We introduce AudioLM, a framework for high-quality audio generation with long-term consistency. AudioLM maps the input audio to a sequence of discrete tokens and casts audio generation as a language modeling task in this representation space. We show how existing audio tokenizers provide different trade-offs between reconstruction quality and long-term structure, and we propose a hybrid tokenization scheme to achieve both objectives. Namely, we leverage the discretized activations of a masked language model pre-trained on audio to capture long-term structure and the discrete codes produced by a neural audio codec to achieve high-quality synthesis. By training on large corpora of raw audio waveforms, AudioLM learns to generate natural and coherent continuations given short prompts. When trained on speech, and without any transcript or annotation, AudioLM generates syntactically and semantically plausible speech continuations while also maintaining speaker identity and prosody for unseen speakers. Furthermore, we demonstrate how our approach extends beyond speech by generating coherent piano music continuations, despite being trained without any symbolic representation of music.

We introduce DIVE, an end-to-end speaker diarization sys-tem. DIVE presents the diarization task as an iterative pro-cess: it repeatedly builds a representation for each speaker before predicting their voice activity conditioned on the ex-tracted representations. This strategy intrinsically resolves the speaker ordering ambiguity without requiring the classi-cal permutation invariant training loss. In contrast with prior work, our model does not rely on pretrained speaker represen-tations and jointly optimizes all parameters of the system with a multi-speaker voice activity loss. DIVE does not require the training speaker identities and allows efficient window-based training. Importantly, our loss explicitly excludes unreliable speaker turn boundaries from training, which is adapted to the standard collar-based Diarization Error Rate (DER) eval-uation. Overall, these contributions yield a system redefining the state-of-the-art on the CALLHOME benchmark, with 6.7% DER compared to 7.8% for the best alternative.

Self-supervised pre-training using so-called “pretext” tasks has recently shown impressive performance across a wide range of modalities. In this work, we advance self-supervised learning from permutations, by pre-training a model to reorder shuffled parts of the spectrogram of an audio signal, to improve downstream classification performance. We make two main contributions. First, we overcome the main challenges of integrating permutation inversions into an end-to-end training scheme, using recent advances in differentiable ranking. This was heretofore sidestepped by casting the reordering task as classification, fundamentally reducing the space of permutations that can be exploited. Our experiments validate that learning from all possible permutations improves the quality of the pre-trained representations over using a limited, fixed set. Second, we show that inverting permutations is a meaningful pretext task for learning audio representations in an unsupervised fashion. In particular, we improve instrument classification and pitch estimation of musical notes by reordering spectrogram patches in the time-frequency space.

The sorting operation is one of the most commonly used building blocks in computer programming. In machine learning, it is often used for robust statistics. However, seen as a function, it is piecewise linear and as a result includes many kinks where it is non-differentiable. More problematic is the related ranking operator, often used for order statistics and ranking metrics. It is a piecewise constant function, meaning that its derivatives are null or undefined. While numerous works have proposed differentiable proxies to sorting and ranking, they do not achieve the $O(n \log n)$ time complexity one would expect from sorting and ranking operations. In this paper, we propose the first differentiable sorting and ranking operators with $O(n \log n)$ time and $O(n)$ space complexity. Our proposal in addition enjoys exact computation and differentiation. We achieve this feat by constructing differentiable operators as projections onto the permutahedron, the convex hull of permutations, and using a reduction to isotonic optimization. Empirically, we confirm that our approach is an order of magnitude faster than existing approaches and showcase two novel applications: differentiable Spearman’s rank correlation coefficient and least trimmed squares.

Low rank matrix factorization is a fundamental building block in machine learning, used for instance to summarize gene expression profile data or word-document counts. To be robust to outliers and differences in scale across features, a matrix factorization step is usually preceded by ad-hoc feature normalization steps, such as tf-idf scaling or data whitening. We propose in this work to learn these normalization operators jointly with the factorization itself. More precisely, given a $d\times n$ matrix $X$ of $d$ features measured on $n$ individuals, we propose to learn the parameters of quantile normalization operators that can operate row-wise on the values of $X$ and/or of its factorization $UV$ to improve the quality of the low-rank representation of $X$ itself. This optimization is facilitated by the introduction of a new differentiable quantile normalization operator built using optimal transport, providing new results on top of existing work by Cuturi et al. (2019). We demonstrate the applicability of these techniques on synthetic and genomics datasets.

Machine learning pipelines often rely on optimizers procedures to make discrete decisions (e.g., sorting, picking closest neighbors, or shortest paths). Although these discrete decisions are easily computed in a forward manner, they break the back-propagation of computational graphs. In order to expand the scope of learning problems that can be solved in an end-to-end fashion, we propose a systematic method to transform optimizers into operations that are differentiable and never locally constant. Our approach relies on stochastically perturbed optimizers, and can be used readily within existing solvers. Their derivatives can be evaluated efficiently, and smoothness tuned via the chosen noise amplitude. We also show how this framework can be connected to a family of losses developed in structured prediction, and give theoretical guarantees for their use in learning tasks. We demonstrate experimentally the performance of our approach on various tasks.

Sorting is used pervasively in machine learning, either to define elementary algorithms, such as $k$-nearest neighbors ($k$-NN) rules, or to define test-time metrics, such as top-$k$ classification accuracy or ranking losses. Sorting is however a poor match for the end-to-end, automatically differentiable pipelines of deep learning. Indeed, sorting procedures output two vectors, neither of which is differentiable: the vector of sorted values is piecewise linear, while the sorting permutation itself (or its inverse, the vector of ranks) has no differentiable properties to speak of, since it is integer-valued. We propose in this paper to replace the usual \texttt{sort} procedure with a differentiable proxy. Our proxy builds upon the fact that sorting can be seen as an optimal assignment problem, one in which the $n$ values to be sorted are matched to an \emph{auxiliary} probability measure supported on any \emph{increasing} family of $n$ target values. From this observation, we propose extended rank and sort operators by considering optimal transport (OT) problems (the natural relaxation for assignments) where the auxiliary measure can be any weighted measure supported on $m$ increasing values, where $m \ne n$. We recover differentiable operators by regularizing these OT problems with an entropic penalty, and solve them by applying Sinkhorn iterations. Using these smoothed rank and sort operators, we propose differentiable proxies for the classification 0/1 loss as well as for the quantile regression loss.

In this paper, we use shape grammars (SGs) for facade parsing, which amounts to segmenting 2D building facades into balconies, walls, windows, and doors in an architecturally meaningful manner. The main thrust of our work is the introduction of reinforcement learning (RL) techniques to deal with the computational complexity of the problem. RL provides us with techniques such as Q-learning and state aggregation which we exploit to efficiently solve facade parsing. We initially phrase the 1D parsing problem in terms of a Markov Decision Process, paving the way for the application of RL-based tools. We then develop novel techniques for the 2D shape parsing problem that take into account the specificities of the facade parsing problem. Specifically, we use state aggregation to enforce the symmetry of facade floors and demonstrate how to use RL to exploit bottom-up, image-based guidance during optimization. We provide systematic results on the Paris building dataset and obtain state-of-the-art results in a fraction of the time required by previous methods. We validate our method under diverse imaging conditions and make our software and results available online.

In this paper we address multi-view reconstruction of urban environments using 3D shape grammars. Our formulation expresses the solution to the problem as a shape grammar parse tree where both the tree and the corresponding derivation parameters are unknown. Besides the grammar constraint, the solution is guided by an image support that is twofold. First, we seek for a derivation that induces optimal semantic partitions in the different views. Second, using structure-from-motion, noisy depth maps can be determined towards minimizing their distance from to the ones predicted by any potential solution. We show how the underlying data structure can be efficiently optimized using evolutionary algorithms with automatic parameter selection. To the best of our knowledge, it is the first time that the multi-view 3D procedural modeling problem is tackled. Promising results demonstrate the potentials of the method towards producing a compact representation of urban environments.

We address shape grammar parsing for facade segmentation using Reinforcement Learning (RL). Shape parsing entails simultaneously optimizing the geometry and the topology (e.g. number of floors) of the facade, so as to optimize the fit of the predicted shape with the responses of pixel-level 'terminal detectors'. We formulate this problem in terms of a Hierarchical Markov Decision Process, by employing a recursive binary split grammar. This allows us to use RL to efficiently find the optimal parse of a given facade in terms of our shape grammar. Building on the RL paradigm, we exploit state aggregation to speedup computation, and introduce image-driven exploration in RL to accelerate convergence. We achieve state-of-the-art results on facade parsing, with a significant speed-up compared to existing methods, and substantial robustness to initial conditions. We demonstrate that the method can also be applied to interactive segmentation, and to a broad variety of architectural styles.

In this paper we propose a novel approach to the perceptual interpretation of building facades that combines shape grammars, supervised classification and random walks. Procedural modeling is used to model the geometric and the photometric variation of buildings. This is fused with visual classification techniques (randomized forests) that provide a crude probabilistic interpretation of the observation space in order to measure the appropriateness of a procedural generation with respect to the image. A random exploration of the grammar space is used to optimize the sequence of derivation rules towards a semantico-geometric interpretation of the observations. Experiments conducted on complex architecture facades with ground truth validate the approach.

In this paper we introduce a novel approach to single view reconstruction using shape grammars. Our approach consists in modeling architectural styles using a set of basic shapes and a set of parametric rules, corresponding to increasing levels of detail. This approach is able to model elaborate and varying architectural styles, using a tree representation of variable depth and complexity. Towards reconstruction, the parameters of the rules are optimized using image-based and architectural costs. This is done through an efficient MRF formulation based on the shape grammar itself. The resulting framework can produce precise 3D models from single views, can deal with lack of texture and the presence of occlusions and specular reflections, while maintaining the ability to cope with very complex architectural styles. Promising results demonstrate the potential of our approach.