Ultrasonic vocalizations (USV) convey information about individual identity and arousal status in mice. We propose to track USV as ridges in the time–frequency domain via a variant of time– frequency reassignment (TFR). The key idea is to perform TFR with empirical Wiener shrinkage and multitapering to improve robustness to noise. Furthermore, we perform TFR over both the short-term Fourier transform and the constant-Q transform so as to detect both the fundamental frequency and its harmonic partial (if any). Experimental results show that our approach effectively estimates multicomponent ridges with high precision and low frequency deviation.

STONE, the current method in self-supervised learning for tonality estimation in music signals, cannot distinguish relative keys, such as C major versus A minor. In this article, we extend the neural network architecture and learning objective of STONE to perform self-supervised learning of major and minor keys (S-KEY). Our main contribution is an auxiliary pretext task to STONE, formulated using transposition-invariant chroma features as a source of pseudo-labels. S-KEY matches the supervised state of the art in tonality estimation on FMAKv2 and GTZAN datasets while requiring no human annotation and having the same parameter budget as STONE. We build upon this result and expand the training set of S-KEY to a million songs, thus showing the potential of large-scale self-supervised learning in music information retrieval.

STONE, which stands for self-supervised tonality estimator, has recently demonstrated the practical feasibility of recognizing key signatures in music signals given little or no human annotation. In this article, we revisit STONE from a more theoretical standpoint. We show that cross-power spectral density (CPSD) defines a differentiable measure of harmonic discrepancy between key signature profiles (KSP). Having set the CPSD frequency to seven cycles per octave, we offer a geometric interpretation of this discrepancy via the circle of fifths and conduct an algebraic study to prove that all its local minima are global. We rely on the equivariance property of deep convolutional networks to prove that the STONE loss function is invariant to circular frequency shifts of the constant-Q transform. We conclude by identifying a phenomenon of spontaneous symmetry breaking in STONE: since modes of limited transposition (e.g., augmented, diminished, whole-tone) are multistable in the circle of fifths, the associated CPSD gradient is driven towards more “tonal” (i.e., asymmetric) scales, such as diatonic or pentatonic.

International audience

What makes waveform-based deep learning so hard? Despite numerous attempts at training convolutional neural networks (convnets) for filterbank design, they often fail to outperform hand-crafted baselines. These baselines are linear time-invariant systems: as such, they can be approximated by convnets with wide receptive fields. Yet, in practice, gradient-based optimization leads to suboptimal results. In our article, we approach this problem from the perspective of initialization. We present a theory of large deviations for the energy response of FIR filterbanks with random Gaussian weights. We find that deviations worsen for large filters and locally periodic input signals, which are both typical for audio signal processing applications. Numerical simulations align with our theory and suggest that the condition number of a convolutional layer follows a logarithmic scaling law between the number and length of the filters, which is reminiscent of discrete wavelet bases.

Perceptual sound matching (PSM) aims to find the input parameters to a synthesizer so as to best imitate an audio target. Deep learning for PSM optimizes a neural network to analyze and reconstruct prerecorded samples. In this context, our article addresses the problem of designing a suitable loss function when the training set is generated by a differentiable synthesizer. Our main contribution is perceptual–neural–physical loss (PNP), which aims at addressing a tradeoff between perceptual relevance and computational efficiency. The key idea behind PNP is to linearize the effect of synthesis parameters upon auditory features in the vicinity of each training sample. The linearization procedure is massively parallelizable, can be precomputed, and offers a 100-fold speedup during gradient descent compared to differentiable digital signal processing (DDSP). We demonstrate PNP on two datasets of nonstationary sounds: an AM/FM arpeggiator and a physical model of rectangular membranes. We show that PNP is able to accelerate DDSP with joint time–frequency scattering transform (JTFS) as auditory feature while preserving its perceptual fidelity. Additionally, we evaluate the impact of other design choices in PSM: parameter rescaling, pretraining, auditory representation, and gradient clipping. We report state-of-the-art results on both datasets and find that PNP-accelerated JTFS has greater influence on PSM performance than any other design choice.

Sound event classification has the potential to advance our understanding of bird migration. Although it is long known that migratory species have a vocal signature of their own, previous work on automatic flight call classification has been limited in robustness and scope: e.g., covering few recording sites, short acquisition segments, and simplified biological taxonomies. In this paper, we present BirdVoxDetect (BVD), the first full-fledged solution to bird migration monitoring from acoustic sensor network data. As an open-source software, BVD integrates an original pipeline of three machine learning modules. The first module is a random forest classifier of sensor faults, trained with human-in-the-loop active learning. The second module is a deep convolutional neural network for sound event detection with per-channel energy normalization (PCEN). The third module is a multitask convolutional neural network which predicts the family, genus, and species of flight calls from passerines(Passeriformes)of North America. We evaluate BVD on a new dataset (296 hours from nine locations, the largest to date for this task) and discuss the main sources of estimation error in a real-world deployment: mechanical sensor failures, sensitivity to background noise, misdetection, and taxonomic confusion. Then, we deploy BVD to an unprecedented scale: 6672 hours of audio (approximately one terabyte), corresponding to a full season of bird migration. Running BVD in parallel over the full-season dataset yields 1.6 billion FFT's, 480 million neural network predictions, and over six petabytes of throughput. With this method, our main finding is that deep learning and bioacoustic sensor networks are ready to complement radar observations and crowdsourced surveys for bird migration monitoring, thus benefiting conservation ecology and land-use planning at large.

Sound matching algorithms seek to approximate a target waveform by parametric audio synthesis. Deep neural networks have achieved promising results in matching sustained harmonic tones. However, the task is more challenging when targets are nonstationary and inharmonic, e.g., percussion. We attribute this problem to the inadequacy of loss function. On one hand, mean square error in the parametric domain, known as "P-loss", is simple and fast but fails to accommodate the differing perceptual significance of each parameter. On the other hand, mean square error in the spectrotemporal domain, known as "spectral loss", is perceptually motivated and serves in differentiable digital signal processing (DDSP). Yet, spectral loss is a poor predictor of pitch intervals and its gradient may be computationally expensive; hence a slow convergence. Against this conundrum, we present Perceptual-Neural-Physical loss (PNP). PNP is the optimal quadratic approximation of spectral loss while being as fast as P-loss during training. We instantiate PNP with physical modeling synthesis as decoder and joint time–frequency scattering transform (JTFS) as spectral representation. We demonstrate its potential on matching synthetic drum sounds in comparison with other loss functions.

Deep convolutional networks (convnets) in the time–frequency domain can learn an accurate and fine-grained categorization of sounds. For example, in the context of music signal analysis, this categorization may correspond to a taxonomy of playing techniques: vibrato, tremolo, trill, and so forth. However, convnets lack an explicit connection with the neurophysiological underpinnings of musical timbre perception. In this article, we propose a data-driven approach to explain audio classification in terms of physical attributes in sound production. We borrow from current literature in "explainable AI" (XAI) to study the predictions of a convnet which achieves an almost perfect score on a challenging task: i.e., the classification of five comparable real-world playing techniques from 30 instruments spanning seven octaves. Mapping the signal into the carrier-modulation domain using scattering transform, we decompose the networks’ predictions over this domain with layer-wise relevance propagation. We find that regions highly-relevant to the predictions localized around the physical attributes with which the playing techniques are performed.

Supervised machine learning for music information retrieval requires a large annotated training set, and is thus an expensive and time-consuming process. To circumvent this problem, we propose to train deep neural networks to perceive beats in musical recordings despite having little or no access to human annotations. The key idea is to train two fully convolutional networks in parallel, which we name “Zero-Note Samba” (ZeroNS): the first analyzes the percussive part of a musical piece whilst the second analyzes its non-percussive part. These networks learn a self-supervised pretext task of synchrony prediction (sync-pred), which simulates the ability of musicians to groove together when playing in the same band. Sync-pred encourages the two networks to return similar outputs if the underlying musical parts are synchronized, yet dissimilar outputs if the parts are out of sync. In practice, we obtain the instrumental parts from commercial recordings via an off-the-shelf source separation system: Spleeter. After self-supervised learning with sync-pred, ZeroNS produces a sparse output that resembles a beat detection function. When used in conjunction with a dynamic Bayesian network, ZeroNS surpasses the state of the art in unsupervised beat tracking. Furthermore, fine-tuning ZeroNS to a small set of labeled data (of the order of one to ten songs) matches the performance of a fully supervised network on 96 songs. Lastly, we show that pre-training a supervised model with sync-pred mitigates dataset bias and thus improves cross-dataset generalization, at no extra annotation cost.

Playing techniques contain distinctive information about musical expressivity and interpretation. Yet, current research in music signal analysis suffers from a scarcity of computational models for playing techniques, especially in the context of live performance. To address this problem, our paper develops a general framework for playing technique recognition. We propose the adaptive scattering transform, which refers to any scattering transform that includes a stage of data-driven dimensionality reduction over at least one of its wavelet variables, for representing playing techniques. Two adaptive scattering features are presented: frequency-adaptive scattering and direction-adaptive scattering. We analyse seven playing techniques: vibrato, tremolo, trill, flutter-tongue, acciaccatura, portamento, and glissando. To evaluate the proposed methodology, we create a new dataset containing full-length Chinese bamboo flute performances (CBFdataset) with expert playing technique annotations. Once trained on the proposed scattering representations, a support vector classifier achieves state-of-the-art results. We provide explanatory visualisations of scattering coefficients for each technique and verify the system over three additional datasets with various instrumental and vocal techniques: VPset, SOL, and VocalSet.

Class imbalance in the training data hinders the generalization ability of machine listening systems. In the context of bioacoustics, this issue may be circumvented by aggregating species labels into super-groups of higher taxonomic rank: genus, family, order, and so forth. However, different applications of machine listening to wildlife monitoring may require different levels of granularity. This paper introduces TaxoNet, a deep neural network for structured classification of signals from living organisms. TaxoNet is trained as a multitask and multilabel model, following a new architectural principle in end-to-end learning named "hierarchical composition": shallow layers extract a shared representation to predict a root taxon, while deeper layers specialize recursively to lower-rank taxa. In this way, TaxoNet is capable of handling taxonomic uncertainty, out-of-vocabulary labels, and open-set deployment settings. An experimental benchmark on two new bioacoustic datasets (ANAFCC and BirdVox-14SD) leads to state-of-the-art results in bird species classification. Furthermore, on a task of coarse-grained classification, TaxoNet also outperforms a flat single-task model trained on aggregate labels.

To explain the consonance of octaves, music psychologists represent pitch as a helix where azimuth and axial coordinate correspond to pitch class and pitch height respectively. This article addresses the problem of discovering this helical structure from unlabeled audio data. We measure Pearson correlations in the constant-Q transform (CQT) domain to build a K-nearest neighbor graph between frequency subbands. Then, we run the Isomap manifold learning algorithm to represent this graph in a three-dimensional space in which straight lines approximate graph geodesics. Experiments on isolated musical notes demonstrate that the resulting manifold resembles a helix which makes a full turn at every octave. A circular shape is also found in English speech, but not in urban noise. We discuss the impact of various design choices on the visualization: instrumentarium, loudness mapping function, and number of neighbors K.

Playing techniques are important expressive elements in music signals. In this paper, we propose a recognition system based on the joint time-frequency scattering transform (jTFST) for pitch evolution-based playing techniques (PETs), a group of playing techniques with monotonic pitch changes over time. The jTFST represents spectro-temporal patterns in the time-frequency domain, capturing discriminative information of PETs. As a case study, we analyse three commonly used PETs of the Chinese bamboo flute: acciacatura, portamento, and glissando, and encode their characteristics using the jTFST. To verify the proposed approach, we create a new dataset, the CBF-petsDB, containing PETs played in isolation as well as in the context of whole pieces performed and annotated by professional players. Feeding the jTFST to a machine learning classifier, we obtain F-measures of 71% for acciacatura, 59% for portamento, and 83% for glissando detection, and provide explanatory visualisations of scattering coefficients for each technique.

The wavelet scattering transform is an invariant and stable signal representation suitable for many signal processing and machine learning applications. We present the Kymatio software package, an easy-to-use, high-performance Python implementation of the scattering transform in 1D, 2D, and 3D that is compatible with modern deep learning frameworks, including PyTorch and TensorFlow/Keras. The transforms are implemented on both CPUs and GPUs, the latter offering a significant speedup over the former. The package also has a small memory footprint. Source code, documentation, and examples are available under a BSD license at https://www.kymat.io.

In the context of automatic speech recognition and acoustic event detection, an adaptive procedure named per-channel energy normalization (PCEN) has recently shown to outperform the pointwise logarithm of mel-frequency spectrogram (logmelspec) as an acoustic frontend. This letter investigates the adequacy of PCEN for spectrogram-based pattern recognition in far-field noisy recordings, both from theoretical and practical standpoints. First, we apply PCEN on various datasets of natural acoustic environments and find empirically that it Gaussianizes distributions of magnitudes while decorrelating frequency bands. Second, we describe the asymptotic regimes of each component in PCEN: temporal integration, gain control, and dynamic range compression. Third, we give practical advice for adapting PCEN parameters to the temporal properties of the noise to be mitigated, the signal to be enhanced, and the choice of time-frequency representation. As it converts a large class of real-world soundscapes into additive white Gaussian noise, PCEN is a computationally efficient frontend for robust detection and classification of acoustic events in heterogeneous environments.

This article addresses the automatic detection of vocal, nocturnally migrating birds from a network of acoustic sensors. Thus far, owing to the lack of annotated continuous recordings, existing methods had been benchmarked in a binary classification setting (presence vs. absence). Instead, with the aim of comparing them in event detection, we release BirdVox-full-night, a dataset of 62 hours of audio comprising 35402 flight calls of nocturnally migrating birds, as recorded from 6 sensors. We find a large performance gap between energy-based detection functions and data-driven machine listening. The best model is a deep convolutional neural network trained with data augmentation. We correlate recall with the density of flight calls over time and frequency and identify the main causes of false alarm.