arXiv subjects
Publications and source records attributed to Charalampos Saitis.
We discuss the discontinuities that arise when mapping unordered objects to neural network outputs of fixed permutation, referred to as the responsibility problem. Prior work has proved the existence of the issue by identifying a single discontinuity. Here, we show that discontinuities under such models are uncountably infinite, motivating further research into neural networks for unordered data.
Sinusoidal parameter estimation is a fundamental task in applications from spectral analysis to time-series forecasting. Estimating the sinusoidal frequency parameter by gradient descent is, however, often impossible as the error function is non-convex and densely populated with local minima. The growing family of differentiable signal processing methods has therefore been unable to tune the frequency of oscillatory components, preventing their use in a broad range of applications. This work presents a technique for joint sinusoidal frequency and amplitude estimation using the Wirtinger derivatives of a complex exponential surrogate and any first order gradient-based optimizer, enabling end to-end training of neural network controllers for unconstrained sinusoidal models.
Physical models of rigid bodies are used for sound synthesis in applications from virtual environments to music production. Traditional methods such as modal synthesis often rely on computationally expensive numerical solvers, while recent deep learning approaches are limited by post-processing of their results. In this work we present a novel end-to-end framework for training a deep neural network to generate modal resonators for a given 2D shape and material, using a bank of differentiable IIR filters. We demonstrate our method on a dataset of synthetic objects, but train our model using an audio-domain objective, paving the way for physically-informed synthesisers to be learned directly from recordings of real-world objects.
This study explores the association between music preferences and moral values by applying text analysis techniques to lyrics. Harvesting data from a Facebook-hosted application, we align psychometric scores of 1,386 users to lyrics from the top 5 songs of their preferred music artists as emerged from Facebook Page Likes. We extract a set of lyrical features related to each song's overarching narrative, moral valence, sentiment, and emotion. A machine learning framework was designed to exploit regression approaches and evaluate the predictive power of lyrical features for inferring moral values. Results suggest that lyrics from top songs of artists people like inform their morality. Virtues of hierarchy and tradition achieve higher prediction scores ($.20 \leq r \leq .30$) than values of empathy and equality ($.08 \leq r \leq .11$), while basic demographic variables only account for a small part in the models' explainability. This shows the importance of music listening behaviours, as assessed via lyrical preferences, alone in capturing moral values. We discuss the technological and musicological implications and possible future improvements.
Vocal Percussion Transcription (VPT) is concerned with the automatic detection and classification of vocal percussion sound events, allowing music creators and producers to sketch drum lines on the fly. Classifier algorithms in VPT systems learn best from small user-specific datasets, which usually restrict modelling to small input feature sets to avoid data overfitting. This study explores several deep supervised learning strategies to obtain informative feature sets for amateur vocal percussion classification. We evaluated the performance of these sets on regular vocal percussion classification tasks and compared them with several baseline approaches including feature selection methods and a speech recognition engine. These proposed learning models were supervised with several label sets containing information from four different levels of abstraction: instrument-level, syllable-level, phoneme-level, and boxeme-level. Results suggest that convolutional neural networks supervised with syllable-level annotations produced the most informative embeddings for classification, which can be used as input representations to fit classifiers with. Finally, we used back-propagation-based saliency maps to investigate the importance of different spectrogram regions for feature learning.
Imitating musical instruments with the human voice is an efficient way of communicating ideas between music producers, from sketching melody lines to clarifying desired sonorities. For this reason, there is an increasing interest in building applications that allow artists to efficiently pick target samples from big sound libraries just by imitating them vocally. In this study, we investigated the potential of conditional autoencoder models to learn informative features for Drum Sample Retrieval by Vocalisation (DSRV). We assessed the usefulness of their embeddings using four evaluation metrics, two of them relative to their acoustic properties and two of them relative to their perceptual properties via human listeners' similarity ratings. Results suggest that models conditioned on both sound-type labels (drum vs imitation) and drum-type labels (kick vs snare vs closed hi-hat vs opened hi-hat) learn the most informative embeddings for DSRV. We finally looked into individual differences in vocal imitation style via the Mantel test and found salient differences among participants, highlighting the importance of user information when designing DSRV systems.
The imitation of percussive sounds via the human voice is a natural and effective tool for communicating rhythmic ideas on the fly. Thus, the automatic retrieval of drum sounds using vocal percussion can help artists prototype drum patterns in a comfortable and quick way, smoothing the creative workflow as a result. Here we explore different strategies to perform this type of query, making use of both traditional machine learning algorithms and recent deep learning techniques. The main hyperparameters from the models involved are carefully selected by feeding performance metrics to a grid search algorithm. We also look into several audio data augmentation techniques, which can potentially regularise deep learning models and improve generalisation. We compare the final performances in terms of effectiveness (classification accuracy), efficiency (computational speed), stability (performance consistency), and interpretability (decision patterns), and discuss the relevance of these results when it comes to the design of successful query-by-vocal-percussion systems.
This research project investigates the application of deep learning to timbre transfer, where the timbre of a source audio can be converted to the timbre of a target audio with minimal loss in quality. The adopted approach combines Variational Autoencoders with Generative Adversarial Networks to construct meaningful representations of the source audio and produce realistic generations of the target audio and is applied to the Flickr 8k Audio dataset for transferring the vocal timbre between speakers and the URMP dataset for transferring the musical timbre between instruments. Furthermore, variations of the adopted approach are trained, and generalised performance is compared using the metrics SSIM (Structural Similarity Index) and FAD (Frechét Audio Distance). It was found that a many-to-many approach supersedes a one-to-one approach in terms of reconstructive capabilities, and that the adoption of a basic over a bottleneck residual block design is more suitable for enriching content information about a latent space. It was also found that the decision on whether cyclic loss takes on a variational autoencoder or vanilla autoencoder approach does not have a significant impact on reconstructive and adversarial translation aspects of the model.
We present the Neural Waveshaping Unit (NEWT): a novel, lightweight, fully causal approach to neural audio synthesis which operates directly in the waveform domain, with an accompanying optimisation (FastNEWT) for efficient CPU inference. The NEWT uses time-distributed multilayer perceptrons with periodic activations to implicitly learn nonlinear transfer functions that encode the characteristics of a target timbre. Once trained, a NEWT can produce complex timbral evolutions by simple affine transformations of its input and output signals. We paired the NEWT with a differentiable noise synthesiser and reverb and found it capable of generating realistic musical instrument performances with only 260k total model parameters, conditioned on F0 and loudness features. We compared our method to state-of-the-art benchmarks with a multi-stimulus listening test and the Fréchet Audio Distance and found it performed competitively across the tested timbral domains. Our method significantly outperformed the benchmarks in terms of generation speed, and achieved real-time performance on a consumer CPU, both with and without FastNEWT, suggesting it is a viable basis for future creative sound design tools.
Music is an essential component in our everyday lives and experiences, as it is a way that we use to express our feelings, emotions and cultures. In this study, we explore the association between music genre preferences, demographics and moral values by exploring self-reported data from an online survey administered in Canada. Participants filled in the moral foundations questionnaire, while they also provided their basic demographic information, and music preferences. Here, we predict the moral values of the participants inferring on their musical preferences employing classification and regression techniques. We also explored the predictive power of features estimated from factor analysis on the music genres, as well as the generalist/specialist (GS) score for revealing the diversity of musical choices for each user. Our results show the importance of music in predicting a person's moral values (.55-.69 AUROC); while knowledge of basic demographic features such as age and gender is enough to increase the performance (.58-.71 AUROC).
Convolutional Neural Networks have been extensively explored in the task of automatic music tagging. The problem can be approached by using either engineered time-frequency features or raw audio as input. Modulation filter bank representations that have been actively researched as a basis for timbre perception have the potential to facilitate the extraction of perceptually salient features. We explore end-to-end learned front-ends for audio representation learning, ModNet and SincModNet, that incorporate a temporal modulation processing block. The structure is effectively analogous to a modulation filter bank, where the FIR filter center frequencies are learned in a data-driven manner. The expectation is that a perceptually motivated filter bank can provide a useful representation for identifying music features. Our experimental results provide a fully visualisable and interpretable front-end temporal modulation decomposition of raw audio. We evaluate the performance of our model against the state-of-the-art of music tagging on the MagnaTagATune dataset. We analyse the impact on performance for particular tags when time-frequency bands are subsampled by the modulation filters at a progressively reduced rate. We demonstrate that modulation filtering provides promising results for music tagging and feature representation, without using extensive musical domain knowledge in the design of this front-end.
In this study, we produce a geometrically scaled perceptual timbre space from dissimilarity ratings of subtractive synthesized sounds and correlate the resulting dimensions with a set of acoustic descriptors. We curate a set of 15 sounds, produced by a synthesis model that uses varying source waveforms, frequency modulation (FM) and a lowpass filter with an enveloped cutoff frequency. Pairwise dissimilarity ratings were collected within an online browser-based experiment. We hypothesized that a varied waveform input source and enveloped filter would act as the main vehicles for timbral variation, providing novel acoustic correlates for the perception of synthesized timbres.
In this study, we aim to better understand the cognitive-emotional experience of visually impaired people when navigating in unfamiliar urban environments, both outdoor and indoor. We propose a multimodal framework based on random forest classifiers, which predict the actual environment among predefined generic classes of urban settings, inferring on real-time, non-invasive, ambulatory monitoring of brain and peripheral biosignals. Model performance reached 93% for the outdoor and 87% for the indoor environments (expressed in weighted AUROC), demonstrating the potential of the approach. Estimating the density distributions of the most predictive biomarkers, we present a series of geographic and temporal visualizations depicting the environmental contexts in which the most intense affective and cognitive reactions take place. A linear mixed model analysis revealed significant differences between categories of vision impairment, but not between normal and impaired vision. Despite the limited size of our cohort, these findings pave the way to emotionally intelligent mobility-enhancing systems, capable of implicit adaptation not only to changing environments but also to shifts in the affective state of the user in relation to different environmental and situational factors.