Using a mono camera, in ADAS, as the primary sensor offers a significant advantage by minimizing system complexity since no extra fusion step is considered. This approach also leverages recent advancements in computer vision and deep learning, which enable high levels of environmental understanding and scene analysis from visual data alone. As such, mono-camera setups hold promise for achieving reliable perception at scale and lead to a growing interest in monocular approaches, particularly for detecting and tracking dynamic objects. However, traditional mono-camera methods are often dependent on auxiliary inputs from GPS or maps, which may be unreliable in complex terrain or areas with poor signal coverage. In this work, we introduce an effective monocular 3D multi-object tracking approach, called MonoSORT3D, which operates without requiring additional auxiliary inputs. Evaluation of our method on the KITTI and MOT17 datasets demonstrates the competitive performance against state-of-the-art methods. Additionally, we provide an in-depth analysis of MonoSORT3D architecture by conducting an ablation study on different components within.

Collaborative perception enhances the robustness and accuracy of autonomous systems by leveraging shared perceptual data across agents, particularly through feature-level fusion, which balances communication efficiency with contextual preservation. However, this data-sharing introduces vulnerabilities, as adversaries can inject malicious perturbations, compromising system reliability in safety-critical scenarios. In this work, we address the adversarial robustness of feature-level fusion in collaborative perception under white-box untargeted attack settings. We propose a novel framework that combines adversarial training with a consensus mechanism, enhancing resilience to adversarial perturbations in a model-agnostic manner. Our approach not only improves robustness against attacks but also enhances performance on clean data, achieving at least 5% improvement in average precision. Extensive experiments on the V2XSet dataset with four adversarial attack types and two collaborative perception methods demonstrate the effectiveness of our method, outperforming consensus defense and adversarial training alone consistently under different adversarial perturbation magnitudes. These findings underscore the potential of our approach to advance secure and reliable collaborative perception systems.

Software solutions for advanced driver assistant systems should be culturally fair, in particular when perceiving and reacting to pedestrians. To ensure the generation of suitable training and testing scenarios, a detailed understanding of pedestrian behavior and cultural differences in behavior must be established. However, direct comparisons between the populations in different countries are scarce and complicated to conduct in real environments. In this work, we present the first cross-cultural pedestrian behavior study analyzing and comparing pedestrian behavior between Japanese and German pedestrians. For this purpose, a new and large-scale virtual reality (VR) environment was created and the same experiment was conducted in Germany and in Japan. We identified new insights in gap selection, choice of velocity, following behavior, and route choice. Particularly for the route choice adjacent to zebra crossings, the differences between Japanese and German pedestrians are more complex than past studies might suggest. In addition, the data captured in the VR environment can be directly utilized to train and test algorithms used in intelligent vehicles in a digital domain.

Effective and flexible allocation of visual attention is key for pedestrians who have to navigate to a desired goal under different conditions of urgency and safety preferences. While automatic modelling of pedestrian attention holds great promise to improve simulations of pedestrian behavior, current saliency prediction approaches mostly focus on generic free-viewing scenarios and do not reflect the specific challenges present in pedestrian attention prediction. In this paper, we present Context-SalNET, a novel encoder-decoder architecture that explicitly addresses three key challenges of visual attention prediction in pedestrians: First, Context-SalNET explicitly models the context factors urgency and safety preference in the latent space of the encoder-decoder model. Second, we propose the exponentially weighted mean squared error loss (ew-MSE) that is able to better cope with the fact that only a small part of the ground truth saliency maps consist of non-zero entries. Third, we explicitly model epistemic uncertainty to account for the fact that training data for pedestrian attention prediction is limited. To evaluate Context-SalNET, we recorded the first dataset of pedestrian visual attention in VR that includes explicit variation of the context factors urgency and safety preference. Context-SalNET achieves clear improvements over state-of-the-art saliency prediction approaches as well as over ablations. Our novel dataset will be made fully available and can serve as a valuable resource for further research on pedestrian attention prediction.

The prediction of pedestrian behavior remains a major objective for the development of autonomous vehicles. Pedestrians do not merely represent the most vulnerable traffic participants, but are also a challenge in the prediction process, since their behavior entails a large number of options for possible paths, velocities, and motions. In addition, autonomous vehicles should be able to operate safely in different countries, and thus the incorporation of cultural differences in the training and evaluation of the relevant AI systems is required. This paper provides the first review of Japanese and German pedestrians' behavior in urban traffic. In particular, cultural behavior differences of pedestrians in risk avoidance, compliance, gap acceptance, and walking velocity together with different environmental factors like pedestrian facilities in both countries are addressed.

The pedestrian intention is not only signalized by their past trajectory and head gaze direction, but by their whole body movement. In order to improve existing algorithms, analysis of pedestrian motions before entering a shared environment, like a street crossing, is necessary. In addition, more accurate human models are required for a digital reality in order to enable the scalable evaluation of autonomous systems using synthetic data, most of all in the area of autonomous driving. In this work, we present a first approach to capturing pedestrian locomotion in a field experiment using an existing motion capture solution. The motion of 20 participants in 80 crossing trials was recorded and numerically evaluated. The evaluation suggests, that the hip and shoulder rotation with respect to the street is more pronounced before crossing the street compared to other, non-critical behavior.

In this research we present an application for analyzing a driver's visual focus of attention in both real-life traffic and simulator setups. For this purpose we use the EyeVIUS [1] system featuring stationary eye-tracking and head-tracking functionality. In our real-life experiments, three dimensional representations of both the vehicle's interior and the outside environment are used. A real-time evaluation concerning the object in the driver's visual focus in these environments is then performed. We describe the functionality and the accuracy of our system, which is integrated in a fully functional vehicle in a real traffic environment. Considering the simulator setup, a seamless integration of our system into the OpenDS [2] driving simulator allows to automatically determine whether a user is engaged with virtual content of the simulator or focused on objects in the real world.

This paper presents an automotive interaction concept called GetHomeSafe that allows drivers to book hotels, browse news and use both Wolfram Alpha and Facebook safely during driving. GetHomeSafe includes a natural speech dialog system and a graphical user interface both adapted to drivers' needs. In addition to a description of the interaction concept, we report the results of a real-car driving study. To assess the driving safety subjective and objective data on GetHomeSafe usage were gathered and compared with the usage of respective apps on a mounted tablet. In order to combine information about eye-gaze and head-pose of the driver, we used the EyeVIUS system [1] to get the drivers' focus-of-attention. The results indicate that it seems to be safer to use GetHomeSafe compared to the apps on a mounted tablet. One reason is that the subjective load and the distraction while using the mounted tablet is higher for most of the drivers. Moreover, the GetHomeSafe interaction concept received favorable usability ratings.

What does situation-adaptive technology mean for car drivers and how can it improve their lives? Why is multimodal interaction in the cockpit a critical ingredient? This contribution summarizes several important technological results of the three-year research project SiAM, which investigated these questions. Motivated by the story of an urban commuter, we illustrate three use cases for situation adaptivity: multimodal control of car functions, cognitive load aware interaction with the environment, and a persuasive intermodal travel assistant.

This paper shows that fine-grained linguistic complexity has measurable effects on cognitive load with consequences for the design of in-car spoken dialogue systems. We used synthesized German sentences with grammatical ambiguities to test the additional workload caused by human sentence processing during driving. For the driving task, we used the Continuous Tracking and Reaction (ConTRe) task, which we believe is suitable for the measurement of the fine-grained effects of linguistically-related workload phenomena in automotive environments, as it provides millisecond-level driving deviation measurements on a continuous course. We applied the task in an eye-tracking environment, using a pupillometric measure of cognitive workload called the Index of Cognitive Activity (ICA).

Millions of drivers share the same roads every day. They share the same location but they are socially isolated and anonymous to each other. Anonymity and isolation are considered to be among the most common factors for aggressive behavior on the road (road rage).

We present results of a study on referring to the outside environment from within a moving vehicle. Reference resolution is the first necessary step in integrating the outside environment into the interactive system in the car. It is the problem of finding out which of the objects outside the users is interested in. In our study, we explored eye gaze, head pose, pointing gesture with a smart phone, and the user's view field. We implemented and tested everything in a moving vehicle in a real-life traffic. For safety reasons, the front-seat passenger used the system while the driver was concentrating completely on driving. For analysis and visualization of the user's interaction with the environment, 528 buildings of the city were modeled in 2.5D by using an airborne LIDAR scan, Google Earth, and a spatial database. As a result of our study, we propose in this paper a new algorithm for spatial reference resolution together with a scanning mechanism.

Although intelligent components in modern cars help to contribute to safe mobility, they lack important Smart Environment characteristics like personalization. Moreover, today's in-vehicle infotainment systems do not offer any interaction possibility between the passenger and the visible environment around the car. In this paper we combine several technologies and propose an approach on personalized user interaction in urban environments. We present a showcase that points out the interplay of personalized in-vehicle infotainment system and interaction with visible outside environment.

In this paper, an evaluation method using natural footstep-accurate traces as ground truth is proposed. Said ground truth was obtained by video-taping people who were naturally walking in the foyer of our company building. These video-clips were then manually analyzed, using the tiles of the foyer's floor as a coordinate system. The result of this analysis are not simply the positions of the filmed users as they progress through the foyer, but the coordinates of each single footstep of each user including timestamps for each step. To evaluate the position accuracy of an indoor positioning system, these traces were laid out on the floor and `re-walked' while carrying the hardware for the positioning system. A short sound was played according to the timestamps of each recorded step of the ground truth, to ensure that the walking speed was close to the real walking speed. The distances of these `target system traces' to the ground-truth traces were then calculated and analyzed.

The next big step in embedded, mobile speech recognition will be to allow completely free input as it is needed for messaging like SMS or email. However, unconstrained dictation remains error-prone, especially when the environment is noisy. In this paper, we compare different methods for improving a given free-text dictation system used to enter textbased messages in embedded mobile scenarios, where distraction, interaction cost, and hardware limitations enforce strict constraints over traditional scenarios. We present a corpus-based evaluation, measuring the trade-off between improvement of the word error rate versus the interaction steps that are required under various parameters. Results show that by post-processing the output of a "black box" speech recognizer (e.g. a web-based speech recognition service), a reduction of word error rate by 55% (10.3% abs.) can be obtained. For further error reduction, however, a richer representation of the original hypotheses (e.g. lattice) is necessary.

Cars have been increasingly equipped with technology, meeting the demand of people for safety, connectivity, and comfort. Upcoming HMIs provide access to in-car systems and web services in a personalized manner that facilitates a large array of functionality even while driving, with other passengers also benefiting from an enhanced experience. Such intelligent applications however depend on a solid basis to be effective: Personalization, adaptive HMI, situation-aware intelligent systems -- either of these require semantic knowledge about the user, the vehicle, the current driving situation. Advanced functions coexist with sensors, other functions, and even other vehicles. In such an environment, collaboration can be highly beneficial. Obtaining a common understanding of knowledge and providing a platform to exchange it is essential in order to reach the next level of intelligent in-car systems. This work describes the Automotive Ontology, which is located at the core of such an open platform. We give an overview of design areas relevant to automotive applications, as well as meta aspects that facilitate inference and reasoning.

In this paper, we describe a technique to determine user preferences concerning in-car micro-gesture interaction. The approach is derived from the theater technique [1], and implies a collaborative adjustment of parameters with the experimenter, until the subject has decided about the final settings. We evaluated three systematically selected gestures (zooming, sweeping, and circling) for controlling four exemplary comfort functions of the car (window lifter, air condition, radio volume, and seat heating). The main result of our study is the geometry of a "sweet spot" for micro-gesture recognition close to the steering wheel, which is independent from the underlying technical recognition approach. Additionally, preferred sizes, angles, and pause times for the investigated gestures are provided. We give an indication, which of the gestures is preferred by the users (the sweeping gesture). Finally, we provide a more detailed view on the interaction between gesture preferences and function.

A prototypical conspicuity enhancement (CE) system for vulnerable road users (here e-bikes) is described. We stress that CE is a form of multimodal output. We argue that previous CE approaches have the drawback of affecting uninvolved (road) users. We argue further that augmented reality as an alternative is error prone because objects need to be tracked. Our system implements the hybrid reality modality model, where directed information emanates from the objects themselves and therefore no object recognition/tracking is needed. We describe the components of a functional demonstrator based on standard compliant car-to-car communication components.

Automotive development has been dominated by the constraints of driving. However, natural relations to the more general area of Intelligent User Interfaces exist. Previous research in related fields therefore should be adopted and included. The aim of the 2011 MIAA workshop is to foster discussion between experts in otherwise unrelated fields of research. For example, public interfaces use crossmodal referencing in order to circumvent restrictions by the users current focus on a limited communication channel. Our aim is to raise awareness for this approach, concluding that crossmodal references in the car are helpful to bridge the gap between information inside the car and the environment. Another focus topic of the workshop is eco-friendly driving. Although universally regarded as a necessity, it remains an open question how to encourage drivers to drive ecologically. We discuss for example awards for eco-friendly driving by making it competitive and game-like.

We introduce the "Geremin" approach on in-car 2D microgesture recognition, which belongs to the category of electric field sensing techniques detecting the presence of a human hand near a conductive object (not affected by light and dynamic backgrounds, fast response times). The core component is essentially a modified "Theremin", an early electronic musical instrument named after the Russian inventor Professor Leon Theremin. Gesture recognition is done using a Dynamic Time Warp DTW algorithm. With respect to the application domain, we follow the direction of "selective mapping to theme or function" suggested in the literature. For gesture location, we propose the immediate proximity of the steering wheel, which has the advantage of providing gesture-based interaction without requiring the driver to take off hand(s). The major motivating factor for the proposed approach is reducing installation costs. Although, we use a single-antenna setup for this study, our results indicate that the gain in recognition accuracy justifies the use of two or more.

Interaction with communication and infotainment systems in the car is common while driving. Our research investigates modalities and techniques that enable interaction with interactive applications while driving without compromising safety. In this paper we present the results of an experiment where we use eye-gaze tracking in combination with a button on the steering wheel as explicit input substituting the interaction on the touch screen. This approach combines the advantages of direct interaction on visual displays without the drawbacks of touch screens. In particular the freedom of placement for the screen (even out of reach from the user) and that both hands remain on the steering wheel are the main advantages. The results show that this interaction modality is slightly slower and more distracting than a touch screen but it is significantly faster than automated speech interaction.

Many researchers argue, in assessing the benefits of Advanced Driver Assistance Systems (ADAS) it has to be taken into account that any gains in terms of security may be again reduced by the fact they affect the drivers' behavior. In this paper, we present results of a driving simulation study in which we compare driving performance as well as driver stress in three conditions: 1) when no assistance is available (NOASSIST) ; 2) when the system is working as it should (WORK); 3) when the system suddenly fails (FAIL). Results show that the driving performance is severely affected by system failures. The drivers' ability to effectively react to suddenly appearing obstacles FAIL is significantly lower than in NOASSIST. At the same time, the stress level is significantly higher in FAIL compared to WORK.

We present a novel approach to automatic speaker age classification, which combines regression and classification to achieve competitive classification accuracy on telephone speech. Support vector machine regression is used to generate finer age estimates, which are combined with the posterior probabilities of well-trained discriminative gender classifiers to predict both the age and gender of a speaker. We show that this combination performs better than direct 7-class classifiers. The regressors and classifiers are trained using longterm features such as pitch and formants, as well as short-term (frame-based) features derived from MAP adaptation of GMMs that were trained on MFCCs.

In this paper, we address the question how speech and tangible interfaces can be combined in order to provide effective multimodal interaction in vehicles, taking into account the special requirements induced by the circumstances of driving. Speech is used to set the interaction context (determine the object as is to be manipulated) and a turn-and-push dial is used to manipulate/adjust. An experimental study is presented that measures the distraction induced by manual (conventional), speech-only, and multimodal interaction (combination of speech and turnand-push dial). Results show that while subjects where able to perform more tasks in the manual condition, their driving was significantly safer while using speech-only or multimodal dialog. Supplemental contributions of this paper are descriptions of how a multimodal dialog manager as well as a driving simulation software are connected to the CAN (Controller Area Network) vehicle bus as well as how driver distraction caused by interacting with a system are measured using the standardized lane change task (LCT)

Car manufacturers are faced with a new challenge. While a new generation of “digital natives ” becomes a new customer group, the problem of aging society is still increasing. This em-phasizes the need of providing flexible in-car dialog that take into account the specific needs and preferences of the respective user (group). Along the lines of this year’s Interspeech motto “Spoken Language Processing for All”, we address the ques-tion how we find out which group the current user belongs to.

Most paralinguistic analysis tasks are lacking agreed-upon evaluation procedures and comparability, in contrast to more ‘traditional ’ disciplines in speech analysis. The INTERSPEECH 2010 Paralinguistic Challenge shall help overcome the usually low compatibility of results, by addressing three selected subchallenges. In the Age Sub-Challenge, the age of speakers has to be determined in four groups. In the Gender Sub-Challenge, a three-class classification task has to be solved and finally, the Affect Sub-Challenge asks for speakers ’ interest in ordinal representation. This paper introduces the conditions, the Challenge corpora “aGender ” and “TUM AVIC ” and standard feature sets that may be used. Further, baseline results are given.

Local danger warning is an important function of Advanced Driver Assistance Systems (ADAS) to improve the safety of driving. The user interface (the warning presentation) is particularly crucial to a successful danger avoidance. We present a user study investigating various warning presentations using a scenario of emergent road obstacles. Two presentation factors were selected: modality and level of assistance. The modality factor had 4 variants: speech warning, visual and speech warning, visual warning with blinking cue, and visual warning with sound cue. The level of assistance varied between with or without action suggestions (AS). In accordance with the ISO usability model, a total of 6 measurements were derived to assess the effectiveness and efficiency of the warnings and the drivers' satisfaction. Results indicate that the combination of speech and visual modality leads to the best performance as well as the highest satisfaction. In contrast, purely auditory and purely visual modalities were both insufficient for presenting high-priority warnings. AS generally improved the usability of the warnings especially when they were accompanied by supporting information so that drivers could validate the suggestions.

This paper summarizes the main objectives of the 2nd IUI workshop on multimodal interfaces for automotive applications (MIAA 2010).

Multilinguality represents an area of significant opportunities for automatic speech-processing systems: whereas multilingual societies are commonplace, the majority of speech-processing systems are developed with a single language in mind. As a step towards improved understanding of multilingual speech processing, the current contribution investigates how an important para-linguistic aspect of speech, namely speaker age, depends on the language spoken. In particular, we study how certain speech features affect the performance of an age recognition system for different South African languages in the Lwazi corpus. By optimizing our feature set and performing language-specific tuning, we are working towards true multilingual classifiers. As they are closely related, ASR and dialog systems are likely to benefit from an improved classification of the speaker. In a comprehensive corpus analysis on long-term features, we have identified features that exhibit characteristic behaviors for particular languages. In a follow-up regression experiment, we confirm the suitability of our feature selection for age recognition and present cross-language error rates. The mean absolute error ranges between 7.7 and 12.8 years for same-language predictors and rises to 14.5 years for cross-language predictors.

We present an experimental study on the effectiveness of five modality variants (speech, text-only, icon-only, two combinations of text and icons) for presenting local danger warnings for drivers. Hereby, we focus on sudden appearing road obstacles within a maximum up-to-date scenario as it is envisaged in Car2Car communication research. The effectiveness is measured by the minimum time necessary for fully interpreting the content. Results show that text-only requires the most time while icon only is perceived the fastest. The two combined versions lie in between. The minimum length for speech is determined by the duration of the utterance, which is longer than perception time of text-only in this case. However, speech could be decoded reliably by nearly all subjects. Results indicate further that a blinking visual cue provided through the periphery visual channel is able to enhance the saliency of visual modalities. Subjective judgements by the subjects furthermore suggest a combined use of visual and auditory modalities.

In this paper, we outline a system that supports the communication between passengers by transmitting speech (and maybe also video) of the communication partners back and forth. A study is presented that addresses the questions: 1. Is listening to noisy speech coming from the backseat really distracting the driver? Subjects are rating the truth of common-sense statements played from the back of the car (clear, noisy) while driving with a drive simulator. NOISY is rated significantly more distracting than CLEAR while objective driving performance only degrades for men but not for women.

The following article shows how a state-of-the-art speaker diarization system can be improved by combining traditional short-term features (MFCCs) with prosodic and other long-term features. First, we present a framework to study the speaker discriminability of 70 different long-term features. Then, we show how the top-ranked long-term features can be combined with short-term features to increase the accuracy of speaker diarization. The results were measured on standardized data sets (NIST RT) and show a consistent improvement of about 30% relative in diarization error rate compared to the best system presented at the NIST evaluation in 2007. This result was also verified on a wide set of meetings, which we call CombDev, that contains 21 meetings from previous evaluations. Since the prosodic and long-term features were selected using a diarization-independent speaker-discriminability study, we are confident that the same features are able to improve other systems that perform similar tasks

This paper summarizes the main objectives of the IUI workshop W2 on multimodal interfaces for automotive applications.

Speaker diarization is defined as the task of determining ldquowho spoke whenrdquo given an audio track and no other prior knowledge of any kind. The following article shows how a state-of-the-art speaker diarization system can be improved by combining traditional short-term features (MFCCs) with prosodic and other long-term features. First, we present a framework to study the speaker discriminability of 70 different long-term features. Then, we show how the top-ranked long-term features can be combined with short-term features to increase the accuracy of speaker diarization. The results were measured on standardized datasets (NIST RT) and show a consistent improvement of about 30% relative in diarization error rate compared to the best system presented at the NIST evaluation in 2007.

We present two approaches on acoustic event detection for speech-enabled car applications: a generative GMM-UBM approach and a discriminative GMM-SVM supervector approach. The systems detect whether or not a certain acoustic event occurred while the built-in microphone of the car was active to record a spoken command, either before, while, or after the driver was speaking. These events can be music playing, phone ringing, a passenger different from the driver is talking, laughing, or coughing. The task is formally defined as a detection task along the lines of well established detection tasks such as speaker recognition or language recognition. Similarly, the evaluation procedure has been designed to resemble the respective official evaluation series performed by NIST (i.e. it was a blind ’one-shot’ evaluation on a separately provided dataset). The performance of the system was calculated in terms of detection miss and false alarm probabilities (C Mi ss = C FA =1 , and P Ta rget =0 .5). The performance of the superior GMMSVM system was 0.0345 for known test speakers and 0.1955 for novel test speakers. Frequency-filtered band energy coefficients (FFBE) outperformed MFCCS on that task. The results are promising and suggest further experiments on more data.

During the past few years, speaker diarization has achieved satisfying accuracy in terms of speaker Diarization Error Rate (DER). The most successful approaches, based on agglomerative clustering, however, exhibit an inherent computational complexity which makes real-time processing, especially in combination with further processing steps, almost impossible. In this article we present a framework to speed up agglomerative clustering speaker diarization. The basic idea is to adopt a computationally cheap method to reduce the hypothesis space of the more expensive and accurate model selection via Bayesian Information Criterion (BIC). Two strategies based on the pitch-correlogram and the unscented-transform based approximation of KL-divergence are used independently as a fast-match approach to select the most likely clusters to merge. We performed the experiments using the existing ICSI speaker diarization system. The new system using KL-divergence fast-match strategy only performs 14% of total BIC comparisons needed in the baseline system, speeds up the system by 41% without affecting the speaker Diarization Error Rate (DER). The result is a robust and faster than real-time speaker diarization system.

This paper presents a comparative study of four different approaches to automatic age and gender classification using seven classes on a telephony speech task and also compares the results with human performance on the same data. The automatic approaches compared are based on (1) a parallel phone recognizer, derived from an automatic language identification system; (2) a system using dynamic Bayesian networks to combine several prosodic features; (3) a system based solely on linear prediction analysis; and (4) Gaussian mixture models based on MFCCs for separate recognition of age and gender. On average, the parallel phone recognizer performs as well as Human listeners do, while loosing performance on short utterances. The system based on prosodic features however shows very little dependence on the length of the utterance.

The most successful systems in previous comparative studies on speaker age recognition used short-term cepstral features modeled with Gaussian Mixture Models (GMMs) or applied multiple phone recognizers trained with the data of speakers of the respective class. Acoustic analyses, however, indicate that certain features such as pitch extracted from a longer span of speech correlate clearly with the speaker age although the systems based on those features have been inferior to the before mentioned approaches. In this paper, three novel systems combining short-term cepstral features and long-term features for speaker age recognition are compared to each other. A system combining GMMs using frame-based MFCCs and SupportVector-Machines using long-term pitch performs best. The results indicate that the combination of the two feature types is a promising approach, which corresponds to findings in related fields like speaker recognition.

This paper describes a system that exploits the paralinguistic information in the speech to estimate the speakers’ age and gender. Compared with previously published work, the so called AGENDER approach involves finer grained speaker classes and achieves a significantly higher classification accuracy. The introduction encompasses various application examples representing the actual AGENDER project context. Then hypotheses, method and a representative selection of results from extensive corpus analyses are presented, that build the empirical basis for the machine learning. Finally, the AGENDER approach on speaker classification is outlined, involving the comparison of different classification methods as well as evaluation results. The paper finishes with an outlook on extensions that are scheduled for the next project phase.

Navigation services can be found in different situations and contexts: while connected to the web through a desktop PC, in cars, and more recently on PDAs while on foot. These services are usually well designed for their specific purpose, but fail to work in other situations. In this paper we present an approach that connects a variety of specialized user interfaces to achieve a personal navigation service spanning different situations. We describe the concepts behind the \bf BPN (BMW Personal Navigator), an entirely implemented system that combines a desktop event and route planner, a car navigation system, and a multi-modal, in- and outdoor pedestrian navigation system for a PDA. Rather than designing for one unified UI, we focus on connecting specialized UIs for desktop, in-car and on-foot use.

In this paper we show how to exploit raw speech data to gain higher level information about the user in a mobile context. In particular we introduce an approach for the estimation of age and gender using well known machine learning techniques. On the basis of this information, systems like for example a mobile pedestrian navigation system, can be made adaptive to the special needs of a specific user group (here the elderly). First we provide a motivation why we consider such an adaptation as necessary, then we outline some adaptation strategies that are adequate for mobile assistants. The major part of the paper is about (a) identifying and extracting features of speech that are relevant for age and gender estimation and (b) classifying a particular speaker, treating uncertainty, and updating the user model over time. Finally we provide a short outlook on current work.