Ville Hautamäki

Stephanie Milani, Anssi Kanervisto, Karolis Ramanauskas et al. · berkeley

To facilitate research in the direction of fine-tuning foundation models from human feedback, we held the MineRL BASALT Competition on Fine-Tuning from Human Feedback at NeurIPS 2022. The BASALT challenge asks teams to compete to develop algorithms to solve tasks with hard-to-specify reward functions in Minecraft. Through this competition, we aimed to promote the development of algorithms that use human feedback as channels to learn the desired behavior. We describe the competition and provide an overview of the top solutions. We conclude by discussing the impact of the competition and future directions for improvement.

Anssi Kanervisto, Tomi Kinnunen, Ville Hautamäki

Playing games with cheaters is not fun, and in a multi-billion-dollar video game industry with hundreds of millions of players, game developers aim to improve the security and, consequently, the user experience of their games by preventing cheating. Both traditional software-based methods and statistical systems have been successful in protecting against cheating, but recent advances in the automatic generation of content, such as images or speech, threaten the video game industry; they could be used to generate artificial gameplay indistinguishable from that of legitimate human players. To better understand this threat, we begin by reviewing the current state of multiplayer video game cheating, and then proceed to build a proof-of-concept method, GAN-Aimbot. By gathering data from various players in a first-person shooter game we show that the method improves players' performance while remaining hidden from automatic and manual protection mechanisms. By sharing this work we hope to raise awareness on this issue and encourage further research into protecting the gaming communities.

Federico Malato, Florian Leopold, Amogh Raut et al.

Our aim is to build autonomous agents that can solve tasks in environments like Minecraft. To do so, we used an imitation learning-based approach. We formulate our control problem as a search problem over a dataset of experts' demonstrations, where the agent copies actions from a similar demonstration trajectory of image-action pairs. We perform a proximity search over the BASALT MineRL-dataset in the latent representation of a Video PreTraining model. The agent copies the actions from the expert trajectory as long as the distance between the state representations of the agent and the selected expert trajectory from the dataset do not diverge. Then the proximity search is repeated. Our approach can effectively recover meaningful demonstration trajectories and show human-like behavior of an agent in the Minecraft environment.

Federico Malato, Florian Leopold, Ville Hautamaki et al.

Behavioural cloning uses a dataset of demonstrations to learn a behavioural policy. To overcome various learning and policy adaptation problems, we propose to use latent space to index a demonstration dataset, instantly access similar relevant experiences, and copy behavior from these situations. Actions from a selected similar situation can be performed by the agent until representations of the agent's current situation and the selected experience diverge in the latent space. Thus, we formulate our control problem as a search problem over a dataset of experts' demonstrations. We test our approach on BASALT MineRL-dataset in the latent representation of a Video PreTraining model. We compare our model to state-of-the-art Minecraft agents. Our approach can effectively recover meaningful demonstrations and show human-like behavior of an agent in the Minecraft environment in a wide variety of scenarios. Experimental results reveal that performance of our search-based approach is comparable to trained models, while allowing zero-shot task adaptation by changing the demonstration examples.

Trung Ngo, Najwa Laabid, Ville Hautamäki et al.

Paradoxically, a Variational Autoencoder (VAE) could be pushed in two opposite directions, utilizing powerful decoder model for generating realistic images but collapsing the learned representation, or increasing regularization coefficient for disentangling representation but ultimately generating blurry examples. Existing methods narrow the issues to the rate-distortion trade-off between compression and reconstruction. We argue that a good reconstruction model does learn high capacity latents that encode more details, however, its use is hindered by two major issues: the prior is random noise which is completely detached from the posterior and allow no controllability in the generation; mean-field variational inference doesn't enforce hierarchy structure which makes the task of recombining those units into plausible novel output infeasible. As a result, we develop a system that learns a hierarchy of disentangled representation together with a mechanism for recombining the learned representation for generalization. This is achieved by introducing a minimal amount of inductive bias to learn controllable prior for the VAE. The idea is supported by here developed transitive information theory, that is, the mutual information between two target variables could alternately be maximized through the mutual information to the third variable, thus bypassing the rate-distortion bottleneck in VAE design. In particular, we show that our model, named SemafoVAE (inspired by the similar concept in computer science), could generate high-quality examples in a controllable manner, perform smooth traversals of the disentangled factors and intervention at a different level of representation hierarchy.

Marko Tuononen, Dani Korpi, Ville Hautamäki

We propose a novel method for interpreting neural networks, focusing on convolutional neural network-based receiver model. The method identifies which unit or units of the model contain most (or least) information about the channel parameter(s) of the interest, providing insights at both global and local levels -- with global explanations aggregating local ones. Experiments on link-level simulations demonstrate the method's effectiveness in identifying units that contribute most (and least) to signal-to-noise ratio processing. Although we focus on a radio receiver model, the method generalizes to other neural network architectures and applications, offering robust estimation even in high-dimensional settings.

Anssi Kanervisto, Tomi Kinnunen, Ville Hautamäki

Behavioural characterizations (BCs) of decision-making agents, or their policies, are used to study outcomes of training algorithms and as part of the algorithms themselves to encourage unique policies, match expert policy or restrict changes to policy per update. However, previously presented solutions are not applicable in general, either due to lack of expressive power, computational constraint or constraints on the policy or environment. Furthermore, many BCs rely on the actions of policies. We discuss and demonstrate how these BCs can be misleading, especially in stochastic environments, and propose a novel solution based on what states policies visit. We run experiments to evaluate the quality of the proposed BC against baselines and evaluate their use in studying training algorithms, novelty search and trust-region policy optimization. The code is available at https://github.com/miffyli/policy-supervectors.

Ville Vestman, Bilal Soomro, Anssi Kanervisto et al.

The popularization of science can often be disregarded by scientists as it may be challenging to put highly sophisticated research into words that general public can understand. This work aims to help presenting speaker recognition research to public by proposing a publicly appealing concept for showcasing recognition systems. We leverage data from YouTube and use it in a large-scale voice search web application that finds the celebrity voices that best match to the user's voice. The concept was tested in a public event as well as "in the wild" and the received feedback was mostly positive. The i-vector based speaker identification back end was found to be fast (665 ms per request) and had a high identification accuracy (93 %) for the YouTube target speakers. To help other researchers to develop the idea further, we share the source codes of the web platform used for the demo at https://github.com/bilalsoomro/speech-demo-platform.

Anssi Kanervisto, Ville Hautamäki

We present Toribash Learning Environment (ToriLLE), a learning environment for machine learning agents based on the video game Toribash. Toribash is a MuJoCo-like environment of two humanoid character fighting each other hand-to-hand, controlled by changing actuation modes of the joints. Competitive nature of Toribash as well its focused domain provide a platform for evaluating self-play methods, and evaluating machine learning agents against human players. In this paper we describe the environment with ToriLLE's capabilities and limitations, and experimentally show its applicability as a learning environment. The source code of the environment and conducted experiments can be found at https://github.com/Miffyli/ToriLLE.

Federco Malato, Florian Leopold, Andrew Melnik et al.

Behavioral cloning uses a dataset of demonstrations to learn a policy. To overcome computationally expensive training procedures and address the policy adaptation problem, we propose to use latent spaces of pre-trained foundation models to index a demonstration dataset, instantly access similar relevant experiences, and copy behavior from these situations. Actions from a selected similar situation can be performed by the agent until representations of the agent's current situation and the selected experience diverge in the latent space. Thus, we formulate our control problem as a dynamic search problem over a dataset of experts' demonstrations. We test our approach on BASALT MineRL-dataset in the latent representation of a Video Pre-Training model. We compare our model to state-of-the-art, Imitation Learning-based Minecraft agents. Our approach can effectively recover meaningful demonstrations and show human-like behavior of an agent in the Minecraft environment in a wide variety of scenarios. Experimental results reveal that performance of our search-based approach clearly wins in terms of accuracy and perceptual evaluation over learning-based models.

Yusuke Mikami, Andrew Melnik, Jun Miura et al.

We demonstrate experimental results with LLMs that address robotics task planning problems. Recently, LLMs have been applied in robotics task planning, particularly using a code generation approach that converts complex high-level instructions into mid-level policy codes. In contrast, our approach acquires text descriptions of the task and scene objects, then formulates task planning through natural language reasoning, and outputs coordinate level control commands, thus reducing the necessity for intermediate representation code as policies with pre-defined APIs. Our approach is evaluated on a multi-modal prompt simulation benchmark, demonstrating that our prompt engineering experiments with natural language reasoning significantly enhance success rates compared to its absence. Furthermore, our approach illustrates the potential for natural language descriptions to transfer robotics skills from known tasks to previously unseen tasks. The project website: https://natural-language-as-policies.github.io/

Ivan Kukanov, Janne Laakkonen, Tomi Kinnunen et al.

Current speech deepfake detection approaches perform satisfactorily against known adversaries; however, generalization to unseen attacks remains an open challenge. The proliferation of speech deepfakes on social media underscores the need for systems that can generalize to unseen attacks not observed during training. We address this problem from the perspective of meta-learning, aiming to learn attack-invariant features to adapt to unseen attacks with very few samples available. This approach is promising since generating of a high-scale training dataset is often expensive or infeasible. Our experiments demonstrated an improvement in the Equal Error Rate (EER) from 21.67% to 10.42% on the InTheWild dataset, using just 96 samples from the unseen dataset. Continuous few-shot adaptation ensures that the system remains up-to-date.

Janne Laakkonen, Ivan Kukanov, Ville Hautamäki

Reliable detection of speech deepfakes (spoofs) must remain effective when the distribution of spoofing attacks shifts. We frame the task as domain generalization and show that inserting Low-Rank Adaptation (LoRA) adapters into every attention head of a self-supervised (SSL) backbone, then training only those adapters with Meta-Learning Domain Generalization (MLDG), yields strong zero-shot performance. The resulting model updates about 3.6 million parameters, roughly 1.1% of the 318 million updated in full fine-tuning, yet surpasses a fully fine-tuned counterpart on five of six evaluation corpora. A first-order MLDG loop encourages the adapters to focus on cues that persist across attack types, lowering the average EER from 8.84% for the fully fine-tuned model to 5.30% with our best MLDG-LoRA configuration. Our findings show that combining meta-learning with parameter-efficient adaptation offers an effective method for zero-shot, distribution-shift-aware speech deepfake detection.

Federico Malato, Ville Hautamaki

Sample inefficiency is a long-lasting challenge in deep reinforcement learning (DRL). Despite dramatic improvements have been made, the problem is far from being solved and is especially challenging in environments with sparse or delayed rewards. In our work, we propose to use Adversarial Estimates as a new, simple and efficient approach to mitigate this problem for a class of feedback-based DRL algorithms. Our approach leverages latent similarity search from a small set of human-collected trajectories to boost learning, using only five minutes of human-recorded experience. The results of our study show algorithms trained with Adversarial Estimates converge faster than their original version. Moreover, we discuss how our approach could enable learning in feedback-based algorithms in extreme scenarios with very sparse rewards.

Federico Malato, Ville Hautamäki

World Models have vastly permeated the field of Reinforcement Learning. Their ability to model the transition dynamics of an environment have greatly improved sample efficiency in online RL. Among them, the most notorious example is Dreamer, a model that learns to act in a diverse set of image-based environments. In this paper, we leverage similarity search and stochastic representations to approximate a world model without a training procedure. We establish a comparison with PlaNet, a well-established world model of the Dreamer family. We evaluate the models on the quality of latent reconstruction and on the perceived similarity of the reconstructed image, on both next-step and long horizon dynamics prediction. The results of our study demonstrate that a search-based world model is comparable to a training based one in both cases. Notably, our model show stronger performance in long-horizon prediction with respect to the baseline on a range of visually different environments.

Marko Tuononen, Heikki Penttinen, Ville Hautamäki

We present the first use of influence functions for deep learning-based wireless receivers. Applied to DeepRx, a fully convolutional receiver, influence analysis reveals which training samples drive bit predictions, enabling targeted fine-tuning of poorly performing cases. We show that loss-relative influence with capacity-like binary cross-entropy loss and first-order updates on beneficial samples most consistently improves bit error rate toward genie-aided performance, outperforming random fine-tuning in single-target scenarios. Multi-target adaptation proved less effective, underscoring open challenges. Beyond experiments, we connect influence to self-influence corrections and propose a second-order, influence-aligned update strategy. Our results establish influence functions as both an interpretability tool and a basis for efficient receiver adaptation.

Janne Laakkonen, Ivan Kukanov, Ville Hautamäki

Foundation models such as Wav2Vec2 excel at representation learning in speech tasks, including audio deepfake detection. However, after being fine-tuned on a fixed set of bonafide and spoofed audio clips, they often fail to generalize to novel deepfake methods not represented in training. To address this, we propose a mixture-of-LoRA-experts approach that integrates multiple low-rank adapters (LoRA) into the model's attention layers. A routing mechanism selectively activates specialized experts, enhancing adaptability to evolving deepfake attacks. Experimental results show that our method outperforms standard fine-tuning in both in-domain and out-of-domain scenarios, reducing equal error rates relative to baseline models. Notably, our best MoE-LoRA model lowers the average out-of-domain EER from 8.55\% to 6.08\%, demonstrating its effectiveness in achieving generalizable audio deepfake detection.

Federico Malato, Ville Hautamaki

Imitation learning enables autonomous agents to learn from human examples, without the need for a reward signal. Still, if the provided dataset does not encapsulate the task correctly, or when the task is too complex to be modeled, such agents fail to reproduce the expert policy. We propose to recover from these failures through online adaptation. Our approach combines the action proposal coming from a pre-trained policy with relevant experience recorded by an expert. The combination results in an adapted action that closely follows the expert. Our experiments show that an adapted agent performs better than its pure imitation learning counterpart. Notably, adapted agents can achieve reasonable performance even when the base, non-adapted policy catastrophically fails.

Anssi Kanervisto, Ville Hautamäki, Tomi Kinnunen et al.

As automatic speaker verification (ASV) systems are vulnerable to spoofing attacks, they are typically used in conjunction with spoofing countermeasure (CM) systems to improve security. For example, the CM can first determine whether the input is human speech, then the ASV can determine whether this speech matches the speaker's identity. The performance of such a tandem system can be measured with a tandem detection cost function (t-DCF). However, ASV and CM systems are usually trained separately, using different metrics and data, which does not optimize their combined performance. In this work, we propose to optimize the tandem system directly by creating a differentiable version of t-DCF and employing techniques from reinforcement learning. The results indicate that these approaches offer better outcomes than finetuning, with our method providing a 20% relative improvement in the t-DCF in the ASVSpoof19 dataset in a constrained setting.

Federico Malato, Joona Jehkonen, Ville Hautamäki

Behavioural cloning has been extensively used to train agents and is recognized as a fast and solid approach to teach general behaviours based on expert trajectories. Such method follows the supervised learning paradigm and it strongly depends on the distribution of the data. In our paper, we show how combining behavioural cloning with human-in-the-loop training solves some of its flaws and provides an agent task-specific corrections to overcome tricky situations while speeding up the training time and lowering the required resources. To do this, we introduce a novel approach that allows an expert to take control of the agent at any moment during a simulation and provide optimal solutions to its problematic situations. Our experiments show that this approach leads to better policies both in terms of quantitative evaluation and in human-likeliness.

Yi Ma, Kong Aik Lee, Ville Hautamaki et al.

Speech enhancement aims to improve the perceptual quality of the speech signal by suppression of the background noise. However, excessive suppression may lead to speech distortion and speaker information loss, which degrades the performance of speaker embedding extraction. To alleviate this problem, we propose an end-to-end deep learning framework, dubbed PL-EESR, for robust speaker representation extraction. This framework is optimized based on the feedback of the speaker identification task and the high-level perceptual deviation between the raw speech signal and its noisy version. We conducted speaker verification tasks in both noisy and clean environment respectively to evaluate our system. Compared to the baseline, our method shows better performance in both clean and noisy environments, which means our method can not only enhance the speaker relative information but also avoid adding distortions.

Khaled Hechmi, Trung Ngo Trong, Ville Hautamaki et al.

VoxCeleb datasets are widely used in speaker recognition studies. Our work serves two purposes. First, we provide speaker age labels and (an alternative) annotation of speaker gender. Second, we demonstrate the use of this metadata by constructing age and gender recognition models with different features and classifiers. We query different celebrity databases and apply consensus rules to derive age and gender labels. We also compare the original VoxCeleb gender labels with our labels to identify records that might be mislabeled in the original VoxCeleb data. On modeling side, we design a comprehensive study of multiple features and models for recognizing gender and age. Our best system, using i-vector features, achieved an F1-score of 0.9829 for gender recognition task using logistic regression, and the lowest mean absolute error (MAE) in age regression, 9.443 years, is obtained with ridge regression. This indicates challenge in age estimation from in-the-wild style speech data.

Anssi Kanervisto, Christian Scheller, Yanick Schraner et al.

Reinforcement learning (RL) research focuses on general solutions that can be applied across different domains. This results in methods that RL practitioners can use in almost any domain. However, recent studies often lack the engineering steps ("tricks") which may be needed to effectively use RL, such as reward shaping, curriculum learning, and splitting a large task into smaller chunks. Such tricks are common, if not necessary, to achieve state-of-the-art results and win RL competitions. To ease the engineering efforts, we distill descriptions of tricks from state-of-the-art results and study how well these tricks can improve a standard deep Q-learning agent. The long-term goal of this work is to enable combining proven RL methods with domain-specific tricks by providing a unified software framework and accompanying insights in multiple domains.

Keishi Ishihara, Anssi Kanervisto, Jun Miura et al.

Autonomous driving systems need to handle complex scenarios such as lane following, avoiding collisions, taking turns, and responding to traffic signals. In recent years, approaches based on end-to-end behavioral cloning have demonstrated remarkable performance in point-to-point navigational scenarios, using a realistic simulator and standard benchmarks. Offline imitation learning is readily available, as it does not require expensive hand annotation or interaction with the target environment, but it is difficult to obtain a reliable system. In addition, existing methods have not specifically addressed the learning of reaction for traffic lights, which are a rare occurrence in the training datasets. Inspired by the previous work on multi-task learning and attention modeling, we propose a novel multi-task attention-aware network in the conditional imitation learning (CIL) framework. This does not only improve the success rate of standard benchmarks, but also the ability to react to traffic lights, which we show with standard benchmarks.

Ivan Kukanov, Janne Karttunen, Hannu Sillanpää et al.

Since the invention of cinema, the manipulated videos have existed. But generating manipulated videos that can fool the viewer has been a time-consuming endeavor. With the dramatic improvements in the deep generative modeling, generating believable looking fake videos has become a reality. In the present work, we concentrate on the so-called deepfake videos, where the source face is swapped with the targets. We argue that deepfake detection task should be viewed as a screening task, where the user, such as the video streaming platform, will screen a large number of videos daily. It is clear then that only a small fraction of the uploaded videos are deepfakes, so the detection performance needs to be measured in a cost-sensitive way. Preferably, the model parameters also need to be estimated in the same way. This is precisely what we propose here.

Abrhalei Tela, Abraham Woubie, Ville Hautamaki

In recent years, transformer models have achieved great success in natural language processing (NLP) tasks. Most of the current state-of-the-art NLP results are achieved by using monolingual transformer models, where the model is pre-trained using a single language unlabelled text corpus. Then, the model is fine-tuned to the specific downstream task. However, the cost of pre-training a new transformer model is high for most languages. In this work, we propose a cost-effective transfer learning method to adopt a strong source language model, trained from a large monolingual corpus to a low-resource language. Thus, using XLNet language model, we demonstrate competitive performance with mBERT and a pre-trained target language model on the cross-lingual sentiment (CLS) dataset and on a new sentiment analysis dataset for low-resourced language Tigrinya. With only 10k examples of the given Tigrinya sentiment analysis dataset, English XLNet has achieved 78.88% F1-Score outperforming BERT and mBERT by 10% and 7%, respectively. More interestingly, fine-tuning (English) XLNet model on the CLS dataset has promising results compared to mBERT and even outperformed mBERT for one dataset of the Japanese language.

Anssi Kanervisto, Janne Karttunen, Ville Hautamäki

MineRL 2019 competition challenged participants to train sample-efficient agents to play Minecraft, by using a dataset of human gameplay and a limit number of steps the environment. We approached this task with behavioural cloning by predicting what actions human players would take, and reached fifth place in the final ranking. Despite being a simple algorithm, we observed the performance of such an approach can vary significantly, based on when the training is stopped. In this paper, we detail our submission to the competition, run further experiments to study how performance varied over training and study how different engineering decisions affected these results.

Anssi Kanervisto, Joonas Pussinen, Ville Hautamäki

Behavioural cloning, where a computer is taught to perform a task based on demonstrations, has been successfully applied to various video games and robotics tasks, with and without reinforcement learning. This also includes end-to-end approaches, where a computer plays a video game like humans do: by looking at the image displayed on the screen, and sending keystrokes to the game. As a general approach to playing video games, this has many inviting properties: no need for specialized modifications to the game, no lengthy training sessions and the ability to re-use the same tools across different games. However, related work includes game-specific engineering to achieve the results. We take a step towards a general approach and study the general applicability of behavioural cloning on twelve video games, including six modern video games (published after 2010), by using human demonstrations as training data. Our results show that these agents cannot match humans in raw performance but do learn basic dynamics and rules. We also demonstrate how the quality of the data matters, and how recording data from humans is subject to a state-action mismatch, due to human reflexes.

Anssi Kanervisto, Christian Scheller, Ville Hautamäki

Reinforcement learning (RL) has been successful in training agents in various learning environments, including video-games. However, such work modifies and shrinks the action space from the game's original. This is to avoid trying "pointless" actions and to ease the implementation. Currently, this is mostly done based on intuition, with little systematic research supporting the design decisions. In this work, we aim to gain insight on these action space modifications by conducting extensive experiments in video-game environments. Our results show how domain-specific removal of actions and discretization of continuous actions can be crucial for successful learning. With these insights, we hope to ease the use of RL in new environments, by clarifying what action-spaces are easy to learn.

Anssi Kanervisto, Ville Hautamäki, Tomi Kinnunen et al.

The spoofing countermeasure (CM) systems in automatic speaker verification (ASV) are not typically used in isolation of each other. These systems can be combined, for example, into a cascaded system where CM produces first a decision whether the input is synthetic or bona fide speech. In case the CM decides it is a bona fide sample, then the ASV system will consider it for speaker verification. End users of the system are not interested in the performance of the individual sub-modules, but instead are interested in the performance of the combined system. Such combination can be evaluated with tandem detection cost function (t-DCF) measure, yet the individual components are trained separately from each other using their own performance metrics. In this work we study training the ASV and CM components together for a better t-DCF measure by using reinforcement learning. We demonstrate that such training procedure indeed is able to improve the performance of the combined system, and does so with more reliable results than with the standard supervised learning techniques we compare against.

Bilal Soomro, Anssi Kanervisto, Trung Ngo Trong et al.

Deep neural networks (DNN) are able to successfully process and classify speech utterances. However, understanding the reason behind a classification by DNN is difficult. One such debugging method used with image classification DNNs is activation maximization, which generates example-images that are classified as one of the classes. In this work, we evaluate applicability of this method to speech utterance classifiers as the means to understanding what DNN "listens to". We trained a classifier using the speech command corpus and then use activation maximization to pull samples from the trained model. Then we synthesize audio from features using WaveNet vocoder for subjective analysis. We measure the quality of generated samples by objective measurements and crowd-sourced human evaluations. Results show that when combined with the prior of natural speech, activation maximization can be used to generate examples of different classes. Based on these results, activation maximization can be used to start opening up the DNN black-box in speech tasks.

Abraham Woubie, Anssi Kanervisto, Janne Karttunen et al.

Mapping states to actions in deep reinforcement learning is mainly based on visual information. The commonly used approach for dealing with visual information is to extract pixels from images and use them as state representation for reinforcement learning agent. But, any vision only agent is handicapped by not being able to sense audible cues. Using hearing, animals are able to sense targets that are outside of their visual range. In this work, we propose the use of audio as complementary information to visual only in state representation. We assess the impact of such multi-modal setup in reach-the-goal tasks in ViZDoom environment. Results show that the agent improves its behavior when visual information is accompanied with audio features.

Janne Karttunen, Anssi Kanervisto, Ville Kyrki et al.

Deep reinforcement learning has proven to be successful for learning tasks in simulated environments, but applying same techniques for robots in real-world domain is more challenging, as they require hours of training. To address this, transfer learning can be used to train the policy first in a simulated environment and then transfer it to physical agent. As the simulation never matches reality perfectly, the physics, visuals and action spaces by necessity differ between these environments to some degree. In this work, we study how general video games can be directly used instead of fine-tuned simulations for the sim-to-real transfer. Especially, we study how the agent can learn the new action space autonomously, when the game actions do not match the robot actions. Our results show that the different action space can be learned by re-training only part of neural network and we obtain above 90% mean success rate in simulation and robot experiments.

Kong Aik Lee, Ville Hautamaki, Tomi Kinnunen et al.

The I4U consortium was established to facilitate a joint entry to NIST speaker recognition evaluations (SRE). The latest edition of such joint submission was in SRE 2018, in which the I4U submission was among the best-performing systems. SRE'18 also marks the 10-year anniversary of I4U consortium into NIST SRE series of evaluation. The primary objective of the current paper is to summarize the results and lessons learned based on the twelve sub-systems and their fusion submitted to SRE'18. It is also our intention to present a shared view on the advancements, progresses, and major paradigm shifts that we have witnessed as an SRE participant in the past decade from SRE'08 to SRE'18. In this regard, we have seen, among others, a paradigm shift from supervector representation to deep speaker embedding, and a switch of research challenge from channel compensation to domain adaptation.

Trung Ngo Trong, Ville Hautamäki, Kristiina Jokinen

Language recognition system is typically trained directly to optimize classification error on the target language labels, without using the external, or meta-information in the estimation of the model parameters. However labels are not independent of each other, there is a dependency enforced by, for example, the language family, which affects negatively on classification. The other external information sources (e.g. audio encoding, telephony or video speech) can also decrease classification accuracy. In this paper, we attempt to solve these issues by constructing a deep hierarchical neural network, where different levels of meta-information are encapsulated by attentive prediction units and also embedded into the training progress. The proposed method learns auxiliary tasks to obtain robust internal representation and to construct a variant of attentive units within the hierarchical model. The final result is the structural prediction of the target language and a closely related language family. The algorithm reflects a "staircase" way of learning in both its architecture and training, advancing from the fundamental audio encoding to the language family level and finally to the target language level. This process not only improves generalization but also tackles the issues of imbalanced class priors and channel variability in the deep neural network model. Our experimental findings show that the proposed architecture outperforms the state-of-the-art i-vector approaches on both small and big language corpora by a significant margin.

Rosa González Hautamäki, Anssi Kanervisto, Ville Hautamäki et al.

Voice disguise, purposeful modification of one's speaker identity with the aim of avoiding being identified as oneself, is a low-effort way to fool speaker recognition, whether performed by a human or an automatic speaker verification (ASV) system. We present an evaluation of the effectiveness of age stereotypes as a voice disguise strategy, as a follow up to our recent work where 60 native Finnish speakers attempted to sound like an elderly and like a child. In that study, we presented evidence that both ASV and human observers could easily miss the target speaker but we did not address how believable the presented vocal age stereotypes were; this study serves to fill that gap. The interesting cases would be speakers who succeed in being missed by the ASV system, and which a typical listener cannot detect as being a disguise. We carry out a perceptual test to study the quality of the disguised speech samples. The listening test was carried out both locally and with the help of Amazon's Mechanical Turk (MT) crowd-workers. A total of 91 listeners participated in the test and were instructed to estimate both the speaker's chronological and intended age. The results indicate that age estimations for the intended old and child voices for female speakers were towards the target age groups, while for male speakers, the age estimations corresponded to the direction of the target voice only for elderly voices. In the case of intended child's voice, listeners estimated the age of male speakers to be older than their chronological age for most of the speakers and not the intended target age.

Kong Aik Lee, Ville Hautamäki, Anthony Larcher et al.

This article describes the systems jointly submitted by Institute for Infocomm (I$^2$R), the Laboratoire d'Informatique de l'Université du Maine (LIUM), Nanyang Technology University (NTU) and the University of Eastern Finland (UEF) for 2015 NIST Language Recognition Evaluation (LRE). The submitted system is a fusion of nine sub-systems based on i-vectors extracted from different types of features. Given the i-vectors, several classifiers are adopted for the language detection task including support vector machines (SVM), multi-class logistic regression (MCLR), Probabilistic Linear Discriminant Analysis (PLDA) and Deep Neural Networks (DNN).