Search

Recent research has demonstrated that training a linear connector between speech foundation encoders and large language models (LLMs) enables this architecture to achieve strong ASR capabilities. Despite the impressive results, it remains unclear whether these simple approaches are robust enough across different scenarios and speech conditions, such as domain shifts and different speech perturbations. In this paper, we address these questions by conducting various ablation experiments using a recent and widely adopted approach called SLAM-ASR. We present novel empirical findings that offer insights on how to effectively utilize the SLAM-ASR architecture across a wide range of settings. Our main findings indicate that the SLAM-ASR exhibits poor performance in cross-domain evaluation settings. Additionally, speech perturbations within in-domain data, such as changes in speed or the presence of additive noise, can significantly impact performance. Our findings offer critical insights for fine-tuning and configuring robust LLM-based ASR models, tailored to different data characteristics and computational resources., Comment: Submitted to ICASSP 2025 SALMA Workshop

Bias assessment of news sources is paramount for professionals, organizations, and researchers who rely on truthful evidence for information gathering and reporting. While certain bias indicators are discernible from content analysis, descriptors like political bias and fake news pose greater challenges. In this paper, we propose an extension to a recently presented news media reliability estimation method that focuses on modeling outlets and their longitudinal web interactions. Concretely, we assess the classification performance of four reinforcement learning strategies on a large news media hyperlink graph. Our experiments, targeting two challenging bias descriptors, factual reporting and political bias, showed a significant performance improvement at the source media level. Additionally, we validate our methods on the CLEF 2023 CheckThat! Lab challenge, outperforming the reported results in both, F1-score and the official MAE metric. Furthermore, we contribute by releasing the largest annotated dataset of news source media, categorized with factual reporting and political bias labels. Our findings suggest that profiling news media sources based on their hyperlink interactions over time is feasible, offering a bird's-eye view of evolving media landscapes., Comment: Accepted to CLEF 2024

Despite the recent success of end-to-end models for automatic speech recognition, recognizing special rare and out-of-vocabulary words, as well as fast domain adaptation with text, are still challenging. It often happens that biasing to the special entities leads to a degradation in the overall performance. We propose a light on-the-fly method to improve automatic speech recognition performance by combining a bias list of named entities with a word-level n-gram language model with the shallow fusion approach based on the Aho-Corasick string matching algorithm. The Aho-Corasick algorithm has proved to be more efficient than other methods and allows fast context adaptation. An n-gram language model is introduced as a graph with fail and output arcs, where the arc weights are adapted from the n-gram probabilities. The language model is used as an additional support to keyword biasing when the language model is combined with bias entities in a single context graph to take care of the overall performance. We demonstrate our findings on 4 languages, 2 public and 1 private datasets including performance on named entities and out-of-vocabulary entities. We achieve up to 21.6% relative improvement in the general word error rate with no practical difference in the inverse real-time factor., Comment: Submitted to ICASSP2025

The training of automatic speech recognition (ASR) with little to no supervised data remains an open question. In this work, we demonstrate that streaming Transformer-Transducer (TT) models can be trained from scratch in consumer and accessible GPUs in their entirety with pseudo-labeled (PL) speech from foundational speech models (FSM). This allows training a robust ASR model just in one stage and does not require large data and computational budget compared to the two-step scenario with pre-training and fine-tuning. We perform a comprehensive ablation on different aspects of PL-based streaming TT models such as the impact of (1) shallow fusion of n-gram LMs, (2) contextual biasing with named entities, (3) chunk-wise decoding for low-latency streaming applications, and (4) TT overall performance as the function of the FSM size. Our results demonstrate that TT can be trained from scratch without supervised data, even with very noisy PLs. We validate the proposed framework on 6 languages from CommonVoice and propose multiple heuristics to filter out hallucinated PLs., Comment: Accepted to EMNLP Findings 2024

In traditional conversational intelligence from speech, a cascaded pipeline is used, involving tasks such as voice activity detection, diarization, transcription, and subsequent processing with different NLP models for tasks like semantic endpointing and named entity recognition (NER). Our paper introduces TokenVerse, a single Transducer-based model designed to handle multiple tasks. This is achieved by integrating task-specific tokens into the reference text during ASR model training, streamlining the inference and eliminating the need for separate NLP models. In addition to ASR, we conduct experiments on 3 different tasks: speaker change detection, endpointing, and NER. Our experiments on a public and a private dataset show that the proposed method improves ASR by up to 7.7% in relative WER while outperforming the cascaded pipeline approach in individual task performance. Our code is publicly available: https://github.com/idiap/tokenverse-unifying-speech-nlp, Comment: Accepted at EMNLP 2024 (Main Conference)

Self-supervised pretrained models exhibit competitive performance in automatic speech recognition on finetuning, even with limited in-domain supervised data. However, popular pretrained models are not suitable for streaming ASR because they are trained with full attention context. In this paper, we introduce XLSR-Transducer, where the XLSR-53 model is used as encoder in transducer setup. Our experiments on the AMI dataset reveal that the XLSR-Transducer achieves 4% absolute WER improvement over Whisper large-v2 and 8% over a Zipformer transducer model trained from scratch. To enable streaming capabilities, we investigate different attention masking patterns in the self-attention computation of transformer layers within the XLSR-53 model. We validate XLSR-Transducer on AMI and 5 languages from CommonVoice under low-resource scenarios. Finally, with the introduction of attention sinks, we reduce the left context by half while achieving a relative 12% improvement in WER., Comment: 5 pages, double column

Automatic depression detection from conversational data has gained significant interest in recent years. The DAIC-WOZ dataset, interviews conducted by a human-controlled virtual agent, has been widely used for this task. Recent studies have reported enhanced performance when incorporating interviewer's prompts into the model. In this work, we hypothesize that this improvement might be mainly due to a bias present in these prompts, rather than the proposed architectures and methods. Through ablation experiments and qualitative analysis, we discover that models using interviewer's prompts learn to focus on a specific region of the interviews, where questions about past experiences with mental health issues are asked, and use them as discriminative shortcuts to detect depressed participants. In contrast, models using participant responses gather evidence from across the entire interview. Finally, to highlight the magnitude of this bias, we achieve a 0.90 F1 score by intentionally exploiting it, the highest result reported to date on this dataset using only textual information. Our findings underline the need for caution when incorporating interviewers' prompts into models, as they may inadvertently learn to exploit targeted prompts, rather than learning to characterize the language and behavior that are genuinely indicative of the patient's mental health condition., Comment: Accepted to Clinical NLP workshop at NAACL 2024

Evaluating the reliability of news sources is a routine task for journalists and organizations committed to acquiring and disseminating accurate information. Recent research has shown that predicting sources' reliability represents an important first-prior step in addressing additional challenges such as fake news detection and fact-checking. In this paper, we introduce a novel approach for source reliability estimation that leverages reinforcement learning strategies for estimating the reliability degree of news sources. Contrary to previous research, our proposed approach models the problem as the estimation of a reliability degree, and not a reliability label, based on how all the news media sources interact with each other on the Web. We validated the effectiveness of our method on a news media reliability dataset that is an order of magnitude larger than comparable existing datasets. Results show that the estimated reliability degrees strongly correlates with journalists-provided scores (Spearman=0.80) and can effectively predict reliability labels (macro-avg. F$_1$ score=81.05). We release our implementation and dataset, aiming to provide a valuable resource for the NLP community working on information verification., Comment: Accepted to NAACL 2024 Main Conference

We propose a simple approach for weighting self-connecting edges in a Graph Convolutional Network (GCN) and show its impact on depression detection from transcribed clinical interviews. To this end, we use a GCN for modeling non-consecutive and long-distance semantics to classify the transcriptions into depressed or control subjects. The proposed method aims to mitigate the limiting assumptions of locality and the equal importance of self-connections vs. edges to neighboring nodes in GCNs, while preserving attractive features such as low computational cost, data agnostic, and interpretability capabilities. We perform an exhaustive evaluation in two benchmark datasets. Results show that our approach consistently outperforms the vanilla GCN model as well as previously reported results, achieving an F1=0.84 on both datasets. Finally, a qualitative analysis illustrates the interpretability capabilities of the proposed approach and its alignment with previous findings in psychology., Comment: Paper Accepted to Interspeech 2023

GPU decoding significantly accelerates the output of ASR predictions. While GPUs are already being used for online ASR decoding, post-processing and rescoring on GPUs have not been properly investigated yet. Rescoring with available contextual information can considerably improve ASR predictions. Previous studies have proven the viability of lattice rescoring in decoding and biasing language model (LM) weights in offline and online CPU scenarios. In real-time GPU decoding, partial recognition hypotheses are produced without lattice generation, which makes the implementation of biasing more complex. The paper proposes and describes an approach to integrate contextual biasing in real-time GPU decoding while exploiting the standard Kaldi GPU decoder. Besides the biasing of partial ASR predictions, our approach also permits dynamic context switching allowing a flexible rescoring per each speech segment directly on GPU. The code is publicly released and tested with open-sourced test sets., Comment: Accepted to Interspeech 2023

In this paper, we perform an exhaustive evaluation of different representations to address the intent classification problem in a Spoken Language Understanding (SLU) setup. We benchmark three types of systems to perform the SLU intent detection task: 1) text-based, 2) lattice-based, and a novel 3) multimodal approach. Our work provides a comprehensive analysis of what could be the achievable performance of different state-of-the-art SLU systems under different circumstances, e.g., automatically- vs. manually-generated transcripts. We evaluate the systems on the publicly available SLURP spoken language resource corpus. Our results indicate that using richer forms of Automatic Speech Recognition (ASR) outputs, namely word-consensus-networks, allows the SLU system to improve in comparison to the 1-best setup (5.5% relative improvement). However, crossmodal approaches, i.e., learning from acoustic and text embeddings, obtains performance similar to the oracle setup, a relative improvement of 17.8% over the 1-best configuration, being a recommended alternative to overcome the limitations of working with automatically generated transcripts., Comment: Accepted in ICASSP 2023

In this paper, we describe our shared task submissions for Subtask 2 in CASE-2022, Event Causality Identification with Casual News Corpus. The challenge focused on the automatic detection of all cause-effect-signal spans present in the sentence from news-media. We detect cause-effect-signal spans in a sentence using T5 -- a pre-trained autoregressive language model. We iteratively identify all cause-effect-signal span triplets, always conditioning the prediction of the next triplet on the previously predicted ones. To predict the triplet itself, we consider different causal relationships such as cause$\rightarrow$effect$\rightarrow$signal. Each triplet component is generated via a language model conditioned on the sentence, the previous parts of the current triplet, and previously predicted triplets. Despite training on an extremely small dataset of 160 samples, our approach achieved competitive performance, being placed second in the competition. Furthermore, we show that assuming either cause$\rightarrow$effect or effect$\rightarrow$cause order achieves similar results., Comment: Camera-ready for CASE@EMNLP

In this paper, we describe our participation in the subtask 1 of CASE-2022, Event Causality Identification with Casual News Corpus. We address the Causal Relation Identification (CRI) task by exploiting a set of simple yet complementary techniques for fine-tuning language models (LMs) on a small number of annotated examples (i.e., a few-shot configuration). We follow a prompt-based prediction approach for fine-tuning LMs in which the CRI task is treated as a masked language modeling problem (MLM). This approach allows LMs natively pre-trained on MLM problems to directly generate textual responses to CRI-specific prompts. We compare the performance of this method against ensemble techniques trained on the entire dataset. Our best-performing submission was fine-tuned with only 256 instances per class, 15.7% of the all available data, and yet obtained the second-best precision (0.82), third-best accuracy (0.82), and an F1-score (0.85) very close to what was reported by the winner team (0.86)., Comment: To be published in CASE@EMNLP 2022 (5th Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text)

Engaged costumers are a very import part of current social media marketing. Public figures and brands have to be very careful about what to post online. That is why the need for accurate strategies for anticipating the impact of a post written for an online audience is critical to any public brand. Therefore, in this paper, we propose a method to predict the impact of a given post by accounting for the content, style, and behavioral attributes as well as metadata information. For validating our method we collected Facebook posts from 10 public pages, we performed experiments with almost 14000 posts and found that the content and the behavioral attributes from posts provide relevant information to our prediction model., Comment: Accepted at LKE 2019

Author Profiling (AP) aims at predicting specific characteristics from a group of authors by analyzing their written documents. Many research has been focused on determining suitable features for modeling writing patterns from authors. Reported results indicate that content-based features continue to be the most relevant and discriminant features for solving this task. Thus, in this paper, we present a thorough analysis regarding the appropriateness of different distributional term representations (DTR) for the AP task. In this regard, we introduce a novel framework for supervised AP using these representations and, supported on it. We approach a comparative analysis of representations such as DOR, TCOR, SSR, and word2vec in the AP problem. We also compare the performance of the DTRs against classic approaches including popular topic-based methods. The obtained results indicate that DTRs are suitable for solving the AP task in social media domains as they achieve competitive results while providing meaningful interpretability.

The Multi-Purpose Detector (MPD) is to be installed at the Nuclotron Ion Collider fAcility (NICA) of the Joint Institute for Nuclear Research (JINR). Its main goal is to study the phase diagram of the strongly interacting matter produced in heavy-ion collisions. These studies, while providing insight into the physics of heavy-ion collisions, are relevant for improving our understanding of the evolution of the early Universe and the formation of neutron stars. In order to extend the MPD trigger capabilities, we propose to include a high granularity beam-beam monitoring detector (BE-BE) to provide a level-0 trigger signal with an expected time resolution of 30 ps. This new detector will improve the determination of the reaction plane by the MPD experiment, a key measurement for flow studies that provides physics insight into the early stages of the reaction. In this work, we use simulated Au+Au collisions at NICA energies to show the potential of such a detector to determine the event plane resolution, providing further redundancy to the detectors originally considered for this purpose namely, the Fast Forward Detector (FFD) and the Hadron Calorimeter (HCAL). We also show our results for the time resolution studies of two prototype cells carried out at the T10 beam line at the CERN PS complex., Comment: 16 pages, 12 figures. Updated to published version with added comments and corrections

Resources such as labeled corpora are necessary to train automatic models within the natural language processing (NLP) field. Historically, a large number of resources regarding a broad number of problems are available mostly in English. One of such problems is known as Personality Identification where based on a psychological model (e.g. The Big Five Model), the goal is to find the traits of a subject's personality given, for instance, a text written by the same subject. In this paper we introduce a new corpus in Spanish called Texts for Personality Identification (TxPI). This corpus will help to develop models to automatically assign a personality trait to an author of a text document. Our corpus, TxPI-u, contains information of 416 Mexican undergraduate students with some demographics information such as, age, gender, and the academic program they are enrolled. Finally, as an additional contribution, we present a set of baselines to provide a comparison scheme for further research.

Paraphrase plagiarism identification represents a very complex task given that plagiarized texts are intentionally modified through several rewording techniques. Accordingly, this paper introduces two new measures for evaluating the relatedness of two given texts: a semantically-informed similarity measure and a semantically-informed edit distance. Both measures are able to extract semantic information from either an external resource or a distributed representation of words, resulting in informative features for training a supervised classifier for detecting paraphrase plagiarism. Obtained results indicate that the proposed metrics are consistently good in detecting different types of paraphrase plagiarism. In addition, results are very competitive against state-of-the art methods having the advantage of representing a much more simple but equally effective solution.

The goal of Author Profiling (AP) is to identify demographic aspects (e.g., age, gender) from a given set of authors by analyzing their written texts. Recently, the AP task has gained interest in many problems related to computer forensics, psychology, marketing, but specially in those related with social media exploitation. As known, social media data is shared through a wide range of modalities (e.g., text, images and audio), representing valuable information to be exploited for extracting valuable insights from users. Nevertheless, most of the current work in AP using social media data has been devoted to analyze textual information only, and there are very few works that have started exploring the gender identification using visual information. Contrastingly, this paper focuses in exploiting the visual modality to perform both age and gender identification in social media, specifically in Twitter. Our goal is to evaluate the pertinence of using visual information in solving the AP task. Accordingly, we have extended the Twitter corpus from PAN 2014, incorporating posted images from all the users, making a distinction between tweeted and retweeted images. Performed experiments provide interesting evidence on the usefulness of visual information in comparison with traditional textual representations for the AP task.