Speaker segmentation consists in partitioning a conversation between one or more speakers into speaker turns. Usually addressed as the late combination of three sub-tasks (voice activity detection, speaker change detection, and overlapped speech detection), we propose to train an end-to-end segmentation model that does it directly. Inspired by the original end-to-end neural speaker diarization approach (EEND), the task is modeled as a multi-label classification problem using permutation-invariant training. The main difference is that our model operates on short audio chunks (5 seconds) but at a much higher temporal resolution (every 16ms). Experiments on multiple speaker diarization datasets conclude that our model can be used with great success on both voice activity detection and overlapped speech detection. Our proposed model can also be used as a post-processing step, to detect and correctly assign overlapped speech regions. Relative diarization error rate improvement over the best considered baseline (VBx) reaches 17% on AMI, 13% on DIHARD 3, and 13% on VoxConverse.

In this paper, we introduce How2, a multimodal collection of instructional videos with English subtitles and crowdsourced Portuguese translations. We also present integrated sequence-to-sequence baselines for machine translation, automatic speech recognition, spoken language translation, and multimodal summarization. By making available data and code for several multimodal natural language tasks, we hope to stimulate more research on these and similar challenges, to obtain a deeper understanding of multimodality in language processing.

Many modern NLP systems rely on word embeddings, previously trained in an unsupervised manner on large corpora, as base features. Efforts to obtain embeddings for larger chunks of text, such as sentences, have however not been so successful. Several attempts at learning unsupervised representations of sentences have not reached satisfactory enough performance to be widely adopted. In this paper, we show how universal sentence representations trained using the supervised data of the Stanford Natural Language Inference datasets can consistently outperform unsupervised methods like SkipThought vectors on a wide range of transfer tasks. Much like how computer vision uses ImageNet to obtain features, which can then be transferred to other tasks, our work tends to indicate the suitability of natural language inference for transfer learning to other NLP tasks. Our encoder is publicly available.

Factored neural machine translation (FNMT) is founded on the idea of using the morphological and grammatical decomposition of the words (factors) at the output side of the neural network. This architecture addresses two well-known problems occurring in MT, namely the size of target language vocabulary and the number of unknown tokens produced in the translation. FNMT system is designed to manage larger vocabulary and reduce the training time (for systems with equivalent target language vocabulary size). Moreover, we can produce grammatically correct words that are not part of the vocabulary. FNMT model is evaluated on IWSLT'15 English to French task and compared to the baseline word-based and BPE-based NMT systems. Promising qualitative and quantitative results (in terms of BLEU and METEOR) are reported.

Speech data carries a range of personal information, such as the speaker's identity and emotional state. These attributes can be used for malicious purposes. With the development of virtual assistants, a new generation of privacy threats has emerged. Current studies have addressed the topic of preserving speech privacy. One of them, the VoicePrivacy initiative aims to promote the development of privacy preservation tools for speech technology. The task selected for the VoicePrivacy 2020 Challenge (VPC) is about speaker anonymization. The goal is to hide the source speaker's identity while preserving the linguistic information. The baseline of the VPC makes use of a voice conversion. This paper studies the impact of the speaker anonymization baseline system of the VPC on emotional information present in speech utterances. Evaluation is performed following the VPC rules regarding the attackers' knowledge about the anonymization system. Our results show that the VPC baseline system does not suppress speakers' emotions against informed attackers. When comparing anonymized speech to original speech, the emotion recognition performance is degraded by 15\% relative to IEMOCAP data, similar to the degradation observed for automatic speech recognition used to evaluate the preservation of the linguistic information.

We present a new approach for neural machine translation (NMT) using the morphological and grammatical decomposition of the words (factors) in the output side of the neural network. This architecture addresses two main problems occurring in MT, namely dealing with a large target language vocabulary and the out of vocabulary (OOV) words. By the means of factors, we are able to handle larger vocabulary and reduce the training time (for systems with equivalent target language vocabulary size). In addition, we can produce new words that are not in the vocabulary. We use a morphological analyser to get a factored representation of each word (lemmas, Part of Speech tag, tense, person, gender and number). We have extended the NMT approach with attention mechanism in order to have two different outputs, one for the lemmas and the other for the rest of the factors. The final translation is built using some \textit{a priori} linguistic information. We compare our extension with a word-based NMT system. The experiments, performed on the IWSLT'15 dataset translating from English to French, show that while the performance do not always increase, the system can manage a much larger vocabulary and consistently reduce the OOV rate. We observe up to 2% BLEU point improvement in a simulated out of domain translation setup.

We present an end-to-end approach to extract semantic concepts directly from the speech audio signal. To overcome the lack of data available for this spoken language understanding approach, we investigate the use of a transfer learning strategy based on the principles of curriculum learning. This approach allows us to exploit out-of-domain data that can help to prepare a fully neural architecture. Experiments are carried out on the French MEDIA and PORTMEDIA corpora and show that this end-to-end SLU approach reaches the best results ever published on this task. We compare our approach to a classical pipeline approach that uses ASR, POS tagging, lemmatizer, chunker... and other NLP tools that aim to enrich ASR outputs that feed an SLU text to concepts system. Last, we explore the promising capacity of our end-to-end SLU approach to address the problem of domain portability.

This paper explores various attack scenarios on a voice anonymization system using embeddings alignment techniques. We use Wasserstein-Procrustes (an algorithm initially designed for unsupervised translation) or Procrustes analysis to match two sets of x-vectors, before and after voice anonymization, to mimic this transformation as a rotation function. We compute the optimal rotation and compare the results of this approximation to the official Voice Privacy Challenge results. We show that a complex system like the baseline of the Voice Privacy Challenge can be approximated by a rotation, estimated using a limited set of x-vectors. This paper studies the space of solutions for voice anonymization within the specific scope of rotations. Rotations being reversible, the proposed method can recover up to 62% of the speaker identities from anonymized embeddings.

A step-reinforced random walk is a discrete-time stochastic process with long-range dependence. At each step, with a fixed probability $\alpha$, the so-called positively step-reinforced random walk repeats one of its previous steps, chosen randomly and uniformly from its entire history. Alternatively, with probability $1-\alpha$, it makes an independent move. For the so-called negatively step-reinforced random walk, the process is similar, but any repeated step is taken with its direction reversed. These random walks have been introduced respectively by Simon (1955) and Bertoin (2024) and are sometimes refered to the self-confident step-reinforced random walk and the counterbalanced step-reinforced random walk respectively. In this work, we introduce a new class of unbalanced step-reinforced random walks for which we prove the strong law of large numbers and the central limit theorem. In particular, our work provides a unified treatment of the elephant random walk introduced by Schutz and Trimper (2004) and the positively and negatively step-reinforced random walks.

In this paper, we present a probabilistic numerical method for a class of forward utilities in a stochastic factor model. For this purpose, we use the representation of dynamic consistent utilities with mean of ergodic Backward Stochastic Differential Equations (eBSDEs) introduced by Liang and Zariphopoulou in [27]. We establish a connection between the solution of the ergodic BSDE and the solution of an associated BSDE with random terminal time $\tau$ , defined as the hitting time of the positive recurrent stochastic factor V . The viewpoint based on BSDEs with random horizon yields a new characterization of the ergodic cost $\lambda$ which is a part of the solution of the eBSDEs. In particular, for a certain class of eBSDEs with quadratic generator, the Cole-Hopf transform leads to a semi-explicit representation of the solution as well as a new expression of the ergodic cost $\lambda$. The latter can be estimated with Monte Carlo methods. We also propose two new deep learning numerical schemes for eBSDEs, where the ergodic cost $\lambda$ is optimized according to a loss function at the random horizon $\tau$ or taking into account the whole trajectory. Finally, we present numerical results for different examples of eBSDEs and forward utilities along with the associated investment strategies.

This paper presents a semi-automatic approach to create a diachronic corpus of voices balanced for speaker's age, gender, and recording period, according to 32 categories (2 genders, 4 age ranges and 4 recording periods). Corpora were selected at French National Institute of Audiovisual (INA) to obtain at least 30 speakers per category (a total of 960 speakers; only 874 have be found yet). For each speaker, speech excerpts were extracted from audiovisual documents using an automatic pipeline consisting of speech detection, background music and overlapped speech removal and speaker diarization, used to present clean speaker segments to human annotators identifying target speakers. This pipeline proved highly effective, cutting down manual processing by a factor of ten. Evaluation of the quality of the automatic processing and of the final output is provided. It shows the automatic processing compare to up-to-date process, and that the output provides high quality speech for most of the selected excerpts. This method shows promise for creating large corpora of known target speakers.

Transition Relevance Places are defined as the end of an utterance where the interlocutor may take the floor without interrupting the current speaker --i.e., a place where the turn is terminal. Analyzing turn terminality is useful to study the dynamic of turn-taking in spontaneous conversations. This paper presents an automatic classification of spoken utterances as Terminal or Non-Terminal in multi-speaker settings. We compared audio, text, and fusions of both approaches on a French corpus of TV and Radio extracts annotated with turn-terminality information at each speaker change. Our models are based on pre-trained self-supervised representations. We report results for different fusion strategies and varying context sizes. This study also questions the problem of performance variability by analyzing the differences in results for multiple training runs with random initialization. The measured accuracy would allow the use of these models for large-scale analysis of turn-taking.

The purpose of this paper is to estimate a representative and detailed map of the harvest value in wine using structured and unstructured open data sources. With climate change and new environmental and ecological policies, wine producers are facing new challenges. The ability to model the evolution of these risks is strategic for wine producers and research in order to adapt. Many research projects require the values exposed to risk. For example, to assess the economic impact of risks or the premium of crop insurance, or to choose between different agroecological solutions in a cost-benefit approach. The high spatial heterogeneity and complexity of wine characteristics add to the challenge of these production values and the need to improve our spatial assessment of these harvest-expected values.Structured, exhaustive and detailed historical data are collected by the customs services, but they are not open. To achieve this, we combine the aggregate of the vineyard register and the data of the Public Body for Products of Official Quality and Origin. There are several techniques available to merge, combine or complete missing data. We have chosen to use optimization methods to re-estimate the area by appellation and by county, which can then be converted into expected harvest values using olympic average yields by appellation and crop insurance prices. This approach allows us to capture the heterogeneity in production values faced by different vineyards, thereby facilitating further research on risk assessment in the wine industry.

Biometric recognition systems are security systems based on intrinsic properties of their users, usually encoded in high dimension representations called embeddings, which potential theft would represent a greater threat than a temporary password or a replaceable key. To study the threat of embedding theft, we perform spoofing attacks on two behavioral biometric systems (an automatic speaker verification system and a handwritten digit analysis system) using a set of alignment techniques. Biometric recognition systems based on embeddings work in two phases: enrollment - where embeddings are collected and stored - then authentication - when new embeddings are compared to the stored ones -.The threat of stolen enrollment embeddings has been explored by the template reconstruction attack literature: reconstructing the original data to spoof an authentication system is doable with black-box access to their encoder. In this document, we explore the options available to perform template reconstruction attacks without any access to the encoder. To perform those attacks, we suppose general rules over the distribution of embeddings across encoders and use supervised and unsupervised algorithms to align an unlabeled set of embeddings with a set from a known encoder. The use of an alignment algorithm from the unsupervised translation literature gives promising results on spoofing two behavioral biometric systems.

Homophones present a significant challenge to authors in any languages due to their similarities of pronunciations but different meanings and spellings. This issue is particularly pronounced in the Khmer language, rich in homophones due to its complex structure and extensive character set. This research aims to address the difficulties faced by Khmer authors when using homophones in their writing and proposes potential solutions based on an extensive literature review and survey analysis. A survey of 108 Khmer native speakers, including students, employees, and professionals, revealed that many frequently encounter challenges with homophones in their writing, often struggling to choose the correct word based on context. The survey also highlighted the absence of effective tools to address homophone errors in Khmer, which complicates the writing process. Additionally, a review of existing studies on spelling correction in other languages, such as English, Azerbaijani, and Bangla, identified a lack of research focused specifically on homophones, particularly in the Khmer language. In summary, this research highlights the necessity for a specialized tool to address Khmer homophone errors. By bridging current gaps in research and available resources, such a tool would enhance the confidence and accuracy of Khmer authors in their writing, thereby contributing to the enrichment and preservation of the language. Continued efforts in this domain are essential for ensuring that Khmer can leverage advancements in technology and linguistics effectively.

We study a n-player and mean-field portfolio optimization problem under relative performance concerns with non-zero volatility, for wealth and consumption. The consistency assumption defining forward relative performance processes leads to a sufficient characterization of such processes with mean of a Stochastic HJB equations, which highlights the link between wealth and consumption utility, and also characterizes the optimal strategies. In particular, forward relative performance processes with a wealth utility of CRRA type and separable time and space dependence necessarily have a consumption utility of the same form, with the same risk aversion parameter. This characterization gives a better understanding of the drift condition ensuring time consistency. In this setting, we establish closed form of the Nash equilibrium for both the n-player and mean eld problems. We also provide some numerical examples.

Speech signals contain a lot of sensitive information, such as the speaker's identity, which raises privacy concerns when speech data get collected. Speaker anonymization aims to transform a speech signal to remove the source speaker's identity while leaving the spoken content unchanged. Current methods perform the transformation by relying on content/speaker disentanglement and voice conversion. Usually, an acoustic model from an automatic speech recognition system extracts the content representation while an x-vector system extracts the speaker representation. Prior work has shown that the extracted features are not perfectly disentangled. This paper tackles how to improve features disentanglement, and thus the converted anonymized speech. We propose enhancing the disentanglement by removing speaker information from the acoustic model using vector quantization. Evaluation done using the VoicePrivacy 2022 toolkit showed that vector quantization helps conceal the original speaker identity while maintaining utility for speech recognition.

It is today acknowledged that neural network language models outperform backoff language models in applications like speech recognition or statistical machine translation. However, training these models on large amounts of data can take several days. We present efficient techniques to adapt a neural network language model to new data. Instead of training a completely new model or relying on mixture approaches, we propose two new methods: continued training on resampled data or insertion of adaptation layers. We present experimental results in an CAT environment where the post-edits of professional translators are used to improve an SMT system. Both methods are very fast and achieve significant improvements without overfitting the small adaptation data.

In this paper we study a zero-sum switching game and its verification theorems expressed in terms of either a system of Reflected Backward Stochastic Differential Equations (RBSDEs in short) with bilateral interconnected obstacles or a system of parabolic partial differential equations (PDEs in short) with bilateral interconnected obstacles as well. We show that each one of the systems has a unique solution. Then we show that the game has a value.

In this paper, we propose a novel end-to-end sequence-to-sequence spoken language understanding model using an attention mechanism. It reliably selects contextual acoustic features in order to hypothesize semantic contents. An initial architecture capable of extracting all pronounced words and concepts from acoustic spans is designed and tested. With a shallow fusion language model, this system reaches a 13.6 concept error rate (CER) and an 18.5 concept value error rate (CVER) on the French MEDIA corpus, achieving an absolute 2.8 points reduction compared to the state-of-the-art. Then, an original model is proposed for hypothesizing concepts and their values. This transduction reaches a 15.4 CER and a 21.6 CVER without any new type of context.