Numerous examples in the literature proved that deep learning models have the ability to work well with multimodal data. Recently, CLIP has enabled deep learning systems to learn shared latent spaces between images and text descriptions, with outstanding zero- or few-shot results in downstream tasks. In this paper we explore the same idea proposed by CLIP but applied to the speech domain, where the phonetic and acoustic spaces usually coexist. We train a CLIP-based model with the aim to learn shared representations of phonetic and acoustic spaces. The results show that the proposed model is sensible to phonetic changes, with a 91% of score drops when replacing 20% of the phonemes at random, while providing substantial robustness against different kinds of noise, with a 10% performance drop when mixing the audio with 75% of Gaussian noise. We also provide empirical evidence showing that the resulting embeddings are useful for a variety of downstream applications, such as intelligibility evaluation and the ability to leverage rich pre-trained phonetic embeddings in speech generation task. Finally, we discuss potential applications with interesting implications for the speech generation and recognition fields.

Recently normalizing flows have been gaining traction in text-to-speech (TTS) and voice conversion (VC) due to their state-of-the-art (SOTA) performance. Normalizing flows are unsupervised generative models. In this paper, we introduce supervision to the training process of normalizing flows, without the need for parallel data. We call this training paradigm AutoEncoder Normalizing Flow (AE-Flow). It adds a reconstruction loss forcing the model to use information from the conditioning to reconstruct an audio sample. Our goal is to understand the impact of each component and find the right combination of the negative log-likelihood (NLL) and the reconstruction loss in training normalizing flows with coupling blocks. For that reason we will compare flow-based mapping model trained with: (i) NLL loss, (ii) NLL and reconstruction losses, as well as (iii) reconstruction loss only. Additionally, we compare our model with SOTA VC baseline. The models are evaluated in terms of naturalness, speaker similarity, intelligibility in many-to-many and many-to-any VC settings. The results show that the proposed training paradigm systematically improves speaker similarity and naturalness when compared to regular training methods of normalizing flows. Furthermore, we show that our method improves speaker similarity and intelligibility over the state-of-the-art.

Non-parallel voice conversion (VC) is typically achieved using lossy representations of the source speech. However, ensuring only speaker identity information is dropped whilst all other information from the source speech is retained is a large challenge. This is particularly challenging in the scenario where at inference-time we have no knowledge of the text being read, i.e., text-free VC. To mitigate this, we investigate information-preserving VC approaches.Normalising flows have gained attention for text-to-speech synthesis, however have been under-explored for VC. Flows utilize invertible functions to learn the likelihood of the data, thus provide a lossless encoding of speech. We investigate normalising flows for VC in both text-conditioned and text-free scenarios. Furthermore, for text-free VC we compare pre-trained and jointly-learnt priors. Flow-based VC evaluations show no degradation between text-free and text-conditioned VC, resulting in improvements over the state-of-the-art. Also, joint-training of the prior is found to negatively impact text-free VC quality.

Creating realistic and natural-sounding synthetic speech remains a big challenge for voice identities unseen during training. As there is growing interest in synthesizing voices of new speakers, here we investigate the ability of normalizing flows in text-to-speech (TTS) and voice conversion (VC) modes to extrapolate from speakers observed during training to create unseen speaker identities. Firstly, we create an approach for TTS and VC, and then we comprehensively evaluate our methods and baselines in terms of intelligibility, naturalness, speaker similarity, and ability to create new voices. We use both objective and subjective metrics to benchmark our techniques on 2 evaluation tasks: zero-shot and new voice speech synthesis. The goal of the former task is to measure the precision of the conversion to an unseen voice. The goal of the latter is to measure the ability to create new voices. Extensive evaluations demonstrate that the proposed approach systematically allows to obtain state-of-the-art performance in zero-shot speech synthesis and creates various new voices, unobserved in the training set. We consider this work to be the first attempt to synthesize new voices based on mel-spectrograms and normalizing flows, along with a comprehensive analysis and comparison of the TTS and VC modes.

Regional accents of the same language affect not only how words are pronounced (i.e., phonetic content), but also impact prosodic aspects of speech such as speaking rate and intonation. This paper investigates a novel flow-based approach to accent conversion using normalizing flows. The proposed approach revolves around three steps: remapping the phonetic conditioning, to better match the target accent, warping the duration of the converted speech, to better suit the target phonemes, and an attention mechanism that implicitly aligns source and target speech sequences. The proposed remap-warp-attend system enables adaptation of both phonetic and prosodic aspects of speech while allowing for source and converted speech signals to be of different lengths. Objective and subjective evaluations show that the proposed approach significantly outperforms a competitive CopyCat baseline model in terms of similarity to the target accent, naturalness and intelligibility.

Extreme precipitation events, such as violent rainfall and hail storms, routinely ravage economies and livelihoods around the developing world. Climate change further aggravates this issue. Data-driven deep learning approaches could widen the access to accurate multi-day forecasts, to mitigate against such events. However, there is currently no benchmark dataset dedicated to the study of global precipitation forecasts. In this paper, we introduce RainBench, a new multi-modal benchmark dataset for data-driven precipitation forecasting. It includes simulated satellite data, a selection of relevant meteorological data from the ERA5 reanalysis product, and IMERG precipitation data. We also release PyRain, a library to process large precipitation datasets efficiently. We present an extensive analysis of our novel dataset and establish baseline results for two benchmark medium-range precipitation forecasting tasks. Finally, we discuss existing data-driven weather forecasting methodologies and suggest future research avenues.

Creating realistic human videos entails the challenge of being able to simultaneously generate both appearance, as well as motion. To tackle this challenge, we introduce G3AN, a novel spatio-temporal generative model, which seeks to capture the distribution of high dimensional video data and to model appearance and motion in disentangled manner. The latter is achieved by decomposing appearance and motion in a three-stream Generator, where the main stream aims to model spatio-temporal consistency, whereas the two auxiliary streams augment the main stream with multi-scale appearance and motion features, respectively. An extensive quantitative and qualitative analysis shows that our model systematically and significantly outperforms state-of-the-art methods on the facial expression datasets MUG and UvA-NEMO, as well as the Weizmann and UCF101 datasets on human action. Additional analysis on the learned latent representations confirms the successful decomposition of appearance and motion.

Generating human videos based on single images entails the challenging simultaneous generation of realistic and visual appealing appearance and motion. In this context, we propose a novel conditional GAN architecture, namely ImaGINator, which given a single image, a condition (label of a facial expression or action) and noise, decomposes appearance and motion in both latent and high level feature spaces, generating realistic videos. This is achieved by (i) a novel spatio-temporal fusion scheme, which generates dynamic motion, while retaining appearance throughout the full video sequence by transmitting appearance (originating from the single image) through all layers of the network. In addition, we propose (ii) a novel transposed (1+2)D convolution, factorizing the transposed 3D convolutional filters into separate transposed temporal and spatial components, which yields significantly gains in video quality and speed. We extensively evaluate our approach on the facial expression datasets MUG and UvA-NEMO, as well as on the action datasets NATOPS and Weizmann. We show that our approach achieves significantly better quantitative and qualitative results than the state-of-the-art. The source code and models are available under https://github.com/wyhsirius/ImaGINator.

We propose a novel approach for rapid segmentation of flooded buildings by fusing multiresolution, multisensor, and multitemporal satellite imagery in a convolutional neural network. Our model significantly expedites the generation of satellite imagery-based flood maps, crucial for first responders and local authorities in the early stages of flood events. By incorporating multitemporal satellite imagery, our model allows for rapid and accurate post-disaster damage assessment and can be used by governments to better coordinate medium- and long-term financial assistance programs for affected areas. The network consists of multiple streams of encoder-decoder architectures that extract spatiotemporal information from medium-resolution images and spatial information from high-resolution images before fusing the resulting representations into a single medium-resolution segmentation map of flooded buildings. We compare our model to state-of-the-art methods for building footprint segmentation as well as to alternative fusion approaches for the segmentation of flooded buildings and find that our model performs best on both tasks. We also demonstrate that our model produces highly accurate segmentation maps of flooded buildings using only publicly available medium-resolution data instead of significantly more detailed but sparsely available very high-resolution data. We release the first open-source dataset of fully preprocessed and labeled multiresolution, multispectral, and multitemporal satellite images of disaster sites along with our source code.

Informal settlements are home to the most socially and economically vulnerable people on the planet. In order to deliver effective economic and social aid, non-government organizations (NGOs), such as the United Nations Children's Fund (UNICEF), require detailed maps of the locations of informal settlements. However, data regarding informal and formal settlements is primarily unavailable and if available is often incomplete. This is due, in part, to the cost and complexity of gathering data on a large scale. To address these challenges, we, in this work, provide three contributions. 1) A brand new machine learning dataset purposely developed for informal settlement detection. 2) We show that it is possible to detect informal settlements using freely available low-resolution (LR) data, in contrast to previous studies that use very-high resolution~(VHR) satellite and aerial imagery, something that is cost-prohibitive for NGOs. 3) We demonstrate two effective classification schemes on our curated data set, one that is cost-efficient for NGOs and another that is cost-prohibitive for NGOs, but has additional utility. We integrate these schemes into a semi-automated pipeline that converts either a LR or VHR satellite image into a binary map that encodes the locations of informal settlements.

We propose a novel end-to-end trainable, deep, encoder-decoder architecture for single-pass semantic segmentation. Our approach is based on a cascaded architecture with feature-level long-range skip connections. The encoder incorporates the structure of ResNeXt's residual building blocks and adopts the strategy of repeating a building block that aggregates a set of transformations with the same topology. The decoder features a novel architecture, consisting of blocks, that (i) capture context information, (ii) generate semantic features, and (iii) enable fusion between different output resolutions. Crucially, we introduce dense decoder shortcut connections to allow decoder blocks to use semantic feature maps from all previous decoder levels, i.e. from all higher-level feature maps. The dense decoder connections allow for effective information propagation from one decoder block to another, as well as for multi-level feature fusion that significantly improves the accuracy. Importantly, these connections allow our method to obtain state-of-the-art performance on several challenging datasets, without the need of time-consuming multi-scale averaging of previous works.

Body height, weight, as well as the associated and composite body mass index (BMI) are human attributes of pertinence due to their use in a number of applications including surveillance, re-identification, image retrieval systems, as well as healthcare. Previous work on automated estimation of height, weight and BMI has predominantly focused on 2D and 3D full-body images and videos. Little attention has been given to the use of face for estimating such traits. Motivated by the above, we here explore the possibility of estimating height, weight and BMI from single-shot facial images by proposing a regression method based on the 50-layers ResNet-architecture. In addition, we present a novel dataset consisting of 1026 subjects and show results, which suggest that facial images contain discriminatory information pertaining to height, weight and BMI, comparable to that of body-images and videos. Finally, we perform a gender-based analysis of the prediction of height, weight and BMI.

In this paper, we focus on the important topic of violence recognition and detection in surveillance videos. Our goal is to determine if a violence occurs in a video (recognition) and when it happens (detection). Firstly, we propose an extension of the Improved Fisher Vectors (IFV) for videos, which allows to represent a video using both local features and their spatio-temporal positions. Then, we study the popular sliding window approach for violence detection, and we re-formulate the Improved Fisher Vectors and use the summed area table data structure to speed up the approach. We present an extensive evaluation, comparison and analysis of the proposed improvements on 4 state-of-the-art datasets. We show that the proposed improvements make the violence recognition more accurate (as compared to the standard IFV, IFV with spatio-temporal grid, and other state-of-the-art methods) and make the violence detection significantly faster.

This paper addresses a problem of recognizing human actions in video sequences. Recent studies have shown that methods which use bag-of-features and space-time features achieve high recognition accuracy. Such methods extract both appearance-based and motion-based features. This paper focuses only on appearance features. We propose to model relationships between different pixel-level appearance features such as intensity and gradient using Brownian covariance, which is a natural extension of classical covariance measure. While classical covariance can model only linear relationships, Brownian covariance models all kinds of possible relationships. We propose a method to compute Brownian covariance on space-time volume of a video sequence. We show that proposed Video Brownian Covariance (VBC) descriptor carries complementary information to the Histogram of Oriented Gradients (HOG) descriptor. The fusion of these two descriptors gives a significant improvement in performance on three challenging action recognition datasets.

We propose a method for the synthesis of the magnitudes of Head-related Transfer Functions (HRTFs) using a sparse representation of anthropometric features. Our approach treats the HRTF synthesis problem as finding a sparse representation of the subject's anthropometric features w.r.t. the anthropometric features in the training set. The fundamental assumption is that the magnitudes of a given HRTF set can be described by the same sparse combination as the anthropometric data. Thus, we learn a sparse vector that represents the subject's anthropometric features as a linear superposition of the anthropometric features of a small subset of subjects from the training data. Then, we apply the same sparse vector directly on the HRTF tensor data. For evaluation purpose we use a new dataset, containing both anthropometric features and HRTFs. We compare the proposed sparse representation based approach with ridge regression and with the data of a manikin (which was designed based on average anthropometric data), and we simulate the best and the worst possible classifiers to select one of the HRTFs from the dataset. For instrumental evaluation we use log-spectral distortion. Experiments show that our sparse representation outperforms all other evaluated techniques, and that the synthesized HRTFs are almost as good as the best possible HRTF classifier.

Recent development in affordable depth sensors opens new possibilities in action recognition problem. Depth information improves skeleton detection, therefore many authors focused on analyzing pose for action recognition. But still skeleton detection is not robust and fail in more challenging scenarios, where sensor is placed outside of optimal working range and serious occlusions occur. In this paper we investigate state-of-the-art methods designed for RGB videos, which have proved their performance. Then we extend current state-of-the-art algorithms to benefit from depth information without need of skeleton detection. In this paper we propose two novel video descriptors. First combines motion and 3D information. Second improves performance on actions with low movement rate. We validate our approach on challenging MSR Daily Activty 3D dataset.

The bag-of-words approach with local spatio-temporal features have become a popular video representation for action recognition. Recent methods have typically focused on capturing global and local statistics of features. However, existing approaches ignore relations between the features, particularly space-time arrangement of features, and thus may not be discriminative enough. Therefore, we propose a novel figure-centric representation which captures both local density of features and statistics of space-time ordered features. Using two benchmark datasets for human action recognition, we demonstrate that our representation enhances the discriminative power of features and improves action recognition performance, achieving 96.16% recognition rate on popular KTH action dataset and 93.33% on challenging ADL dataset.