arXiv preprint arXiv:2504.06710. Miron, Marius, David Robinson, Milad Alizadeh, Ellen Gilsenan-McMahon, Gagan Narula, Emmanuel Chemla, Maddie Cusimano, Felix Effenberger, Masato Hagiwara, Benjamin Hoffman, et al. (2025). “What matters for bioacoustic encoding”. In: arXiv preprint arXiv:2508.11845. Miron, Marius, David Robinson, Masato Hagiwara, Parcollet Titouan, Jules Cauzinille, Gagan Narula, Milad Alizadeh, Ellen Gilsenan-McMahon, Sara Keen, Emmanuel Chemla, Benjamin Hoffman, Maddie Cusimano, Diane Kim, Felix Effenberger, Jane K. Lawton, Aza Raskin, Olivier Pietquin, and Matthieu Geist (2026). “Multi-layer attentive probing improves transfer of audio representations for bioacoustics”. In: ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, pp. 1–5. Parker, James EK (2025). “The Planetization of Machine Listening”. In: Critical Inquiry 52.1, pp. 21–47.
10
Earth Species Project
DECODING BIOACOUSTIC EMBEDDINGS
Pasad, Ankita, Ju-Chieh Chou, and Karen Livescu (2021). “Layer-wise analysis of a self-supervised speech representation model”. In: 2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, pp. 914–921. Pasad, Ankita, Bowen Shi, and Karen Livescu (2023). “Comparative layer-wise analysis of self-supervised speech models”. In: ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, pp. 1–5. Payne, Katharine B, William R Langbauer Jr, and Elizabeth M Thomas (1986). “Infrasonic calls of the Asian elephant (Elephas maximus)”. In: Behavioral Ecology and Sociobiology 18.4, pp. 297–301. Plachouras, Christos, Julien Guinot, George Fazekas, Elio Quinton, Emmanouil Benetos, and Johan Pauwels (2025). “Towards a Unified Representation Evaluation Framework Beyond Downstream Tasks”. In: arXiv preprint arXiv:2505.06224. Rauch, Lukas, Ilyass Moummad, René Heinrich, Alexis Joly, Bernhard Sick, and Christoph Scholz (2025). “Can Masked Autoencoders Also Listen to Birds?” In: arXiv:2504.12880. Raymondaud, Quentin, Mickael Rouvier, and Richard Dufour (2024). “Probing the information encoded in neural-based acoustic models of automatic speech recognition systems”. In: arXiv preprint arXiv:2402.19443. Robinson, David, Marius Miron, Masato Hagiwara, Benno Weck, Sara Keen, Milad Alizadeh, Gagan Narula, Matthieu Geist, and Olivier Pietquin (2024). “NatureLM-audio: An audio-language foundation model for bioacoustics”. In: arXiv preprint arXiv:2411.07186. Schuller, Björn, Stefan Steidl, Anton Batliner, Alessandro Vinciarelli, Klaus Scherer, Fabien Ringeval, Mohamed Chetouani, Felix Weninger, Florian Eyben, Erik Marchi, et al. (2013). “The INTERSPEECH 2013 computational paralinguistics challenge: Social signals, conflict, emotion, autism”. In: Proceedings INTERSPEECH 2013, 14th Annual Conference of the International Speech Communication Association, Lyon, France. Schwinger, Raphael, Paria Vali Zadeh, Lukas Rauch, Mats Kurz, Tom Hauschild, Sam Lapp, and Sven Tomforde (2025). “Foundation Models for Bioacoustics–a Comparative Review”. In: arXiv preprint arXiv:2508.01277. Sharma, Garima, Kartikeyan Umapathy, and Sridhar Krishnan (2020). “Trends in audio signal feature extraction methods”. In: Applied Acoustics 158, p. 107020. Somervuo, Panu, Aki Harma, and Seppo Fagerlund (2006). “Parametric representations of bird sounds for automatic species recognition”. In: IEEE Transactions on Audio, Speech, and Language Processing 14.6, pp. 2252–2263. Stowell, Dan (2018). “Computational Bioacoustic Scene Analysis”. In: Computational Analysis of Sound Scenes and Events. Ed. by Tuomas Virtanen, Mark D. Plumbley, and Dan Ellis. Vol. 9. Springer Topics in Signal Processing. Springer, pp. 303–333. doi: 10.1007/978-3-319-63450-0_11. —
(2022). “Computational bioacoustics with deep learning: a review and roadmap”. In: PeerJ 10, e13152.
Torrisi, Antonella, Inês Nolasco, Paola Sgadò, Elisabetta Versace, and Emmanouil Benetos (2026). “Embryonic Exposure to VPA Influences Chick Vocalisations: A Computational Study”. In: arXiv preprint arXiv:2601.12203. Turian, Joseph, Jordie Shier, Humair Raj Khan, Bhiksha Raj, Björn W Schuller, Christian J Steinmetz, Colin Malloy, George Tzanetakis, Gissel Velarde, Kirk McNally, et al. (2022). “Hear: Holistic evaluation of audio representations”. In: NeurIPS 2021 Competitions and Demonstrations Track. PMLR, pp. 125–145. Ulyanov, Dmitry, Andrea Vedaldi, and Victor Lempitsky (2018). “Deep image prior”. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 9446–9454. Yang, Shu-wen, Po-Han Chi, Yung-Sung Chuang, Cheng-I Jeff Lai, Kushal Lakhotia, Yist Y Lin, Andy T Liu, Jiatong Shi, Xuankai Chang, Guan-Ting Lin, et al. (2021). “Superb: Speech processing universal performance benchmark”. In: arXiv preprint arXiv:2105.01051.
11
Earth Species Project