Stefan Lattner

Mathias Rose Bjare, Stefan Lattner, Gerhard Widmer

In recent years, the quality and public interest in music generation systems have grown, encouraging research into various ways to control these systems. We propose a novel method for controlling surprisal in music generation using sequence models. To achieve this goal, we define a metric called Instantaneous Information Content (IIC). The IIC serves as a proxy function for the perceived musical surprisal (as estimated from a probabilistic model) and can be calculated at any point within a music piece. This enables the comparison of surprisal across different musical content even if the musical events occur in irregular time intervals. We use beam search to generate musical material whose IIC curve closely approximates a given target IIC. We experimentally show that the IIC correlates with harmonic and rhythmic complexity and note density. The correlation decreases with the length of the musical context used for estimating the IIC. Finally, we conduct a qualitative user study to test if human listeners can identify the IIC curves that have been used as targets when generating the respective musical material. We provide code for creating IIC interpolations and IIC visualizations on https://github.com/muthissar/iic.

Mathias Rose Bjare, Stefan Lattner, Gerhard Widmer

Research in natural language processing has demonstrated that the quality of generations from trained autoregressive language models is significantly influenced by the used sampling strategy. In this study, we investigate the impact of different sampling techniques on musical qualities such as diversity and structure. To accomplish this, we train a high-capacity transformer model on a vast collection of highly-structured Irish folk melodies and analyze the musical qualities of the samples generated using distribution truncation sampling techniques. Specifically, we use nucleus sampling, the recently proposed "typical sampling", and conventional ancestral sampling. We evaluate the effect of these sampling strategies in two scenarios: optimal circumstances with a well-calibrated model and suboptimal circumstances where we systematically degrade the model's performance. We assess the generated samples using objective and subjective evaluations. We discover that probability truncation techniques may restrict diversity and structural patterns in optimal circumstances, but may also produce more musical samples in suboptimal circumstances.

Alain Riou, Stefan Lattner, Gaëtan Hadjeres et al.

This paper explores the automated process of determining stem compatibility by identifying audio recordings of single instruments that blend well with a given musical context. To tackle this challenge, we present Stem-JEPA, a novel Joint-Embedding Predictive Architecture (JEPA) trained on a multi-track dataset using a self-supervised learning approach. Our model comprises two networks: an encoder and a predictor, which are jointly trained to predict the embeddings of compatible stems from the embeddings of a given context, typically a mix of several instruments. Training a model in this manner allows its use in estimating stem compatibility - retrieving, aligning, or generating a stem to match a given mix - or for downstream tasks such as genre or key estimation, as the training paradigm requires the model to learn information related to timbre, harmony, and rhythm. We evaluate our model's performance on a retrieval task on the MUSDB18 dataset, testing its ability to find the missing stem from a mix and through a subjective user study. We also show that the learned embeddings capture temporal alignment information and, finally, evaluate the representations learned by our model on several downstream tasks, highlighting that they effectively capture meaningful musical features.

Stefan Lattner

Modern digital music production typically involves combining numerous acoustic elements to compile a piece of music. Important types of such elements are drum samples, which determine the characteristics of the percussive components of the piece. Artists must use their aesthetic judgement to assess whether a given drum sample fits the current musical context. However, selecting drum samples from a potentially large library is tedious and may interrupt the creative flow. In this work, we explore the automatic drum sample retrieval based on aesthetic principles learned from data. As a result, artists can rank the samples in their library by fit to some musical context at different stages of the production process (i.e., by fit to incomplete song mixtures). To this end, we use contrastive learning to maximize the score of drum samples originating from the same song as the mixture. We conduct a listening test to determine whether the human ratings match the automatic scoring function. We also perform objective quantitative analyses to evaluate the efficacy of our approach.

Mathias Rose Bjare, Stefan Lattner

It has been shown in a recent publication that words in human-produced English language tend to have an information content close to the conditional entropy. In this paper, we show that the same is true for events in human-produced monophonic musical sequences. We also show how "typical sampling" influences the distribution of information around the entropy for single events and sequences.

Bernardo Torres, Stefan Lattner, Gaël Richard

Significant strides have been made in creating voice identity representations using speech data. However, the same level of progress has not been achieved for singing voices. To bridge this gap, we suggest a framework for training singer identity encoders to extract representations suitable for various singing-related tasks, such as singing voice similarity and synthesis. We explore different self-supervised learning techniques on a large collection of isolated vocal tracks and apply data augmentations during training to ensure that the representations are invariant to pitch and content variations. We evaluate the quality of the resulting representations on singer similarity and identification tasks across multiple datasets, with a particular emphasis on out-of-domain generalization. Our proposed framework produces high-quality embeddings that outperform both speaker verification and wav2vec 2.0 pre-trained baselines on singing voice while operating at 44.1 kHz. We release our code and trained models to facilitate further research on singing voice and related areas.

Marco Pasini, Javier Nistal, Stefan Lattner et al.

Autoregressive models are typically applied to sequences of discrete tokens, but recent research indicates that generating sequences of continuous embeddings in an autoregressive manner is also feasible. However, such Continuous Autoregressive Models (CAMs) can suffer from a decline in generation quality over extended sequences due to error accumulation during inference. We introduce a novel method to address this issue by injecting random noise into the input embeddings during training. This procedure makes the model robust against varying error levels at inference. We further reduce error accumulation through an inference procedure that introduces low-level noise. Experiments on musical audio generation show that CAM substantially outperforms existing autoregressive and non-autoregressive approaches while preserving audio quality over extended sequences. This work paves the way for generating continuous embeddings in a purely autoregressive setting, opening new possibilities for real-time and interactive generative applications.

Alain Riou, Bernardo Torres, Ben Hayes et al.

In this paper, we introduce PESTO, a self-supervised learning approach for single-pitch estimation using a Siamese architecture. Our model processes individual frames of a Variable-$Q$ Transform (VQT) and predicts pitch distributions. The neural network is designed to be equivariant to translations, notably thanks to a Toeplitz fully-connected layer. In addition, we construct pitch-shifted pairs by translating and cropping the VQT frames and train our model with a novel class-based transposition-equivariant objective, eliminating the need for annotated data. Thanks to this architecture and training objective, our model achieves remarkable performances while being very lightweight ($130$k parameters). Evaluations on music and speech datasets (MIR-1K, MDB-stem-synth, and PTDB) demonstrate that PESTO not only outperforms self-supervised baselines but also competes with supervised methods, exhibiting superior cross-dataset generalization. Finally, we enhance PESTO's practical utility by developing a streamable VQT implementation using cached convolutions. Combined with our model's low latency (less than 10 ms) and minimal parameter count, this makes PESTO particularly suitable for real-time applications.

Alain Riou, Stefan Lattner, Gaëtan Hadjeres et al.

This paper addresses the problem of self-supervised general-purpose audio representation learning. We explore the use of Joint-Embedding Predictive Architectures (JEPA) for this task, which consists of splitting an input mel-spectrogram into two parts (context and target), computing neural representations for each, and training the neural network to predict the target representations from the context representations. We investigate several design choices within this framework and study their influence through extensive experiments by evaluating our models on various audio classification benchmarks, including environmental sounds, speech and music downstream tasks. We focus notably on which part of the input data is used as context or target and show experimentally that it significantly impacts the model's quality. In particular, we notice that some effective design choices in the image domain lead to poor performance on audio, thus highlighting major differences between these two modalities.

Alain Riou, Antonin Gagneré, Gaëtan Hadjeres et al.

In this paper, we tackle the task of musical stem retrieval. Given a musical mix, it consists in retrieving a stem that would fit with it, i.e., that would sound pleasant if played together. To do so, we introduce a new method based on Joint-Embedding Predictive Architectures, where an encoder and a predictor are jointly trained to produce latent representations of a context and predict latent representations of a target. In particular, we design our predictor to be conditioned on arbitrary instruments, enabling our model to perform zero-shot stem retrieval. In addition, we discover that pretraining the encoder using contrastive learning drastically improves the model's performance. We validate the retrieval performances of our model using the MUSDB18 and MoisesDB datasets. We show that it significantly outperforms previous baselines on both datasets, showcasing its ability to support more or less precise (and possibly unseen) conditioning. We also evaluate the learned embeddings on a beat tracking task, demonstrating that they retain temporal structure and local information.

Javier Nistal, Marco Pasini, Cyran Aouameur et al.

Recent advancements in deep generative models present new opportunities for music production but also pose challenges, such as high computational demands and limited audio quality. Moreover, current systems frequently rely solely on text input and typically focus on producing complete musical pieces, which is incompatible with existing workflows in music production. To address these issues, we introduce "Diff-A-Riff," a Latent Diffusion Model designed to generate high-quality instrumental accompaniments adaptable to any musical context. This model offers control through either audio references, text prompts, or both, and produces 48kHz pseudo-stereo audio while significantly reducing inference time and memory usage. We demonstrate the model's capabilities through objective metrics and subjective listening tests, with extensive examples available on the accompanying website: sonycslparis.github.io/diffariff-companion/

Javier Nistal, Cyran Aouameur, Stefan Lattner et al.

Influenced by the field of Computer Vision, Generative Adversarial Networks (GANs) are often adopted for the audio domain using fixed-size two-dimensional spectrogram representations as the "image data". However, in the (musical) audio domain, it is often desired to generate output of variable duration. This paper presents VQCPC-GAN, an adversarial framework for synthesizing variable-length audio by exploiting Vector-Quantized Contrastive Predictive Coding (VQCPC). A sequence of VQCPC tokens extracted from real audio data serves as conditional input to a GAN architecture, providing step-wise time-dependent features of the generated content. The input noise z (characteristic in adversarial architectures) remains fixed over time, ensuring temporal consistency of global features. We evaluate the proposed model by comparing a diverse set of metrics against various strong baselines. Results show that, even though the baselines score best, VQCPC-GAN achieves comparable performance even when generating variable-length audio. Numerous sound examples are provided in the accompanying website, and we release the code for reproducibility.

Javier Nistal, Stefan Lattner, Gaël Richard

In this paper, we compare different audio signal representations, including the raw audio waveform and a variety of time-frequency representations, for the task of audio synthesis with Generative Adversarial Networks (GANs). We conduct the experiments on a subset of the NSynth dataset. The architecture follows the benchmark Progressive Growing Wasserstein GAN. We perform experiments both in a fully non-conditional manner as well as conditioning the network on the pitch information. We quantitatively evaluate the generated material utilizing standard metrics for assessing generative models, and compare training and sampling times. We show that complex-valued as well as the magnitude and Instantaneous Frequency of the Short-Time Fourier Transform achieve the best results, and yield fast generation and inversion times. The code for feature extraction, training and evaluating the model is available online.

Stefan Lattner

In recent years, artificial neural networks (ANNs) have become a universal tool for tackling real-world problems. ANNs have also shown great success in music-related tasks including music summarization and classification, similarity estimation, computer-aided or autonomous composition, and automatic music analysis. As structure is a fundamental characteristic of Western music, it plays a role in all these tasks. Some structural aspects are particularly challenging to learn with current ANN architectures. This is especially true for mid- and high-level self-similarity, tonal and rhythmic relationships. In this thesis, I explore the application of ANNs to different aspects of musical structure modeling, identify some challenges involved and propose strategies to address them. First, using probability estimations of a Restricted Boltzmann Machine (RBM), a probabilistic bottom-up approach to melody segmentation is studied. Then, a top-down method for imposing a high-level structural template in music generation is presented, which combines Gibbs sampling using a convolutional RBM with gradient-descent optimization on the intermediate solutions. Furthermore, I motivate the relevance of musical transformations in structure modeling and show how a connectionist model, the Gated Autoencoder (GAE), can be employed to learn transformations between musical fragments. For learning transformations in sequences, I propose a special predictive training of the GAE, which yields a representation of polyphonic music as a sequence of intervals. Furthermore, the applicability of these interval representations to a top-down discovery of repeated musical sections is shown. Finally, a recurrent variant of the GAE is proposed, and its efficacy in music prediction and modeling of low-level repetition structure is demonstrated.

Stefan Lattner, Maarten Grachten

Spurred by the potential of deep learning, computational music generation has gained renewed academic interest. A crucial issue in music generation is that of user control, especially in scenarios where the music generation process is conditioned on existing musical material. Here we propose a model for conditional kick drum track generation that takes existing musical material as input, in addition to a low-dimensional code that encodes the desired relation between the existing material and the new material to be generated. These relational codes are learned in an unsupervised manner from a music dataset. We show that codes can be sampled to create a variety of musically plausible kick drum tracks and that the model can be used to transfer kick drum patterns from one song to another. Lastly, we demonstrate that the learned codes are largely invariant to tempo and time-shift.

Stefan Lattner, Monika Dörfler, Andreas Arzt

Learning features from data has shown to be more successful than using hand-crafted features for many machine learning tasks. In music information retrieval (MIR), features learned from windowed spectrograms are highly variant to transformations like transposition or time-shift. Such variances are undesirable when they are irrelevant for the respective MIR task. We propose an architecture called Complex Autoencoder (CAE) which learns features invariant to orthogonal transformations. Mapping signals onto complex basis functions learned by the CAE results in a transformation-invariant "magnitude space" and a transformation-variant "phase space". The phase space is useful to infer transformations between data pairs. When exploiting the invariance-property of the magnitude space, we achieve state-of-the-art results in audio-to-score alignment and repeated section discovery for audio. A PyTorch implementation of the CAE, including the repeated section discovery method, is available online.

Andreas Arzt, Stefan Lattner

Audio-to-score alignment is an important pre-processing step for in-depth analysis of classical music. In this paper, we apply novel transposition-invariant audio features to this task. These low-dimensional features represent local pitch intervals and are learned in an unsupervised fashion by a gated autoencoder. Our results show that the proposed features are indeed fully transposition-invariant and enable accurate alignments between transposed scores and performances. Furthermore, they can even outperform widely used features for audio-to-score alignment on `untransposed data', and thus are a viable and more flexible alternative to well-established features for music alignment and matching.

Stefan Lattner, Maarten Grachten, Gerhard Widmer

Connectionist sequence models (e.g., RNNs) applied to musical sequences suffer from two known problems: First, they have strictly "absolute pitch perception". Therefore, they fail to generalize over musical concepts which are commonly perceived in terms of relative distances between pitches (e.g., melodies, scale types, modes, cadences, or chord types). Second, they fall short of capturing the concepts of repetition and musical form. In this paper we introduce the recurrent gated autoencoder (RGAE), a recurrent neural network which learns and operates on interval representations of musical sequences. The relative pitch modeling increases generalization and reduces sparsity in the input data. Furthermore, it can learn sequences of copy-and-shift operations (i.e. chromatically transposed copies of musical fragments)---a promising capability for learning musical repetition structure. We show that the RGAE improves the state of the art for general connectionist sequence models in learning to predict monophonic melodies, and that ensembles of relative and absolute music processing models improve the results appreciably. Furthermore, we show that the relative pitch processing of the RGAE naturally facilitates the learning and the generation of sequences of copy-and-shift operations, wherefore the RGAE greatly outperforms a common absolute pitch recurrent neural network on this task.

Stefan Lattner, Maarten Grachten, Gerhard Widmer

Many music theoretical constructs (such as scale types, modes, cadences, and chord types) are defined in terms of pitch intervals---relative distances between pitches. Therefore, when computer models are employed in music tasks, it can be useful to operate on interval representations rather than on the raw musical surface. Moreover, interval representations are transposition-invariant, valuable for tasks like audio alignment, cover song detection and music structure analysis. We employ a gated autoencoder to learn fixed-length, invertible and transposition-invariant interval representations from polyphonic music in the symbolic domain and in audio. An unsupervised training method is proposed yielding an organization of intervals in the representation space which is musically plausible. Based on the representations, a transposition-invariant self-similarity matrix is constructed and used to determine repeated sections in symbolic music and in audio, yielding competitive results in the MIREX task "Discovery of Repeated Themes and Sections".

Stefan Lattner, Maarten Grachten, Gerhard Widmer

Music is usually highly structured and it is still an open question how to design models which can successfully learn to recognize and represent musical structure. A fundamental problem is that structurally related patterns can have very distinct appearances, because the structural relationships are often based on transformations of musical material, like chromatic or diatonic transposition, inversion, retrograde, or rhythm change. In this preliminary work, we study the potential of two unsupervised learning techniques - Restricted Boltzmann Machines (RBMs) and Gated Autoencoders (GAEs) - to capture pre-defined transformations from constructed data pairs. We evaluate the models by using the learned representations as inputs in a discriminative task where for a given type of transformation (e.g. diatonic transposition), the specific relation between two musical patterns must be recognized (e.g. an upward transposition of diatonic steps). Furthermore, we measure the reconstruction error of models when reconstructing musical transformed patterns. Lastly, we test the models in an analogy-making task. We find that it is difficult to learn musical transformations with the RBM and that the GAE is much more adequate for this task, since it is able to learn representations of specific transformations that are largely content-invariant. We believe these results show that models such as GAEs may provide the basis for more encompassing music analysis systems, by endowing them with a better understanding of the structures underlying music.

Stefan Lattner, Maarten Grachten

Content-invariance in mapping codes learned by GAEs is a useful feature for various relation learning tasks. In this paper we show that the content-invariance of mapping codes for images of 2D and 3D rotated objects can be substantially improved by extending the standard GAE loss (symmetric reconstruction error) with a regularization term that penalizes the symmetric cross-reconstruction error. This error term involves reconstruction of pairs with mapping codes obtained from other pairs exhibiting similar transformations. Although this would principally require knowledge of the transformations exhibited by training pairs, our experiments show that a bootstrapping approach can sidestep this issue, and that the regularization term can effectively be used in an unsupervised setting.

Stefan Lattner, Maarten Grachten, Gerhard Widmer

We introduce a method for imposing higher-level structure on generated, polyphonic music. A Convolutional Restricted Boltzmann Machine (C-RBM) as a generative model is combined with gradient descent constraint optimisation to provide further control over the generation process. Among other things, this allows for the use of a "template" piece, from which some structural properties can be extracted, and transferred as constraints to the newly generated material. The sampling process is guided with Simulated Annealing to avoid local optima, and to find solutions that both satisfy the constraints, and are relatively stable with respect to the C-RBM. Results show that with this approach it is possible to control the higher-level self-similarity structure, the meter, and the tonal properties of the resulting musical piece, while preserving its local musical coherence.