arXiv subjects

Gerhard Widmer

Publications and source records attributed to Gerhard Widmer.

At least 109 records · Page 6Linked to original sources

arXiv2021

Over-Parameterization and Generalization in Audio Classification

Convolutional Neural Networks (CNNs) have been dominating classification tasks in various domains, such as machine vision, machine listening, and natural language processing. In machine listening, while generally exhibiting very good generalization capabilities, CNNs are sensitive to the specific audio recording device used, which has been recognized as a substantial problem in the acoustic scene classification (DCASE) community. In this study, we investigate the relationship between over-parameterization of acoustic scene classification models, and their resulting generalization abilities. Specifically, we test scaling CNNs in width and depth, under different conditions. Our results indicate that increasing width improves generalization to unseen devices, even without an increase in the number of parameters.

cs.SD↗

arXiv2021

Tracing Back Music Emotion Predictions to Sound Sources and Intuitive Perceptual Qualities

Music emotion recognition is an important task in MIR (Music Information Retrieval) research. Owing to factors like the subjective nature of the task and the variation of emotional cues between musical genres, there are still significant challenges in developing reliable and generalizable models. One important step towards better models would be to understand what a model is actually learning from the data and how the prediction for a particular input is made. In previous work, we have shown how to derive explanations of model predictions in terms of spectrogram image segments that connect to the high-level emotion prediction via a layer of easily interpretable perceptual features. However, that scheme lacks intuitive musical comprehensibility at the spectrogram level. In the present work, we bridge this gap by merging audioLIME -- a source-separation based explainer -- with mid-level perceptual features, thus forming an intuitive connection chain between the input audio and the output emotion predictions. We demonstrate the usefulness of this method by applying it to debug a biased emotion prediction model.

cs.SD↗

arXiv2021

Exploiting Temporal Dependencies for Cross-Modal Music Piece Identification

This paper addresses the problem of cross-modal musical piece identification and retrieval: finding the appropriate recording(s) from a database given a sheet music query, and vice versa, working directly with audio and scanned sheet music images. The fundamental approach to this is to learn a cross-modal embedding space with a suitable similarity structure for audio and sheet image snippets, using a deep neural network, and identifying candidate pieces by cross-modal near neighbour search in this space. However, this method is oblivious of temporal aspects of music. In this paper, we introduce two strategies that address this shortcoming. First, we present a strategy that aligns sequences of embeddings learned from sheet music scans and audio snippets. A series of experiments on whole piece and fragment-level retrieval on 24 hours worth of classical piano recordings demonstrates significant improvement. Second, we show that the retrieval can be further improved by introducing an attention mechanism to the embedding learning model that reduces the effects of tempo variations in music. To conclude, we assess the scalability of our method and discuss potential measures to make it suitable for truly large-scale applications.

eess.AS↗

arXiv2021

Receptive Field Regularization Techniques for Audio Classification and Tagging with Deep Convolutional Neural Networks

In this paper, we study the performance of variants of well-known Convolutional Neural Network (CNN) architectures on different audio tasks. We show that tuning the Receptive Field (RF) of CNNs is crucial to their generalization. An insufficient RF limits the CNN's ability to fit the training data. In contrast, CNNs with an excessive RF tend to over-fit the training data and fail to generalize to unseen testing data. As state-of-the-art CNN architectures-in computer vision and other domains-tend to go deeper in terms of number of layers, their RF size increases and therefore they degrade in performance in several audio classification and tagging tasks. We study well-known CNN architectures and how their building blocks affect their receptive field. We propose several systematic approaches to control the RF of CNNs and systematically test the resulting architectures on different audio classification and tagging tasks and datasets. The experiments show that regularizing the RF of CNNs using our proposed approaches can drastically improve the generalization of models, out-performing complex architectures and pre-trained models on larger datasets. The proposed CNNs achieve state-of-the-art results in multiple tasks, from acoustic scene classification to emotion and theme detection in music to instrument recognition, as demonstrated by top ranks in several pertinent challenges (DCASE, MediaEval).

cs.SD↗

arXiv2021

Handling Structural Mismatches in Real-time Opera Tracking

Algorithms for reliable real-time score following in live opera promise a lot of useful applications such as automatic subtitles display, or real-time video cutting in live streaming. Until now, such systems were based on the strong assumption that an opera performance follows the structure of the score linearly. However, this is rarely the case in practice, because of different opera versions and directors' cutting choices. In this paper, we propose a two-level solution to this problem. We introduce a real-time-capable, high-resolution (HR) tracker that can handle jumps or repetitions at specific locations provided to it. We then combine this with an additional low-resolution (LR) tracker that can handle all sorts of mismatches that can occur at any time, with some imprecision, and can re-direct the HR tracker if the latter is `lost' in the score. We show that the combination of the two improves tracking robustness in the presence of strong structural mismatches.

eess.AS↗

arXiv2021

Multi-modal Conditional Bounding Box Regression for Music Score Following

This paper addresses the problem of sheet-image-based on-line audio-to-score alignment also known as score following. Drawing inspiration from object detection, a conditional neural network architecture is proposed that directly predicts x,y coordinates of the matching positions in a complete score sheet image at each point in time for a given musical performance. Experiments are conducted on a synthetic polyphonic piano benchmark dataset and the new method is compared to several existing approaches from the literature for sheet-image-based score following as well as an Optical Music Recognition baseline. The proposed approach achieves new state-of-the-art results and furthermore significantly improves the alignment performance on a set of real-world piano recordings by applying Impulse Responses as a data augmentation technique.

cs.SD↗

arXiv2021

Towards Explaining Expressive Qualities in Piano Recordings: Transfer of Explanatory Features via Acoustic Domain Adaptation

Emotion and expressivity in music have been topics of considerable interest in the field of music information retrieval. In recent years, mid-level perceptual features have been suggested as means to explain computational predictions of musical emotion. We find that the diversity of musical styles and genres in the available dataset for learning these features is not sufficient for models to generalise well to specialised acoustic domains such as solo piano music. In this work, we show that by utilising unsupervised domain adaptation together with receptive-field regularised deep neural networks, it is possible to significantly improve generalisation to this domain. Additionally, we demonstrate that our domain-adapted models can better predict and explain expressive qualities in classical piano performances, as perceived and described by human listeners.

cs.SD↗

arXiv2020

Anomalous Sound Detection as a Simple Binary Classification Problem with Careful Selection of Proxy Outlier Examples

Unsupervised anomalous sound detection is concerned with identifying sounds that deviate from what is defined as 'normal', without explicitly specifying the types of anomalies. A significant obstacle is the diversity and rareness of outliers, which typically prevent us from collecting a representative set of anomalous sounds. As a consequence, most anomaly detection methods use unsupervised rather than supervised machine learning methods. Nevertheless, we will show that anomalous sound detection can be effectively framed as a supervised classification problem if the set of anomalous samples is carefully substituted with what we call proxy outliers. Candidates for proxy outliers are available in abundance as they potentially include all recordings that are neither normal nor abnormal sounds. We experiment with the machine condition monitoring data set of the 2020's DCASE Challenge and find proxy outliers with matching recording conditions and high similarity to the target sounds particularly informative. If no data with similar sounds and matching recording conditions is available, data sets with a larger diversity in these two dimensions are preferable. Our models based on supervised training with proxy outliers achieved rank three in Task 2 of the DCASE2020 Challenge.

eess.AS↗

arXiv2020

Low-Complexity Models for Acoustic Scene Classification Based on Receptive Field Regularization and Frequency Damping

Deep Neural Networks are known to be very demanding in terms of computing and memory requirements. Due to the ever increasing use of embedded systems and mobile devices with a limited resource budget, designing low-complexity models without sacrificing too much of their predictive performance gained great importance. In this work, we investigate and compare several well-known methods to reduce the number of parameters in neural networks. We further put these into the context of a recent study on the effect of the Receptive Field (RF) on a model's performance, and empirically show that we can achieve high-performing low-complexity models by applying specific restrictions on the RFs, in combination with parameter reduction methods. Additionally, we propose a filter-damping technique for regularizing the RF of models, without altering their architecture and changing their parameter counts. We will show that incorporating this technique improves the performance in various low-complexity settings such as pruning and decomposed convolution. Using our proposed filter damping, we achieved the 1st rank at the DCASE-2020 Challenge in the task of Low-Complexity Acoustic Scene Classification.

cs.LG↗

arXiv2020

Addressing the Recitative Problem in Real-time Opera Tracking

Robust real-time opera tracking (score following) would be extremely useful for many processes surrounding live opera staging and streaming, including automatic lyrics displays, camera control, or live video cutting. Recent work has shown that, with some appropriate measures to account for common problems such as breaks and interruptions, spontaneous applause, various noises and interludes, current audio-to-audio alignment algorithms can be made to follow an entire opera from beginning to end, in a relatively robust way. However, they remain inaccurate when the textual content becomes prominent against the melody or music -- notably, during recitativo passages. In this paper, we address this specific problem by proposing to use two specialized trackers in parallel, one focusing on music-, the other on speech-sensitive features. We first carry out a systematic study on speech-related features, targeting the precise alignment of corresponding recitatives from different performances of the same opera. Then we propose different solutions, based on pre-trained music and speech classifiers, to combine the two trackers in order to improve the global accuracy over the course of the entire opera.

eess.AS↗

arXiv2020

audioLIME: Listenable Explanations Using Source Separation

Deep neural networks (DNNs) are successfully applied in a wide variety of music information retrieval (MIR) tasks but their predictions are usually not interpretable. We propose audioLIME, a method based on Local Interpretable Model-agnostic Explanations (LIME) extended by a musical definition of locality. The perturbations used in LIME are created by switching on/off components extracted by source separation which makes our explanations listenable. We validate audioLIME on two different music tagging systems and show that it produces sensible explanations in situations where a competing method cannot.

cs.SD↗

arXiv2020

Towards Musically Meaningful Explanations Using Source Separation

Deep neural networks (DNNs) are successfully applied in a wide variety of music information retrieval (MIR) tasks. Such models are usually considered "black boxes", meaning that their predictions are not interpretable. Prior work on explainable models in MIR has generally used image processing tools to produce explanations for DNN predictions, but these are not necessarily musically meaningful, or can be listened to (which, arguably, is important in music). We propose audioLIME, a method based on Local Interpretable Model-agnostic Explanation (LIME), extended by a musical definition of locality. LIME learns locally linear models on perturbations of an example that we want to explain. Instead of extracting components of the spectrogram using image segmentation as part of the LIME pipeline, we propose using source separation. The perturbations are created by switching on/off sources which makes our explanations listenable. We first validate audioLIME on a classifier that was deliberately trained to confuse the true target with a spurious signal, and show that this can easily be detected using our method. We then show that it passes a sanity check that many available explanation methods fail. Finally, we demonstrate the general applicability of our (model-agnostic) method on a third-party music tagger.

cs.SD↗

arXiv2020

The Impact of Label Noise on a Music Tagger

We explore how much can be learned from noisy labels in audio music tagging. Our experiments show that carefully annotated labels result in highest figures of merit, but even high amounts of noisy labels contain enough information for successful learning. Artificial corruption of curated data allows us to quantize this contribution of noisy labels.

eess.AS↗

arXiv2020

On the Characterization of Expressive Performance in Classical Music: First Results of the Con Espressione Game

A piece of music can be expressively performed, or interpreted, in a variety of ways. With the help of an online questionnaire, the Con Espressione Game, we collected some 1,500 descriptions of expressive character relating to 45 performances of 9 excerpts from classical piano pieces, played by different famous pianists. More specifically, listeners were asked to describe, using freely chosen words (preferably: adjectives), how they perceive the expressive character of the different performances. In this paper, we offer a first account of this new data resource for expressive performance research, and provide an exploratory analysis, addressing three main questions: (1) how similarly do different listeners describe a performance of a piece? (2) what are the main dimensions (or axes) for expressive character emerging from this?; and (3) how do measurable parameters of a performance (e.g., tempo, dynamics) and mid- and high-level features that can be predicted by machine learning models (e.g., articulation, arousal) relate to these expressive dimensions? The dataset that we publish along with this paper was enriched by adding hand-corrected score-to-performance alignments, as well as descriptive audio features such as tempo and dynamics curves.

cs.SD↗

arXiv2020

Receptive-Field Regularized CNNs for Music Classification and Tagging

Convolutional Neural Networks (CNNs) have been successfully used in various Music Information Retrieval (MIR) tasks, both as end-to-end models and as feature extractors for more complex systems. However, the MIR field is still dominated by the classical VGG-based CNN architecture variants, often in combination with more complex modules such as attention, and/or techniques such as pre-training on large datasets. Deeper models such as ResNet -- which surpassed VGG by a large margin in other domains -- are rarely used in MIR. One of the main reasons for this, as we will show, is the lack of generalization of deeper CNNs in the music domain. In this paper, we present a principled way to make deep architectures like ResNet competitive for music-related tasks, based on well-designed regularization strategies. In particular, we analyze the recently introduced Receptive-Field Regularization and Shake-Shake, and show that they significantly improve the generalization of deep CNNs on music-related tasks, and that the resulting deep CNNs can outperform current more complex models such as CNNs augmented with pre-training and attention. We demonstrate this on two different MIR tasks and two corresponding datasets, thus offering our deep regularized CNNs as a new baseline for these datasets, which can also be used as a feature-extracting module in future, more complex approaches.

eess.AS↗

arXiv2020

Learning to Read and Follow Music in Complete Score Sheet Images

This paper addresses the task of score following in sheet music given as unprocessed images. While existing work either relies on OMR software to obtain a computer-readable score representation, or crucially relies on prepared sheet image excerpts, we propose the first system that directly performs score following in full-page, completely unprocessed sheet images. Based on incoming audio and a given image of the score, our system directly predicts the most likely position within the page that matches the audio, outperforming current state-of-the-art image-based score followers in terms of alignment precision. We also compare our method to an OMR-based approach and empirically show that it can be a viable alternative to such a system.

cs.LG↗

arXiv2020

On Data Augmentation and Adversarial Risk: An Empirical Analysis

Data augmentation techniques have become standard practice in deep learning, as it has been shown to greatly improve the generalisation abilities of models. These techniques rely on different ideas such as invariance-preserving transformations (e.g, expert-defined augmentation), statistical heuristics (e.g, Mixup), and learning the data distribution (e.g, GANs). However, in the adversarial settings it remains unclear under what conditions such data augmentation methods reduce or even worsen the misclassification risk. In this paper, we therefore analyse the effect of different data augmentation techniques on the adversarial risk by three measures: (a) the well-known risk under adversarial attacks, (b) a new measure of prediction-change stress based on the Laplacian operator, and (c) the influence of training examples on prediction. The results of our empirical analysis disprove the hypothesis that an improvement in the classification performance induced by a data augmentation is always accompanied by an improvement in the risk under adversarial attack. Further, our results reveal that the augmented data has more influence than the non-augmented data, on the resulting models. Taken together, our results suggest that general-purpose data augmentations that do not take into the account the characteristics of the data and the task, must be applied with care.

cs.LG↗

arXiv2020

Beneath (or beyond) the surface: Discovering voice-leading patterns with skip-grams

Recurrent voice-leading patterns like the Mi-Re-Do compound cadence (MRDCC) rarely appear on the musical surface in complex polyphonic textures, so finding these patterns using computational methods remains a tremendous challenge. The present study extends the canonical n-gram approach by using skip-grams, which include sub-sequences in an n-gram list if their constituent members occur within a certain number of skips. We compiled four data sets of Western tonal music consisting of symbolic encodings of the notated score and a recorded performance, created a model pipeline for defining, counting, filtering, and ranking skip-grams, and ranked the position of the MRDCC in every possible model configuration. We found that the MRDCC receives a higher rank in the list when the pipeline employs 5 skips, filters the list by excluding n-gram types that do not reflect a genuine harmonic change between adjacent members, and ranks the remaining types using a statistical association measure.

cs.SD↗