Filip Korzeniowski

Igor Pereira, Felipe Araújo, Filip Korzeniowski et al.

In this paper, we introduce the MoisesDB dataset for musical source separation. It consists of 240 tracks from 45 artists, covering twelve musical genres. For each song, we provide its individual audio sources, organized in a two-level hierarchical taxonomy of stems. This will facilitate building and evaluating fine-grained source separation systems that go beyond the limitation of using four stems (drums, bass, other, and vocals) due to lack of data. To facilitate the adoption of this dataset, we publish an easy-to-use Python library to download, process and use MoisesDB. Alongside a thorough documentation and analysis of the dataset contents, this work provides baseline results for open-source separation models for varying separation granularities (four, five, and six stems), and discuss their results.

Matthew C. McCallum, Filip Korzeniowski, Sergio Oramas et al.

In this work, we provide a broad comparative analysis of strategies for pre-training audio understanding models for several tasks in the music domain, including labelling of genre, era, origin, mood, instrumentation, key, pitch, vocal characteristics, tempo and sonority. Specifically, we explore how the domain of pre-training datasets (music or generic audio) and the pre-training methodology (supervised or unsupervised) affects the adequacy of the resulting audio embeddings for downstream tasks. We show that models trained via supervised learning on large-scale expert-annotated music datasets achieve state-of-the-art performance in a wide range of music labelling tasks, each with novel content and vocabularies. This can be done in an efficient manner with models containing less than 100 million parameters that require no fine-tuning or reparameterization for downstream tasks, making this approach practical for industry-scale audio catalogs. Within the class of unsupervised learning strategies, we show that the domain of the training dataset can significantly impact the performance of representations learned by the model. We find that restricting the domain of the pre-training dataset to music allows for training with smaller batch sizes while achieving state-of-the-art in unsupervised learning -- and in some cases, supervised learning -- for music understanding. We also corroborate that, while achieving state-of-the-art performance on many tasks, supervised learning can cause models to specialize to the supervised information provided, somewhat compromising a model's generality.

Matthias Dorfer, Jan Schlüter, Andreu Vall et al.

Cross-modality retrieval encompasses retrieval tasks where the fetched items are of a different type than the search query, e.g., retrieving pictures relevant to a given text query. The state-of-the-art approach to cross-modality retrieval relies on learning a joint embedding space of the two modalities, where items from either modality are retrieved using nearest-neighbor search. In this work, we introduce a neural network layer based on Canonical Correlation Analysis (CCA) that learns better embedding spaces by analytically computing projections that maximize correlation. In contrast to previous approaches, the CCA Layer (CCAL) allows us to combine existing objectives for embedding space learning, such as pairwise ranking losses, with the optimal projections of CCA. We show the effectiveness of our approach for cross-modality retrieval on three different scenarios (text-to-image, audio-sheet-music and zero-shot retrieval), surpassing both Deep CCA and a multi-view network using freely learned projections optimized by a pairwise ranking loss, especially when little training data is available (the code for all three methods is released at: https://github.com/CPJKU/cca_layer).

Sebastian Böck, Filip Korzeniowski, Jan Schlüter et al.

In this paper, we present madmom, an open-source audio processing and music information retrieval (MIR) library written in Python. madmom features a concise, NumPy-compatible, object oriented design with simple calling conventions and sensible default values for all parameters, which facilitates fast prototyping of MIR applications. Prototypes can be seamlessly converted into callable processing pipelines through madmom's concept of Processors, callable objects that run transparently on multiple cores. Processors can also be serialised, saved, and re-run to allow results to be easily reproduced anywhere. Apart from low-level audio processing, madmom puts emphasis on musically meaningful high-level features. Many of these incorporate machine learning techniques and madmom provides a module that implements some in MIR commonly used methods such as hidden Markov models and neural networks. Additionally, madmom comes with several state-of-the-art MIR algorithms for onset detection, beat, downbeat and meter tracking, tempo estimation, and piano transcription. These can easily be incorporated into bigger MIR systems or run as stand-alone programs.

Filip Korzeniowski, Sergio Oramas, Fabien Gouyon

Artist similarity plays an important role in organizing, understanding, and subsequently, facilitating discovery in large collections of music. In this paper, we present a hybrid approach to computing similarity between artists using graph neural networks trained with triplet loss. The novelty of using a graph neural network architecture is to combine the topology of a graph of artist connections with content features to embed artists into a vector space that encodes similarity. To evaluate the proposed method, we compile the new OLGA dataset, which contains artist similarities from AllMusic, together with content features from AcousticBrainz. With 17,673 artists, this is the largest academic artist similarity dataset that includes content-based features to date. Moreover, we also showcase the scalability of our approach by experimenting with a much larger proprietary dataset. Results show the superiority of the proposed approach over current state-of-the-art methods for music similarity. Finally, we hope that the OLGA dataset will facilitate research on data-driven models for artist similarity.

Filip Korzeniowski, Oriol Nieto, Matthew McCallum et al.

The mood of a song is a highly relevant feature for exploration and recommendation in large collections of music. These collections tend to require automatic methods for predicting such moods. In this work, we show that listening-based features outperform content-based ones when classifying moods: embeddings obtained through matrix factorization of listening data appear to be more informative of a track mood than embeddings based on its audio content. To demonstrate this, we compile a subset of the Million Song Dataset, totalling 67k tracks, with expert annotations of 188 different moods collected from AllMusic. Our results on this novel dataset not only expose the limitations of current audio-based models, but also aim to foster further reproducible research on this timely topic.

Filip Korzeniowski, Gerhard Widmer

Common temporal models for automatic chord recognition model chord changes on a frame-wise basis. Due to this fact, they are unable to capture musical knowledge about chord progressions. In this paper, we propose a temporal model that enables explicit modelling of chord changes and durations. We then apply N-gram models and a neural-network-based acoustic model within this framework, and evaluate the effect of model overconfidence. Our results show that model overconfidence plays only a minor role (but target smoothing still improves the acoustic model), and that stronger chord language models do improve recognition results, however their effects are small compared to other domains.

Filip Korzeniowski, Gerhard Widmer

We propose modifications to the model structure and training procedure to a recently introduced Convolutional Neural Network for musical key classification. These modifications enable the network to learn a genre-independent model that performs better than models trained for specific music styles, which has not been the case in existing work. We analyse this generalisation capability on three datasets comprising distinct genres. We then evaluate the model on a number of unseen data sets, and show its superior performance compared to the state of the art. Finally, we investigate the model's performance on short excerpts of audio. From these experiments, we conclude that models need to consider the harmonic coherence of the whole piece when classifying the local key of short segments of audio.

Filip Korzeniowski, Gerhard Widmer

Chord recognition systems typically comprise an acoustic model that predicts chords for each audio frame, and a temporal model that casts these predictions into labelled chord segments. However, temporal models have been shown to only smooth predictions, without being able to incorporate musical information about chord progressions. Recent research discovered that it might be the low hierarchical level such models have been applied to (directly on audio frames) which prevents learning musical relationships, even for expressive models such as recurrent neural networks (RNNs). However, if applied on the level of chord sequences, RNNs indeed can become powerful chord predictors. In this paper, we disentangle temporal models into a harmonic language model---to be applied on chord sequences---and a chord duration model that connects the chord-level predictions of the language model to the frame-level predictions of the acoustic model. In our experiments, we explore the impact of each model on the chord recognition score, and show that using harmonic language and duration models improves the results.

David R. W. Sears, Filip Korzeniowski, Gerhard Widmer

This study borrows and extends probabilistic language models from natural language processing to discover the syntactic properties of tonal harmony. Language models come in many shapes and sizes, but their central purpose is always the same: to predict the next event in a sequence of letters, words, notes, or chords. However, few studies employing such models have evaluated the most state-of-the-art architectures using a large-scale corpus of Western tonal music, instead preferring to use relatively small datasets containing chord annotations from contemporary genres like jazz, pop, and rock. Using symbolic representations of prominent instrumental genres from the common-practice period, this study applies a flexible, data-driven encoding scheme to (1) evaluate Finite Context (or n-gram) models and Recurrent Neural Networks (RNNs) in a chord prediction task; (2) compare predictive accuracy from the best-performing models for chord onsets from each of the selected datasets; and (3) explain differences between the two model architectures in a regression analysis. We find that Finite Context models using the Prediction by Partial Match (PPM) algorithm outperform RNNs, particularly for the piano datasets, with the regression model suggesting that RNNs struggle with particularly rare chord types.

Filip Korzeniowski, David R. W. Sears, Gerhard Widmer

We conduct a large-scale study of language models for chord prediction. Specifically, we compare N-gram models to various flavours of recurrent neural networks on a comprehensive dataset comprising all publicly available datasets of annotated chords known to us. This large amount of data allows us to systematically explore hyper-parameter settings for the recurrent neural networks---a crucial step in achieving good results with this model class. Our results show not only a quantitative difference between the models, but also a qualitative one: in contrast to static N-gram models, certain RNN configurations adapt to the songs at test time. This finding constitutes a further step towards the development of chord recognition systems that are more aware of local musical context than what was previously possible.

Filip Korzeniowski, Gerhard Widmer

We present an end-to-end system for musical key estimation, based on a convolutional neural network. The proposed system not only out-performs existing key estimation methods proposed in the academic literature; it is also capable of learning a unified model for diverse musical genres that performs comparably to existing systems specialised for specific genres. Our experiments confirm that different genres do differ in their interpretation of tonality, and thus a system tuned e.g. for pop music performs subpar on pieces of electronic music. They also reveal that such cross-genre setups evoke specific types of error (predicting the relative or parallel minor). However, using the data-driven approach proposed in this paper, we can train models that deal with multiple musical styles adequately, and without major losses in accuracy.

Filip Korzeniowski, Gerhard Widmer

Chord recognition systems use temporal models to post-process frame-wise chord preditions from acoustic models. Traditionally, first-order models such as Hidden Markov Models were used for this task, with recent works suggesting to apply Recurrent Neural Networks instead. Due to their ability to learn longer-term dependencies, these models are supposed to learn and to apply musical knowledge, instead of just smoothing the output of the acoustic model. In this paper, we argue that learning complex temporal models at the level of audio frames is futile on principle, and that non-Markovian models do not perform better than their first-order counterparts. We support our argument through three experiments on the McGill Billboard dataset. The first two show 1) that when learning complex temporal models at the frame level, improvements in chord sequence modelling are marginal; and 2) that these improvements do not translate when applied within a full chord recognition system. The third, still rather preliminary experiment gives first indications that the use of complex sequential models for chord prediction at higher temporal levels might be more promising.

Rainer Kelz, Matthias Dorfer, Filip Korzeniowski et al.

In an attempt at exploring the limitations of simple approaches to the task of piano transcription (as usually defined in MIR), we conduct an in-depth analysis of neural network-based framewise transcription. We systematically compare different popular input representations for transcription systems to determine the ones most suitable for use with neural networks. Exploiting recent advances in training techniques and new regularizers, and taking into account hyper-parameter tuning, we show that it is possible, by simple bottom-up frame-wise processing, to obtain a piano transcriber that outperforms the current published state of the art on the publicly available MAPS dataset -- without any complex post-processing steps. Thus, we propose this simple approach as a new baseline for this dataset, for future transcription research to build on and improve.

Filip Korzeniowski, Gerhard Widmer

Chord recognition systems depend on robust feature extraction pipelines. While these pipelines are traditionally hand-crafted, recent advances in end-to-end machine learning have begun to inspire researchers to explore data-driven methods for such tasks. In this paper, we present a chord recognition system that uses a fully convolutional deep auditory model for feature extraction. The extracted features are processed by a Conditional Random Field that decodes the final chord sequence. Both processing stages are trained automatically and do not require expert knowledge for optimising parameters. We show that the learned auditory system extracts musically interpretable features, and that the proposed chord recognition system achieves results on par or better than state-of-the-art algorithms.

Filip Korzeniowski, Gerhard Widmer

We explore frame-level audio feature learning for chord recognition using artificial neural networks. We present the argument that chroma vectors potentially hold enough information to model harmonic content of audio for chord recognition, but that standard chroma extractors compute too noisy features. This leads us to propose a learned chroma feature extractor based on artificial neural networks. It is trained to compute chroma features that encode harmonic information important for chord recognition, while being robust to irrelevant interferences. We achieve this by feeding the network an audio spectrum with context instead of a single frame as input. This way, the network can learn to selectively compensate noise and resolve harmonic ambiguities. We compare the resulting features to hand-crafted ones by using a simple linear frame-wise classifier for chord recognition on various data sets. The results show that the learned feature extractor produces superior chroma vectors for chord recognition.