Weipeng He, Petr Motlicek, Jean-Marc Odobez
We propose to use neural networks for simultaneous detection and localization of multiple sound sources in human-robot interaction. In contrast to conventional signal processing techniques, neural network-based sound source localization methods require fewer strong assumptions about the environment. Previous neural network-based methods have been focusing on localizing a single sound source, which do not extend to multiple sources in terms of detection and localization. In this paper, we thus propose a likelihood-based encoding of the network output, which naturally allows the detection of an arbitrary number of sources. In addition, we investigate the use of sub-band cross-correlation information as features for better localization in sound mixtures, as well as three different network architectures based on different motivations. Experiments on real data recorded from a robot show that our proposed methods significantly outperform the popular spatial spectrum-based approaches.
Amrutha Prasad, Petr Motlicek, Srikanth Madikeri
State-of-the-art hybrid automatic speech recognition (ASR) system exploits deep neural network (DNN) based acoustic models (AM) trained with Lattice Free-Maximum Mutual Information (LF-MMI) criterion and n-gram language models. The AMs typically have millions of parameters and require significant parameter reduction to operate on embedded devices. The impact of parameter quantization on the overall word recognition performance is studied in this paper. Following approaches are presented: (i) AM trained in Kaldi framework with conventional factorized TDNN (TDNN-F) architecture, (ii) the TDNN AM built in Kaldi loaded into the PyTorch toolkit using a C++ wrapper for post-training quantization, (iii) quantization-aware training in PyTorch for Kaldi TDNN model, (iv) quantization-aware training in Kaldi. Results obtained on standard Librispeech setup provide an interesting overview of recognition accuracy w.r.t. applied quantization scheme.
Mael Fabien, Seyyed Saeed Sarfjoo, Petr Motlicek, Srikanth Madikeri
Criminal investigations mostly rely on the collection of speech conversational data in order to identify speakers and build or enrich an existing criminal network. Social network analysis tools are then applied to identify the most central characters and the different communities within the network. We introduce two candidate datasets for criminal conversational data, Crime Scene Investigation (CSI), a television show, and the ROXANNE simulated data. We also introduce the metric of conversation accuracy in the context of criminal investigations. By re-ranking candidate speakers based on the frequency of previous interactions, we improve the speaker identification baseline by 1.2% absolute (1.3% relative), and the conversation accuracy by 2.6% absolute (3.4% relative) on CSI data, and by 1.1% absolute (1.2% relative), and 2% absolute (2.5% relative) respectively on the ROXANNE simulated data.
Juan Zuluaga-Gomez, Petr Motlicek, Qingran Zhan, Karel Vesely, Rudolf Braun
Advances in Automatic Speech Recognition (ASR) over the last decade opened new areas of speech-based automation such as in Air-Traffic Control (ATC) environment. Currently, voice communication and data links communications are the only way of contact between pilots and Air-Traffic Controllers (ATCo), where the former is the most widely used and the latter is a non-spoken method mandatory for oceanic messages and limited for some domestic issues. ASR systems on ATCo environments inherit increasing complexity due to accents from non-English speakers, cockpit noise, speaker-dependent biases, and small in-domain ATC databases for training. Hereby, we introduce CleanSky EC-H2020 ATCO2, a project that aims to develop an ASR-based platform to collect, organize and automatically pre-process ATCo speech-data from air space. This paper conveys an exploratory benchmark of several state-of-the-art ASR models trained on more than 170 hours of ATCo speech-data. We demonstrate that the cross-accent flaws due to speakers' accents are minimized due to the amount of data, making the system feasible for ATC environments. The developed ASR system achieves an averaged word error rate (WER) of 7.75% across four databases. An additional 35% relative improvement in WER is achieved on one test set when training a TDNNF system with byte-pair encoding.
Mary Ellen Foster, Bart Craenen, Amol Deshmukh, Oliver Lemon, Emanuele Bastianelli, Christian Dondrup, Ioannis Papaioannou, Andrea Vanzo, Jean-Marc Odobez, Olivier Canévet, Yuanzhouhan Cao, Weipeng He, Angel Martínez-González, Petr Motlicek, Rémy Siegfried, Rachid Alami, Kathleen Belhassein, Guilhem Buisan, Aurélie Clodic, Amandine Mayima, Yoan Sallami, Guillaume Sarthou, Phani-Teja Singamaneni, Jules Waldhart, Alexandre Mazel, Maxime Caniot, Marketta Niemelä, Päivi Heikkilä, Hanna Lammi, Antti Tammela
In the EU-funded MuMMER project, we have developed a social robot designed to interact naturally and flexibly with users in public spaces such as a shopping mall. We present the latest version of the robot system developed during the project. This system encompasses audio-visual sensing, social signal processing, conversational interaction, perspective taking, geometric reasoning, and motion planning. It successfully combines all these components in an overarching framework using the Robot Operating System (ROS) and has been deployed to a shopping mall in Finland interacting with customers. In this paper, we describe the system components, their interplay, and the resulting robot behaviours and scenarios provided at the shopping mall.
Martin Fajcik, Petr Motlicek, Pavel Smrz
We present Claim-Dissector: a novel latent variable model for fact-checking and analysis, which given a claim and a set of retrieved evidences jointly learns to identify: (i) the relevant evidences to the given claim, (ii) the veracity of the claim. We propose to disentangle the per-evidence relevance probability and its contribution to the final veracity probability in an interpretable way -- the final veracity probability is proportional to a linear ensemble of per-evidence relevance probabilities. In this way, the individual contributions of evidences towards the final predicted probability can be identified. In per-evidence relevance probability, our model can further distinguish whether each relevant evidence is supporting (S) or refuting (R) the claim. This allows to quantify how much the S/R probability contributes to the final verdict or to detect disagreeing evidence. Despite its interpretable nature, our system achieves results competitive with state-of-the-art on the FEVER dataset, as compared to typical two-stage system pipelines, while using significantly fewer parameters. It also sets new state-of-the-art on FAVIQ and RealFC datasets. Furthermore, our analysis shows that our model can learn fine-grained relevance cues while using coarse-grained supervision, and we demonstrate it in 2 ways. (i) We show that our model can achieve competitive sentence recall while using only paragraph-level relevance supervision. (ii) Traversing towards the finest granularity of relevance, we show that our model is capable of identifying relevance at the token level. To do this, we present a new benchmark TLR-FEVER focusing on token-level interpretability -- humans annotate tokens in relevant evidences they considered essential when making their judgment. Then we measure how similar are these annotations to the tokens our model is focusing on.
Andres Carofilis, Pradeep Rangappa, Srikanth Madikeri, Shashi Kumar, Sergio Burdisso, Jeena Prakash, Esau Villatoro-Tello, Petr Motlicek, Bidisha Sharma, Kadri Hacioglu, Shankar Venkatesan, Saurabh Vyas, Andreas Stolcke
Fine-tuning pretrained ASR models for specific domains is challenging when labeled data is scarce. But unlabeled audio and labeled data from related domains are often available. We propose an incremental semi-supervised learning pipeline that first integrates a small in-domain labeled set and an auxiliary dataset from a closely related domain, achieving a relative improvement of 4% over no auxiliary data. Filtering based on multi-model consensus or named entity recognition (NER) is then applied to select and iteratively refine pseudo-labels, showing slower performance saturation compared to random selection. Evaluated on the multi-domain Wow call center and Fisher English corpora, it outperforms single-step fine-tuning. Consensus-based filtering outperforms other methods, providing up to 22.3% relative improvement on Wow and 24.8% on Fisher over single-step fine-tuning with random selection. NER is the second-best filter, providing competitive performance at a lower computational cost.
Pradeep Rangappa, Andres Carofilis, Jeena Prakash, Shashi Kumar, Sergio Burdisso, Srikanth Madikeri, Esau Villatoro-Tello, Bidisha Sharma, Petr Motlicek, Kadri Hacioglu, Shankar Venkatesan, Saurabh Vyas, Andreas Stolcke
Fine-tuning pretrained ASR models for specific domains is challenging for small organizations with limited labeled data and computational resources. Here, we explore different data selection pipelines and propose a robust approach that improves ASR adaptation by filtering pseudo-labels generated using Whisper (encoder-decoder) and Zipformer (transducer) models. Our approach integrates multiple selection strategies -- including word error rate (WER) prediction, named entity recognition (NER), and character error rate (CER) analysis -- to extract high-quality training segments. We evaluate our method on Whisper and Zipformer using a 7500-hour baseline, comparing it to a CER-based approach relying on hypotheses from three ASR systems. Fine-tuning on 7500 hours of pseudo-labeled call center data achieves 12.3% WER, while our filtering reduces the dataset to 100 hours (1.4%) with similar performance; a similar trend is observed on Fisher English.
Hasindri Watawana, Sergio Burdisso, Diego A. Moreno-Galván, Fernando Sánchez-Vega, A. Pastor López-Monroy, Petr Motlicek, Esaú Villatoro-Tello
Automatic depression detection from doctor-patient conversations has gained momentum thanks to the availability of public corpora and advances in language modeling. However, interpretability remains limited: strong performance is often reported without revealing what drives predictions. We analyze three datasets: ANDROIDS, DAIC-WOZ, E-DAIC and identify a systematic bias from interviewer prompts in semi-structured interviews. Models trained on interviewer turns exploit fixed prompts and positions to distinguish depressed from control subjects, often achieving high classification scores without using participant language. Restricting models to participant utterances distributes decision evidence more broadly and reflects genuine linguistic cues. While semi-structured protocols ensure consistency, including interviewer prompts inflates performance by leveraging script artifacts. Our results highlight a cross-dataset, architecture-agnostic bias and emphasize the need for analyses that localize decision evidence by time and speaker to ensure models learn from participants' language.
Andrés Carofilis, Sergio Burdisso, Esaú Villatoro-Tello, Shashi Kumar, Kadri Hacioglu, Srikanth Madikeri, Pradeep Rangappa, Manjunath K E, Petr Motlicek, Shankar Venkatesan, Andreas Stolcke
Adapting large language model (LLM)-based automatic speech recognition (ASR) systems to new domains using text-only data is a significant yet underexplored challenge. Standard fine-tuning of the LLM on the target domain text often disrupts the critical alignment between the speech and text modality learned by the projector, degrading performance. We introduce a novel text-only adaptation method that frames this process as a text denoising task. Our approach trains the LLM to recover clean transcripts from noisy inputs. This process effectively adapts the model to a target domain while preserving cross-modal alignment. Our solution is lightweight, requiring no architectural changes or additional parameters. Extensive evaluation on two datasets demonstrates up to 22.1% relative improvement, outperforming recent state-of-the-art text-only adaptation methods.
Sergio Burdisso, Séverin Baroudi, Yanis Labrak, David Grunert, Pawel Cyrta, Yiyang Chen, Srikanth Madikeri, Esaú Villatoro-Tello, Thomas Schaaf, Ricard Marxer, Petr Motlicek
We present SDialog, an MIT-licensed open-source Python toolkit that unifies dialog generation, evaluation and mechanistic interpretability into a single end-to-end framework for building and analyzing LLM-based conversational agents. Built around a standardized \texttt{Dialog} representation, SDialog provides: (1) persona-driven multi-agent simulation with composable orchestration for controlled, synthetic dialog generation, (2) comprehensive evaluation combining linguistic metrics, LLM-as-a-judge and functional correctness validators, (3) mechanistic interpretability tools for activation inspection and steering via feature ablation and induction, and (4) audio generation with full acoustic simulation including 3D room modeling and microphone effects. The toolkit integrates with all major LLM backends, enabling mixed-backend experiments under a unified API. By coupling generation, evaluation, and interpretability in a dialog-centric architecture, SDialog enables researchers to build, benchmark and understand conversational systems more systematically.
Shashi Kumar, Srikanth Madikeri, Juan Zuluaga-Gomez, Esaú Villatoro-Tello, Iuliia Thorbecke, Petr Motlicek, Manjunath K E, Aravind Ganapathiraju
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.
Thibault Bañeras-Roux, Sergio Burdisso, Esaú Villatoro-Tello, Dairazalia Sánchez-Cortés, Shiran Liu, Severin Baroudi, Shashi Kumar, Hasindri Watawana, Manjunath K E, Kadri Hacioglu, Petr Motlicek, Andreas Stolcke
Conventional end-to-end automatic speech recognition (ASR) systems rely on paired speech-text data for domain adaptation. Recent LLM-based ASR architectures connect a speech encoder to a large language model via a projection module, enabling adaptation with text-only data. However, this introduces a modality gap, as the LLM is not exposed to the noisy representations produced by the speech projector. We investigate whether small amounts of speech can mitigate this mismatch. We compare three strategies: text-only adaptation, paired speech-text adaptation, and mixed batching (MB), which combines both. Experiments in in-domain and out-of-domain settings show that even limited speech consistently improves performance. Notably, MB using only 10% of the target-domain (less than 4 hours) speech achieves word error rates comparable to, or better than, conventional ASR fine-tuning with the full dataset, indicating that small amounts of speech provide a strong modality-alignment signal.
Sergio Burdisso, Esaú Villatoro-Tello, Srikanth Madikeri, Petr Motlicek
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.
Martin Fajcik, Muskaan Singh, Juan Zuluaga-Gomez, Esaú Villatoro-Tello, Sergio Burdisso, Petr Motlicek, Pavel Smrz
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.
Subhadeep Dey, Petr Motlicek, Trung Bui, Franck Dernoncourt
In this paper, we explore various approaches for semi supervised learning in an end to end automatic speech recognition (ASR) framework. The first step in our approach involves training a seed model on the limited amount of labelled data. Additional unlabelled speech data is employed through a data selection mechanism to obtain the best hypothesized output, further used to retrain the seed model. However, uncertainties of the model may not be well captured with a single hypothesis. As opposed to this technique, we apply a dropout mechanism to capture the uncertainty by obtaining multiple hypothesized text transcripts of an speech recording. We assume that the diversity of automatically generated transcripts for an utterance will implicitly increase the reliability of the model. Finally, the data selection process is also applied on these hypothesized transcripts to reduce the uncertainty. Experiments on freely available TEDLIUM corpus and proprietary Adobe's internal dataset show that the proposed approach significantly reduces ASR errors, compared to the baseline model.
Iuliia Nigmatulina, Rudolf Braun, Juan Zuluaga-Gomez, Petr Motlicek
Automatic Speech Recognition (ASR) can be used as the assistance of speech communication between pilots and air-traffic controllers. Its application can significantly reduce the complexity of the task and increase the reliability of transmitted information. Evidently, high accuracy predictions are needed to minimize the risk of errors. Especially, high accuracy is required in recognition of key information, such as commands and callsigns, used to navigate pilots. Our results prove that the surveillance data containing callsigns can help to considerably improve the recognition of a callsign in an utterance when the weights of probable callsign n-grams are reduced per utterance. In this paper, we investigate two approaches: (1) G-boosting, when callsigns weights are adjusted at language model level (G) and followed by the dynamic decoder with an on-the-fly composition, and (2) lattice rescoring when callsign information is introduced on top of lattices generated using a conventional decoder. Boosting callsign n-grams with the combination of two methods allowed us to gain 28.4% of absolute improvement in callsign recognition accuracy and up to 74.2% of relative improvement in WER of callsign recognition.
Dairazalia Sánchez-Cortés, Sergio Burdisso, Esaú Villatoro-Tello, Petr Motlicek
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.
Shashi Kumar, Yacouba Kaloga, John Mitros, Petr Motlicek, Ina Kodrasi
Low-rank adaptation (LoRA) is a widely used method for parameter-efficient finetuning. However, existing LoRA variants lack mechanisms to explicitly disambiguate task-relevant information within the learned low-rank subspace, potentially limiting downstream performance. We propose Factorized Variational Autoencoder LoRA (FVAE-LoRA), which leverages a VAE to learn two distinct latent spaces. Our novel Evidence Lower Bound formulation explicitly promotes factorization between the latent spaces, dedicating one latent space to task-salient features and the other to residual information. Extensive experiments on text, audio, and image tasks demonstrate that FVAE-LoRA consistently outperforms standard LoRA. Moreover, spurious correlation evaluations confirm that FVAE-LoRA better isolates task-relevant signals, leading to improved robustness under distribution shifts. Our code is publicly available at: https://github.com/idiap/FVAE-LoRA
Sergio Burdisso, Srikanth Madikeri, Petr Motlicek
Efficiently deriving structured workflows from unannotated dialogs remains an underexplored and formidable challenge in computational linguistics. Automating this process could significantly accelerate the manual design of workflows in new domains and enable the grounding of large language models in domain-specific flowcharts, enhancing transparency and controllability. In this paper, we introduce Dialog2Flow (D2F) embeddings, which differ from conventional sentence embeddings by mapping utterances to a latent space where they are grouped according to their communicative and informative functions (i.e., the actions they represent). D2F allows for modeling dialogs as continuous trajectories in a latent space with distinct action-related regions. By clustering D2F embeddings, the latent space is quantized, and dialogs can be converted into sequences of region/action IDs, facilitating the extraction of the underlying workflow. To pre-train D2F, we build a comprehensive dataset by unifying twenty task-oriented dialog datasets with normalized per-turn action annotations. We also introduce a novel soft contrastive loss that leverages the semantic information of these actions to guide the representation learning process, showing superior performance compared to standard supervised contrastive loss. Evaluation against various sentence embeddings, including dialog-specific ones, demonstrates that D2F yields superior qualitative and quantitative results across diverse domains.