Quantifying audio aesthetics is challenging due to its subjective nature, influenced by human perception and cultural context. Traditional methods rely on human listeners, leading to inconsistencies and high resource demands. This paper addresses the growing need for automated systems capable of predicting audio aesthetics without human intervention. Such systems are crucial for applications like data filtering, pseudo-labeling, and evaluating generative models.In this paper, we propose new annotation guidelines that break down human listening perspectives into four axes and develop no-reference, peritem prediction models for more nuanced audio quality assessment. Our models are evaluated against human mean opinion scores (MOS) and existing methods, demonstrating comparable or superior performance. This research not only advances the field of audio aesthetics but also provides open-source models and datasets to facilitate future work and benchmarking.

We propose FlowDec, a neural full-band audio codec for general audio sampled at 48 kHz that combines non-adversarial codec training with a stochastic postfilter based on a novel conditional flow matching method. Compared to the prior work ScoreDec which is based on score matching, we generalize from speech to general audio and move from 24 kbit/s to as low as 4 kbit/s, while improving output quality and reducing the required postfilter DNN evaluations from 60 to 6 without any fine-tuning or distillation techniques. We provide theoretical insights and geometric intuitions for our approach in comparison to ScoreDec as well as another recent work that uses flow matching, and conduct ablation studies on our proposed components. We show that FlowDec is a competitive alternative to the recent GAN-dominated stream of neural codecs, achieving FAD scores better than those of the established GAN-based codec DAC and listening test scores that are on par, and producing qualitatively more natural reconstructions for speech and harmonic structures in music.

Generative models have gained more and more attention in recent years for their remarkable success in tasks that required estimating and sampling data distribution to generate high-fidelity synthetic data. In speech, text-to-speech synthesis and neural vocoder are good examples where generative models have shined. While generative models have been applied to different applications in speech, there exists no general-purpose generative model that models speech directly. In this work, we take a step toward this direction by showing a single pre-trained generative model can be adapted to different downstream tasks with strong performance. Specifically, we pre-trained a generative model, named SpeechFlow, on 60k hours of untranscribed speech with Flow Matching and masked conditions. Experiment results show the pre-trained generative model can be fine-tuned with task-specific data to match or surpass existing expert models on speech enhancement, separation, and synthesis. Our work suggested a foundational model for generation tasks in speech can be built with generative pre-training.

We introduce MusicFlow, a cascaded text-to-music generation model based on flow matching. Based on self-supervised representations to bridge between text descriptions and music audios, we construct two flow matching networks to model the conditional distribution of semantic and acoustic features. Additionally, we leverage masked prediction as the training objective, enabling the model to generalize to other tasks such as music infilling and continuation in a zero-shot manner. Experiments on MusicCaps reveal that the music generated by MusicFlow exhibits superior quality and text coherence despite being over $2\sim5$ times smaller and requiring $5$ times fewer iterative steps. Simultaneously, the model can perform other music generation tasks and achieves competitive performance in music infilling and continuation.

This paper introduces Bespoke Non-Stationary (BNS) Solvers, a solver distillation approach to improve sample efficiency of Diffusion and Flow models. BNS solvers are based on a family of non-stationary solvers that provably subsumes existing numerical ODE solvers and consequently demonstrate considerable improvement in sample approximation (PSNR) over these baselines. Compared to model distillation, BNS solvers benefit from a tiny parameter space ($<$200 parameters), fast optimization (two orders of magnitude faster), maintain diversity of samples, and in contrast to previous solver distillation approaches nearly close the gap from standard distillation methods such as Progressive Distillation in the low-medium NFE regime. For example, BNS solver achieves 45 PSNR / 1.76 FID using 16 NFE in class-conditional ImageNet-64. We experimented with BNS solvers for conditional image generation, text-to-image generation, and text-2-audio generation showing significant improvement in sample approximation (PSNR) in all.

Recent successful generative models are trained by fitting a neural network to an a-priori defined tractable probability density path taking noise to training examples. In this paper we investigate the space of Gaussian probability paths, which includes diffusion paths as an instance, and look for an optimal member in some useful sense. In particular, minimizing the Kinetic Energy (KE) of a path is known to make particles' trajectories simple, hence easier to sample, and empirically improve performance in terms of likelihood of unseen data and sample generation quality. We investigate Kinetic Optimal (KO) Gaussian paths and offer the following observations: (i) We show the KE takes a simplified form on the space of Gaussian paths, where the data is incorporated only through a single, one dimensional scalar function, called the *data separation function*. (ii) We characterize the KO solutions with a one dimensional ODE. (iii) We approximate data-dependent KO paths by approximating the data separation function and minimizing the KE. (iv) We prove that the data separation function converges to $1$ in the general case of arbitrary normalized dataset consisting of $n$ samples in $d$ dimension as $n/\sqrt{d}\rightarrow 0$. A consequence of this result is that the Conditional Optimal Transport (Cond-OT) path becomes *kinetic optimal* as $n/\sqrt{d}\rightarrow 0$. We further support this theory with empirical experiments on ImageNet.

Large-scale generative models such as GPT and DALL-E have revolutionized the research community. These models not only generate high fidelity outputs, but are also generalists which can solve tasks not explicitly taught. In contrast, speech generative models are still primitive in terms of scale and task generalization. In this paper, we present Voicebox, the most versatile text-guided generative model for speech at scale. Voicebox is a non-autoregressive flow-matching model trained to infill speech, given audio context and text, trained on over 50K hours of speech that are not filtered or enhanced. Similar to GPT, Voicebox can perform many different tasks through in-context learning, but is more flexible as it can also condition on future context. Voicebox can be used for mono or cross-lingual zero-shot text-to-speech synthesis, noise removal, content editing, style conversion, and diverse sample generation. In particular, Voicebox outperforms the state-of-the-art zero-shot TTS model VALL-E on both intelligibility (5.9\% vs 1.9\% word error rates) and audio similarity (0.580 vs 0.681) while being up to 20 times faster. Audio samples can be found in \url{https://voicebox.metademolab.com}.

Jiajun Shen, Peng-Jen Chen, Matthew Le, Junxian He, Jiatao Gu, Myle Ott, Michael Auli, Marc’Aurelio Ranzato. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021.

Multivariate Hawkes Processes (MHPs) are an important class of temporal point processes that have enabled key advances in understanding and predicting social information systems. However, due to their complex modeling of temporal dependencies, MHPs have proven to be notoriously difficult to scale, what has limited their applications to relatively small domains. In this work, we propose a novel model and computational approach to overcome this important limitation. By exploiting a characteristic sparsity pattern in real-world diffusion processes, we show that our approach allows to compute the exact likelihood and gradients of an MHP – independently of the ambient dimensions of the underlying network. We show on synthetic and real-world datasets that our method does not only achieve state-of-the-art modeling results, but also improves runtime performance by multiple orders of magnitude on sparse event sequences. In combination with easily interpretable latent variables and influence structures, this allows us to analyze diffusion processes in networks at previously unattainable scale.

We consider the task of inferring “is-a” relationships from large text corpora. For this purpose, we propose a new method combining hyperbolic embeddings and Hearst patterns. This approach allows us to set appropriate constraints for inferring concept hierarchies from distributional contexts while also being able to predict missing “is-a”-relationships and to correct wrong extractions. Moreover – and in contrast with other methods – the hierarchical nature of hyperbolic space allows us to learn highly efficient representations and to improve the taxonomic consistency of the inferred hierarchies. Experimentally, we show that our approach achieves state-of-the-art performance on several commonly-used benchmarks.

Matthew Le, Y-Lan Boureau, Maximilian Nickel. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

OBJECTIVE: Community health systems operating in remote areas require accurate information about where people live to efficiently provide services across large regions. We sought to determine whether a machine learning analyses of satellite imagery can be used to map remote communities to facilitate service delivery and planning. MATERIALS AND METHODS: We developed a method for mapping communities using a deep learning approach that excels at detecting objects within images. We trained an algorithm to detect individual buildings, then examined building clusters to identify groupings suggestive of communities. The approach was validated in southeastern Liberia, by comparing algorithmically generated results with community location data collected manually by enumerators and community health workers. RESULTS: The deep learning approach achieved 86.47% positive predictive value and 79.49% sensitivity with respect to individual building detection. The approach identified 75.67% (n = 451) of communities registered through the community enumeration process, and identified an additional 167 potential communities not previously registered. Several instances of false positives and false negatives were identified. DISCUSSION: Analysis of satellite images is a promising solution for mapping remote communities rapidly, and with relatively low costs. Further research is needed to determine whether the communities identified algorithmically, but not registered in the manual enumeration process, are currently inhabited. CONCLUSIONS: To our knowledge, this study represents the first effort to apply image recognition algorithms to rural healthcare delivery. Results suggest that these methods have the potential to enhance community health worker scale-up efforts in underserved remote communities.

Software Transactional Memory (STM) has become very popular in Haskell. Currently, there are nearly 500 packages on Haskell’s package archive that directly use STM. Despite the widespread use in real world applications, Haskell’s STM implementation has seen very few updates since its introduction in 2005.

Software transactional memory (STM) has proven to be a useful abstraction for developing concurrent applications, where programmers denote transactions with an atomic construct that delimits a collection of reads and writes to shared mutable references. The runtime system then guarantees that all transactions are observed to execute atomically with respect to each other. Traditionally, when the runtime system detects that one transaction conflicts with another, it aborts one of the transactions and restarts its execution from the beginning. This can lead to problems with both execution time and throughput. In this paper, we present a novel approach that uses first-class continuations to restart a conflicting transaction at the point of a conflict, avoiding the re-execution of any work from the beginning of the transaction that has not been compromised. In practice, this allows transactions to complete more quickly, decreasing execution time and increasing throughput. We have implemented this idea in the context of the Manticore project, an ML-family language with support for parallelism and concurrency. Crucially, we rely on constant-time continuation capturing via a continuation-passing-style (CPS) transformation and heap-allocated continuations. When comparing our STM that performs partial aborts against one that performs full aborts, we achieve a decrease in execution time of up to 31% and an increase in throughput of up to 351%.

Mesoscale ocean eddies are a critical component of the Earth System as they dominate the ocean's kinetic energy and impact the global distribution of oceanic heat, salinity, momentum, and nutrients. Therefore, accurately representing these dynamic features is critical for our planet's sustainability. The majority of methods that identify eddies from satellite observations analyze the data in a frame-by-frame basis despite the fact that eddies are dynamic objects that propagate across space and time. We introduce the notion of spatio-temporal consistency to identify eddies in a continuous spatio-temporal field, to simultaneously ensure that the features detected are both spatially and temporally consistent. Our spatio-temporal consistency approach allows us to remove most of the expert criteria used in traditional methods to reduce false negatives. The removal of arbitrary heuristics enables us to render more complete eddy dynamics by identifying smaller and longer lived eddies compared to existing methods.

Inlining is an optimization that replaces a call to a function with that function's body. This optimization not only reduces the overhead of a function call, but can expose additional optimization opportunities to the compiler, such as removing redundant operations or unused conditional branches. Another optimization, copy propagation, replaces a redundant copy of a still-live variable with the original. Copy propagation can reduce the total number of live variables, reducing register pressure and memory usage, and possibly eliminating redundant memory-to-memory copies. In practice, both of these optimizations are implemented in nearly every modern compiler.

This paper describes a compiler extension to our prototype extensible C translator that adds new features for parallel execution of matrix operations and shows their application to problems in spatio-temporal data mining. The extension provides new language features for constructing new matrices, mapping functions over elements of a matrix, and accumulating operations that, for example, can sum values in a matrix. It also provides the appropriate semantic analysis to check for errors before translating the constructs down to parallel C code. The extension also provides features that let the programmer indicate how the extension translates these matrix constructs down to C code. Programmers seeking higher levels of performance can specify how the underlying for-loops are structured so that code using, for example, loop-tiling techniques or vector processors, is generated. In general, compiler extensions supported by our approach allow new domain-specific syntax and semantic analyses to be easily added to the host language. Specifications of the host C language and the extensions are composed to create a custom translator that maps extended C programs down to plain (parallel) C code, checking for domain-specific errors and applying high-level domain-specific optimizations in the process.

Mesoscale ocean eddies transport heat, salt, energy, and nutrients across oceans. As a result, accurately identifying and tracking such phenomena are crucial for understanding ocean dynamics and marine ecosystem sustainability. Traditionally, ocean eddies are monitored through two phases: identification and tracking. A major challenge for such an approach is that the tracking phase is dependent on the performance of the identification scheme, which can be susceptible to noise and sampling errors. In this paper, we focus on tracking, and introduce the concept of multiple hypothesis assignment (MHA), which extends traditional multiple hypothesis tracking for cases where the features tracked are noisy or uncertain. Under this scheme, features are assigned to multiple potential tracks, and the final assignment is deferred until more data are available to make a relatively unambiguous decision. Unlike the most widely used methods in the eddy tracking literature, MHA uses contextual spatio-temporal information to take corrective measures autonomously on the detection step a pos- teriori and performs significantly better in the presence of noise. This study is also the first to empirically analyze the relative robustness of eddy tracking algorithms.

As spatio-temporal data have become ubiquitous, an increasing challenge facing computer scientists is that of identifying discrete patterns in continuous spatio-temporal fields. In this paper, we introduce a parameter-free pattern mining application that is able to identify dynamic anomalies in ocean data, known as ocean eddies. Despite ocean eddy monitoring being an active field of research, we provide one of the first quantitative analyses of the performance of the most used monitoring algorithms. We present an incomplete information validation technique, that uses the performance of two methods to construct an imperfect ground truth to test the significance of patterns discovered as well as the relative performance of pattern mining algorithms. These methods, in addition to the validation schemes discussed provide researchers new directions in analyzing large unlabeled climate datasets.