Late Breaking / Demo papers

Friday, November 8, 14.30-17.00 h

LBD Poster Layout
LBD Demo Layout

[L-01]   Modeling Artist Preferences for Personalized Music Recommendations Dominik Kowald; Elisabeth Lex; Markus Schedl"The objective of our work is to provide a novel approach for modeling artist preferences of users with different music consumption patterns and listening habits. With that, we aim to realize music recommendations that are not biased towards the mainstream prevalent in a community."

[L-02]   Computation and Visualization of Differences between Two XML Music Score Files Francesco Foscarin; Florent Jacquemard; Raphael Fournier-S'niehotta"The goal of the paper is to provide a tool to visually compare the differences between two similar music scores, similarly to the Unix diff for text files."

[L-03]   Jamming with Yating: Interactive Demonstration of a Music Composition AI Wen-Yi Hsiao; Yin-Cheng Yeh; Yu-Siang Huang; Chung-Yang Wang; Jen-Yu Liu; Tsu-Kuang Hsieh; Hsiao-Tzu Hung; Jun-Yuan Wang; Yi-Hsuan Yang"We present a demo, which is made up with two AI functions: jamming and auto-accompaniment, as part of our endeavor in creating interactive music listening and creation experiences at the Taiwan AI Labs."

[L-04]   Reinforcement Learning Recommender System for Modelling Listening Sessions Tomáš Gajarský"A reinforcement learning recommender system for generating playlists, which continuously learns from users’ implicit feedback to maximize their satisfaction, was developed and compared to an established sequence-aware system and two traditional similarity-based methods."

[L-05]   Learning to generate Jazz and Pop Piano Music from Audio via MIR Techniques Yin-Cheng Yeh; Jen-Yu Liu; Wen-Yi Hsiao; Yu-Siang Huang; Yi-Hsuan Yang"We explore an approach that learns to compose music from musical audio recordings, by capitalizing state-of-the-art music information retrieval (MIR) techniques."

[L-06]   Tools for Semi-Automatic Bounding Box Annotation of Musical Measures in Sheet Music Frank Zalkow; Angel Villar Corrales; TJ Tsai; Vlora Arifi-Müller; Meinard Müller"In this contribution, we introduce various tools that are useful in the context of score following applications, where measures are highlighted synchronously to audio playback. In particular, we present tools for solving a subtask: the annotation of bounding boxes (given in pixels) of measure positions in digital scans of sheet music—a task that is extremely tedious when being done manually. "

[L-07]   Improving Music Tagging from Audio with User-Track Interactions Andres Ferraro; Jae Ho Jeon; Jisang Yoon; Xavier Serra; Dmitry Bogdanov"We propose to improve the tagging of music by using audio and collaborative filtering information (user-track interactions). We use Matrix Factorization (MF) to obtain a representation of the tracks from the user-track interactions and map those representations to the tags predicted from audio. The preliminary results show that following this approach we can increase the tagging performance."

[L-08]   Weak Multi-Label Audio-Tagging with Class Noise Katharina Prinz; Arthur Flexer"We evaluate whether deep neural networks are able to learn multiple annotations from music data with weak noisy labels, even in a domain-mismatch scenario. "

[L-09]   Automatic Music Tagging with Harmonic CNN Minz Won; Sanghyuk Chun; Oriol Nieto; Xavier Serra"We introduce the Harmonic Convolutional Neural Network (Harmonic CNN), a music representation model that exploits the inherent harmonic structure of audio signals."

[L-10]   Creating a Tool for Faciltiating and Researching Human Annotation of Musical Patterns Stephan Wells; Anja Volk; Iris Yuping Ren"It has been identified that there is a distinct lack of digital software for musical pattern annotation in the field of MIR. The goal of this paper is to present a novel digital software that allows users to intuitively and efficiently annotate musical patterns and evaluate this software through an extensive user study."

[L-11]   nnAudio: A PyTorch Audio Processing Tool Using 1D Convolution Neural Networks Kin Wai Cheuk; Kat Agres; Dorien Herremans"A GPU audio processing tool for waveform to spectrogram conversion"

[L-12]   Generative Audio Synthesis with a Parametric Model Krishna Subramani; Alexandre D'Hooge; Preeti Rao"Use a parametric representation of audio to train a generative model in the interest of obtaining more flexible control over the generated sound."

[L-13]   partitura: A Python Package for Handling Symbolic Musical Data Maarten Grachten; Carlos Eduardo Cancino-Chacón; Thassilo Gadermaier"Partitura is a Python package for handling symbolic musical information that is conveyed by modern staff notation. It provides a much wider range of possibilities to deal with music than the more reductive (but very common) pianoroll-oriented approach inspired by the MIDI standard."

[L-14]   ‘’ – A Tool for Identifying Small Musical Forms in Larger Music Corpora Beate Kutschke; Tobias Bachmann"Presents a digital program to identify the form of extensive corpora with small pieces (symbolic music). Based on the concept of the toy-block, the focus of the program is on the constitutive role of the repetition of note sequences, and the visualization of the form of individual pieces in direct comparison with each other."

[L-15]   Effects of Musical Stimulus Habituation and Musical Training on Felt Tension Courtney Reed; Elaine Chew"Preliminary results indicate that musical background and possibly habituation to repetition of musical phrases will cause differences in felt tension annotations over time. Given these results, the features are viewed against an existing tension model and impacts of time-varying emotion inclusion in predictive modelling are discussed."

[L-16]   MIREX 2019 Evaluation Results Stephen Downie; Yun Hao"To present MIREX 2019 overall evaluation results. The results are expected to be ready in October."

[L-17]   Human and Automated Judgements of Similarity in a Global Music Sample: A Preliminary Analysis Ding Shenghao; Hideo Daikoku; Ujwal Sriharsha Sanne; Marino Kinoshita; Rei Konno; Yoichi Kitayama; Shinya Fujii; Patrick E. Savage"This paper attempts to combine both human ground-truth ratings and automated algorithms to evaluate the reliability of human ratings and evaluate Panteli et al.'s automated algorithm against ground-truth data."

[L-18]   Evaluating Non-aligned Musical Score Transcriptions with MV2H Andrew McLeod; Kazuyoshi Yoshii"This paper presents an improvement to the existing MV2H metric for evaluating complete audio-to-score transcriptions, introducing a simple alignment method which allows it to be used for transcriptions whose notes are not explicitly aligned with the ground truth musical score."

[L-19]   Naturalistic Music EEG Dataset-Hindi (NMED-H) 2.0: New Release and Cross-Dataset Compatibility Blair Kaneshiro; Duc Nguyen; Jacek Dmochowski; Anthony Norcia; Jonathan Berger"We present a new (July 2019) release of an open EEG dataset, which was originally published in 2016 but has had low visibility compared to our other published EEG dataset. We have improved usability and cross-dataset compatibility in the new release to heighten its usefulness and relevance to MIR researchers interested in neuroscience."

[L-20]   CBF-periDB: A Chinese Bamboo Flute Dataset for Periodic Modulation Analysis Changhong Wang; Emmanouil Benetos; Elaine Chew"We present CBF-periDB, a dataset of Chinese bamboo flute performances, for ecologically valid analysis of periodic modulations in context."

[L-21]   Open Broadcast Media Audio from TV: A Dataset of TV Broadcast Audio with Relative Music Loudness Annotations Blai Meléndez-Catalán"This paper presents and describes a new open dataset for music detection and relative music loudness estimation."

[L-22]   An Automatic Music Generation Method Based on Given Patterns of Rhythm Yinji JING; Shengchen Li"This paper presents a method that generates music with given rhythm using LSTM-RNN"

[L-23]   Neural Content Based Collaborative Filtering For Recommendation Systems Prateek Verma; Jonathan Berger"The goal of this paper is to propose an algorithm to do recommendations combining content and collaborative filtering based methods, based upon a neural network-based architecture."

[L-24]   All-Convnet for Frame Level Music and Speech Segmentation Rhythm Rajiv Bhatia"An all-convolutional neural network (all-conv net) architecture designed for frame level music/speech classification with input as raw audio and processed using Kapre layers."

[L-25]   M-DJCUE: A Manually Annotated Dataset of Cue Points Mickael Zehren"Our research focuses on the development of algorithms for the identification of cue points; for testing purposes, we built a dataset to be used as a ground truth. To do so, we involved several DJs and musicians to annotate tracks with cue points."

[L-26]   Fractal Modeling of Carnatic Rhythm Sequence: Case-study on a Generative Model Kaustuv Kanti Ganguli; Akshay Anantapadmanabhan; Carlos Guedes"The aim of our current study is to analyze rhythmic patterns at different time-scales and search for possible fractal geometry in the rhythmic progression. This would, additionally, help understand the mental 'schema' a performer uses while performing familiar, yet not memorized, rhythm sequences."

[L-27]   ACMus - Advancing Computational Musicology: Semi-Supervised and Unsupervised Segmentation and Annotation of Musical Collections Estefania Cano; Antonio Escamilla; Sascha Grollmisch; Christian Kehling; Gustavo Adolfo López Gil; Fernando Mora Ángel; José Ricardo Zapata"This paper describes the goal and preliminary results of the ACMus research project. ACMus applies MIR techniques for musicological research on Colombian music from the Andes region."

[L-28]   Music Emotion Recognition for Indian Film Music Makarand Velankar; Parag Kulkarni"Automatic capturing music emotions from Indian film music using acoustic features. Different feature sets and machine learning classifiers used to test accuracy of the model. "

[L-29]   CARAT: Computer-aided Rhythmic Analysis Toolbox Martín Rocamora; Luis Jure; Magdalena Fuentes; Lucas S Maia; Luiz W. P. Biscainho"CARAT is a toolbox for computer-aided rhythm analysis from audio recordings that includes a set of ready-to-use tools in order to ease the adoption of computational tools by musicologists, while at the same time making readily available to MIR researchers some recently developed techniques for rhythm analysis, along with musical examples of their applicability. Pieces from three different corpora of music from the Afro-Atlantic tradition were taken as case studies for rhythmic pattern analysis. "

[L-30]   Midi Miner - A Python Library for Tonal Tension and Track Classification Rui Guo; Dorien Herremans; Thor Magnusson"This work provides a python library to calculate tonal tension, key and key changes in the MIDI file based on the spiral array concept. It also trains a melody, bass and harmony track classifier to extract those tracks for tasks such as music generation and transcription."

[L-31]   New Implementation Method for Generalized Frequency Modulation Synthesizer Keiji Hirata"We propose a generalized frequency modulation (GFM) synthesizer that employs the network architecture like conventional neural networks, every activation function of which is a vibrating function, and is optimized by the well-known backpropagation technique. We think that the tentative results of the experiments done so far suggest a possibility to create new sounds. "

[L-32]   Correlations Between Text Topics and Music Dimensions in Metal Music Using Latent Dirichlet Allocation and High-Level Audio Features Isabella Czedik-Eysenberg; Oliver Wieczorek; Christoph Reuter"By combining audio feature analysis and text processing (latent Dirichlet allocation), we (1) offer a comprehensive overview of the lyrical topics present within the metal genre and (2) are able to address whether or not levels of "hardness” and other music dimensions are associated with the occurrence of brutal (and other) textual topics."

[L-33]   Highly Expressive Peking Opera Synthesis with Durian System Yusong Wu; Shengchen Li; Chengzhu Yu; Heng Lu; Chao Weng; Dong Yu"In this paper, we present a prototype Jingju synthesis method which can generate expressive Jingju synthesis given the phoneme and note-pitch duration."

[L-34]   Android App for Recreating Old-recording Sound Effects for Voice Juan I. Garcia; Ana M Barbancho; Isabel Barbancho; Lorenzo J. Tardón"In this contribution, an Android App for recreating old-recording sound effects for voice is presented. The old-recording sound effects recreated are: Vinyl effect, Cylinder effect and Tape effect. Also, two nonlinear audio effects: Tube effect and Overdrive effect, are implemented since analog recording and reproduction are based on nonlinear signal processing. "

[L-35]   Automated Time-frequency Domain Audio Crossfades Using Graph Cuts Kyle Robinson; Dan Brown"We present the first implementation of a new method to automatically transition between songs by finding an optimal seam in the time-frequency spectrum. "

[L-36]   Spleeter: A Fast And State-of-the Art Music Source Separation Tool With Pre-trained Models Romain Hennequin; Anis KHLIF; Felix Voituret; Manuel Moussalam"We present and release Spleeter, a new source separation tool that performs state-of-the-art music source separation (to 2,4 or 5 stems). Spleeter can be run on both GPU or CPU and is able to perform 4-stems separation in more than 100x real-time on a single GPU."

[L-37]   Dig That Lick: Exploring Melodic Patterns in Jazz Improvisation Frank Höger; Klaus Frieler; Martin Pfleiderer; Simon Dixon"We present a web application for finding melodic patterns in jazz improvisations similar to a given query. Search results can be listened to and interactively explored via various visualizations."

[L-38]   musicnn: Pre-trained Convolutional Neural Networks for Music Audio Tagging Jordi Pons; Xavier Serra"Pronounced as "musician", the musicnn library contains a set of pre-trained musically motivated convolutional neural networks for music audio tagging: This repository also includes some pre-trained vgg-like baselines, and these models can be used as out-of-the box music audio taggers, as music feature extractors, or as pre-trained models for transfer learning."

[L-39]   GrooveCell - Music Learning for Bass and Drums Jakob Abeßer"The GrooveCell project aims to create a large database of grooves, i. e. bass lines and drum patterns, from various music styles. The paper presents the results of an initial user survey, which confirm the necessity of a music learning platform tailored to bass and drums."

[L-40]   Automatic Melody Composition Inspired by Short Melodies Using a Probabilistic Model and Harmonic Rules Lorenzo J. Tardón; Isabel Barbancho; Ana M Barbancho; Carles Roig; George Tzanetakis"This demo shows how automatic melody composition of melodies that follow the style of a certain single short melodic excerpt can be achieved in such a way that the sample excerpt can be considered an inspirational piece of music for the automatic compositor."

[L-41]   Multi-singer Singing Voice Synthesis System Juheon Lee; Hyeong-Seok Choi; Chang-Bin Jeon; Junghyun Koo; Kyogu Lee"We proposed a multi-singer singing voice synthesis system. We also suggested ways to model timbre and singing styles independently from the singing data of various singers."

[L-42]   MidiMe: Personalizing a MusicVAE Model with User Data Monica Dinculescu; Jesse Engel; Adam Roberts"MidiMe is an approach to quickly train a small personalized model to control a larger pretrained latent variable model, which allows us to generate samples from only the portions of the latent space we are interested in without having to retrain the large model from scratch. We demonstrate this in a web demo, using Magenta.js."

[L-43]   An Interactive Multimedia Companion to Wagner's 'Lohengrin': Encoding and Visualising a Motivic Study David Lewis; Kevin R Page; Laurence Dreyfus"We will demo an iPad app for exploring a motivic analysis of an opera, accompanying a musicological article and video. The app includes interactive visualisations of scores, libretto text, commentary and audio."

[L-44]   The Con Espressione! Exhibit: Exploring Human-Machine Collaboration in Expressive Performance Carlos Eduardo Cancino-Chacón; Stefan Balke; Florian Henkel; Christian Stussak; Gerhard Widmer"The Con Espressione! Exhibit is an interactive system designed for popular science exhibitions. It demonstrates and enables joint human--computer control of expressive performance."

[L-45]   An Opensource Web-based Pattern Annotation Framework - PAF Matevž Pesek; Darian Tomašević; Iris Ren; Matija Marolt"The Pattern Annotation Framework (PAF) tool collects the data about the annotator and the annotation process, to enable an analysis of relations between the user's experience/background and the annotations. The tool tracks the user's actions, such as the start and end time of an individual annotation and its changes, midi player actions and other. By open-sourcing the tool, we hope to aid other researchers in the MIR field dealing with pattern-related data gathering."

[L-46]   FlowSynth: Semantic and Vocal Synthesis Control Philippe Esling; Naotake Masuda; Adrien Bardet; Romeo Despres; Axel Chemla--Romeu-Santos"We propose a model able to learn semantic controls of a synthesizer, and infer its parameters based on vocal imitations. We also introduce a Max4Live and VST demo interface that implements all of these mechanisms."

[L-47]   BachDuet: A Human-Machine Duet Improvisation System Christodoulos Benetatos; Zhiyao Duan"BachDuet, is a system that enables real-time counterpoint improvisation between a human and a machine. We hope that it will serve as both an entertainment and practice tool for classical musicians to develop their improvisation skills"

[L-48]   Visual Pattern Analysis using Digital Sheet Music Matthias Miller; Hanna Schäfer; Alexandra Bonnici; Mennatallah El-Assady"We present an application that enables people interested in music analysis to analyze sheet music in MusicXML format using our interactive and intuitive visual pattern analysis prototype."

[L-49]   Linking and Visualising Performance Data and Semantic Music Encodings in Real-time David M. Weigl; Carlos Eduardo Cancino-Chacón; Martin Bonev; Werner Goebl"We present a visualisation interface for real-time performance-to-score alignment based on the Music Encoding and Linked Data (MELD) framework for semantic digital notation, employing an HMM-based polyphonic score-following system for symbolic (MIDI) piano performances. "

[L-50]   Melody Slot Machine Masatoshi Hamanaka"This paper describes melody manupilation using time-span tree of the generative theory of tonal music (GTTM). "