Audio visual automatic speech recognition in vehicles

Acoustically, car cabins are extremely noisy and as a consequence, existing audio-only speech recognition systems, for voice-based control of vehicle functions such as the GPS based navigator, perform poorly. Audio-only speech recognition systems fail to make use of the visual modality of speech (eg: lip movements). As the visual modality is immune to acoustic noise, utilising this visual information in conjunction with an audio only speech recognition system has the potential to improve the accuracy of the system. The field of recognising speech using both auditory and visual inputs is known as Audio Visual Speech Recognition (AVSR). Continuous research in AVASR field has been ongoing for the past twenty-five years with notable progress being made. However, the practical deployment of AVASR systems for use in a variety of real-world applications has not yet emerged. The main reason is due to most research to date neglecting to address variabilities in the visual domain such as illumination and viewpoint in the design of the visual front-end of the AVSR system. In this paper we present an AVASR system in a real-world car environment using the AVICAR database [1], which is publicly available in-car database and we show that the use of visual speech conjunction with the audio modality is a better approach to improve the robustness and effectiveness of voice-only recognition systems in car cabin environments.

Improving visual noise insensitivity in small vocabulary audio visual speech recognition applications

Visual noise insensitivity is important to audio visual speech recognition (AVSR). Visual noise can take on a number of forms such as varying frame rate, occlusion, lighting or speaker variabilities. The use of a high dimensional secondary classifier on the word likelihood scores from both the audio and video modalities is investigated for the purposes of adaptive fusion. Preliminary results are presented demonstrating performance above the catastrophic fusion boundary for our confidence measure irrespective of the type of visual noise presented to it. Our experiments were restricted to small vocabulary applications.

Target Detection and Tracking With Heterogeneous Sensors

We present a multimodal detection and tracking algorithm for sensors composed of a camera mounted between two microphones. Target localization is performed on color-based change detection in the video modality and on time difference of arrival (TDOA) estimation between the two microphones in the audio modality. The TDOA is computed by multiband generalized cross correlation (GCC) analysis. The estimated directions of arrival are then postprocessed using a Riccati Kalman filter. The visual and audio estimates are finally integrated, at the likelihood level, into a particle filter (PF) that uses a zero-order motion model, and a weighted probabilistic data association (WPDA) scheme. We demonstrate that the Kalman filtering (KF) improves the accuracy of the audio source localization and that the WPDA helps to enhance the tracking performance of sensor fusion in reverberant scenarios. The combination of multiband GCC, KF, and WPDA within the particle filtering framework improves the performance of the algorithm in noisy scenarios. We also show how the proposed audiovisual tracker summarizes the observed scene by generating metadata that can be transmitted to other network nodes instead of transmitting the raw images and can be used for very low bit rate communication. Moreover, the generated metadata can also be used to detect and monitor events of interest.

Multimodal Biometric Human Recognition for Perceptual Human-Computer Interaction

In this paper, a novel video-based multimodal biometric verification scheme using the subspace-based low-level feature fusion of face and speech is developed for specific speaker recognition for perceptual human--computer interaction (HCI). In the proposed scheme, human face is tracked and face pose is estimated to weight the detected facelike regions in successive frames, where ill-posed faces and false-positive detections are assigned with lower credit to enhance the accuracy. In the audio modality, mel-frequency cepstral coefficients are extracted for voice-based biometric verification. In the fusion step, features from both modalities are projected into nonlinear Laplacian Eigenmap subspace for multimodal speaker recognition and combined at low level. The proposed approach is tested on the video database of ten human subjects, and the results show that the proposed scheme can attain better accuracy in comparison with the conventional multimodal fusion using latent semantic analysis as well as the single-modality verifications. The experiment on MATLAB shows the potential of the proposed scheme to attain the real-time performance for perceptual HCI applications.

Lip detection for audio-visual speech recognition in-car environment

Acoustically, car cabins are extremely noisy and as a consequence audio-only, in-car voice recognition systems perform poorly. As the visual modality is immune to acoustic noise, using the visual lip information from the driver is seen as a viable strategy in circumventing this problem by using audio visual automatic speech recognition (AVASR). However, implementing AVASR requires a system being able to accurately locate and track the drivers face and lip area in real-time. In this paper we present such an approach using the Viola-Jones algorithm. Using the AVICAR [1] in-car database, we show that the Viola- Jones approach is a suitable method of locating and tracking the driver’s lips despite the visual variability of illumination and head pose for audio-visual speech recognition system.

Recognising audio-visual speech in vehicles using the AVICAR database

Interacting with technology within a vehicle environment using a voice interface can greatly reduce the effects of driver distraction. Most current approaches to this problem only utilise the audio signal, making them susceptible to acoustic noise. An obvious approach to circumvent this is to use the visual modality in addition. However, capturing, storing and distributing audio-visual data in a vehicle environment is very costly and difficult. One current dataset available for such research is the AVICAR [1] database. Unfortunately this database is largely unusable due to timing mismatch between the two streams and in addition, no protocol is available. We have overcome this problem by re-synchronising the streams on the phone-number portion of the dataset and established a protocol for further research. This paper presents the first audio-visual results on this dataset for speaker-independent speech recognition. We hope this will serve as a catalyst for future research in this area.

Can audio-visual speech recognition outperform acoustically enhanced speech recognition in automotive environment?

The use of visual features in the form of lip movements to improve the performance of acoustic speech recognition has been shown to work well, particularly in noisy acoustic conditions. However, whether this technique can outperform speech recognition incorporating well-known acoustic enhancement techniques, such as spectral subtraction, or multi-channel beamforming is not known. This is an important question to be answered especially in an automotive environment, for the design of an efficient human-vehicle computer interface. We perform a variety of speech recognition experiments on a challenging automotive speech dataset and results show that synchronous HMM-based audio-visual fusion can outperform traditional single as well as multi-channel acoustic speech enhancement techniques. We also show that further improvement in recognition performance can be obtained by fusing speech-enhanced audio with the visual modality, demonstrating the complementary nature of the two robust speech recognition approaches.

The impact of interface modality on police officers' visual behaviour when using an in-vehicle system

Background Standard operating procedures state that police officers should not drive while interacting with their mobile data terminal (MDT) which provides in-vehicle information essential to police work. Such interactions do however occur in practice and represent a potential source of driver distraction. The MDT comprises visual output with manual input via touch screen and keyboard. This study investigated the potential for alternative input and output methods to mitigate driver distraction with specific focus on eye movements. Method Nineteen experienced drivers of police vehicles (one female) from the NSW Police Force completed four simulated urban drives. Three drives included a concurrent secondary task: imitation licence plate search using an emulated MDT. Three different interface methods were examined: Visual-Manual, Visual-Voice, and Audio-Voice (“Visual” and “Audio” = output modality; “Manual” and “Voice” = input modality). During each drive, eye movements were recorded using FaceLAB™ (Seeing Machines Ltd, Canberra, ACT). Gaze direction and glances on the MDT were assessed. Results The Visual-Voice and Visual-Manual interfaces resulted in a significantly greater number of glances towards the MDT than Audio-Voice or Baseline. The Visual-Manual and Visual-Voice interfaces resulted in significantly more glances to the display than Audio-Voice or Baseline. For longer duration glances (>2s and 1-2s) the Visual-Manual interface resulted in significantly more fixations than Baseline or Audio-Voice. The short duration glances (<1s) were significantly greater for both Visual-Voice and Visual-Manual compared with Baseline and Audio-Voice. There were no significant differences between Baseline and Audio-Voice. Conclusion An Audio-Voice interface has the greatest potential to decrease visual distraction to police drivers. However, it is acknowledged that an audio output may have limitations for information presentation compared with visual output. The Visual-Voice interface offers an environment where the capacity to present information is sustained, whilst distraction to the driver is reduced (compared to Visual-Manual) by enabling adaptation of fixation behaviour.

Acoustic adaptation in cross database audio visual SHMM training for phonetic spoken term detection

Visual information in the form of lip movements of the speaker has been shown to improve the performance of speech recognition and search applications. In our previous work, we proposed cross database training of synchronous hidden Markov models (SHMMs) to make use of external large and publicly available audio databases in addition to the relatively small given audio visual database. In this work, the cross database training approach is improved by performing an additional audio adaptation step, which enables audio visual SHMMs to benefit from audio observations of the external audio models before adding visual modality to them. The proposed approach outperforms the baseline cross database training approach in clean and noisy environments in terms of phone recognition accuracy as well as spoken term detection (STD) accuracy.

Cross database training of audio-visual hidden Markov models for phone recognition

Speech recognition can be improved by using visual information in the form of lip movements of the speaker in addition to audio information. To date, state-of-the-art techniques for audio-visual speech recognition continue to use audio and visual data of the same database for training their models. In this paper, we present a new approach to make use of one modality of an external dataset in addition to a given audio-visual dataset. By so doing, it is possible to create more powerful models from other extensive audio-only databases and adapt them on our comparatively smaller multi-stream databases. Results show that the presented approach outperforms the widely adopted synchronous hidden Markov models (HMM) trained jointly on audio and visual data of a given audio-visual database for phone recognition by 29% relative. It also outperforms the external audio models trained on extensive external audio datasets and also internal audio models by 5.5% and 46% relative respectively. We also show that the proposed approach is beneficial in noisy environments where the audio source is affected by the environmental noise.

Robust Audio-Visual Speech Recognition under Noisy Audio-Video Conditions

This paper presents the maximum weighted stream posterior (MWSP) model as a robust and efficient stream integration method for audio-visual speech recognition in environments, where the audio or video streams may be subjected to unknown and time-varying corruption. A significant advantage of MWSP is that it does not require any specific measurements of the signal in either stream to calculate appropriate stream weights during recognition, and as such it is modality-independent. This also means that MWSP complements and can be used alongside many of the other approaches that have been proposed in the literature for this problem. For evaluation we used the large XM2VTS database for speaker-independent audio-visual speech recognition. The extensive tests include both clean and corrupted utterances with corruption added in either/both the video and audio streams using a variety of types (e.g., MPEG-4 video compression) and levels of noise. The experiments show that this approach gives excellent performance in comparison to another well-known dynamic stream weighting approach and also compared to any fixed-weighted integration approach in both clean conditions or when noise is added to either stream. Furthermore, our experiments show that the MWSP approach dynamically selects suitable integration weights on a frame-by-frame basis according to the level of noise in the streams and also according to the naturally fluctuating relative reliability of the modalities even in clean conditions. The MWSP approach is shown to maintain robust recognition performance in all tested conditions, while requiring no prior knowledge about the type or level of noise.

N1 enhancement in synesthesia during visual and audio-visual perception in semantic cross-modal conflict situations: an ERP study

Synesthesia entails a special kind of sensory perception, where stimulation in one sensory modality leads to an internally generated perceptual experience of another, not stimulated sensory modality. This phenomenon can be viewed as an abnormal multisensory integration process as here the synesthetic percept is aberrantly fused with the stimulated modality. Indeed, recent synesthesia research has focused on multimodal processing even outside of the specific synesthesia-inducing context and has revealed changed multimodal integration, thus suggesting perceptual alterations at a global level. Here, we focused on audio-visual processing in synesthesia using a semantic classification task in combination with visually or auditory-visually presented animated and in animated objects in an audio-visual congruent and incongruent manner. Fourteen subjects with auditory-visual and/or grapheme-color synesthesia and 14 control subjects participated in the experiment. During presentation of the stimuli, event-related potentials were recorded from 32 electrodes. The analysis of reaction times and error rates revealed no group differences with best performance for audio-visually congruent stimulation indicating the well-known multimodal facilitation effect. We found enhanced amplitude of the N1 component over occipital electrode sites for synesthetes compared to controls. The differences occurred irrespective of the experimental condition and therefore suggest a global influence on early sensory processing in synesthetes.

Robust indoor speaker recognition in a network of audio and video sensors

Situational awareness is achieved naturally by the human senses of sight and hearing in combination. Automatic scene understanding aims at replicating this human ability using microphones and cameras in cooperation. In this paper, audio and video signals are fused and integrated at different levels of semantic abstractions. We detect and track a speaker who is relatively unconstrained, i.e., free to move indoors within an area larger than the comparable reported work, which is usually limited to round table meetings. The system is relatively simple: consisting of just 4 microphone pairs and a single camera. Results show that the overall multimodal tracker is more reliable than single modality systems, tolerating large occlusions and cross-talk. System evaluation is performed on both single and multi-modality tracking. The performance improvement given by the audio–video integration and fusion is quantified in terms of tracking precision and accuracy as well as speaker diarisation error rate and precision–recall (recognition). Improvements vs. the closest works are evaluated: 56% sound source localisation computational cost over an audio only system, 8% speaker diarisation error rate over an audio only speaker recognition unit and 36% on the precision–recall metric over an audio–video dominant speaker recognition method.

Audio-Visual Information Processing across different age groups

People possess different sensory modalities to detect, interpret, and efficiently act upon various events in a complex and dynamic environment (Fetsch, DeAngelis, & Angelaki, 2013). Much empirical work has been done to understand the interplay of modalities (e.g. audio-visual interactions, see Calvert, Spence, & Stein, 2004). On the one hand, integration of multimodal input as a functional principle of the brain enables the versatile and coherent perception of the environment (Lewkowicz & Ghazanfar, 2009). On the other hand, sensory integration does not necessarily mean that input from modalities is always weighted equally (Ernst, 2008). Rather, when two or more modalities are stimulated concurrently, one often finds one modality dominating over another. Study 1 and 2 of the dissertation addressed the developmental trajectory of sensory dominance. In both studies, 6-year-olds, 9-year-olds, and adults were tested in order to examine sensory (audio-visual) dominance across different age groups. In Study 3, sensory dominance was put into an applied context by examining verbal and visual overshadowing effects among 4- to 6-year olds performing a face recognition task. The results of Study 1 and Study 2 support default auditory dominance in young children as proposed by Napolitano and Sloutsky (2004) that persists up to 6 years of age. For 9-year-olds, results on privileged modality processing were inconsistent. Whereas visual dominance was revealed in Study 1, privileged auditory processing was revealed in Study 2. Among adults, a visual dominance was observed in Study 1, which has also been demonstrated in preceding studies (see Spence, Parise, & Chen, 2012). No sensory dominance was revealed in Study 2 for adults. Potential explanations are discussed. Study 3 referred to verbal and visual overshadowing effects in 4- to 6-year-olds. The aim was to examine whether verbalization (i.e., verbally describing a previously seen face), or visualization (i.e., drawing the seen face) might affect later face recognition. No effect of visualization on recognition accuracy was revealed. As opposed to a verbal overshadowing effect, a verbal facilitation effect occurred. Moreover, verbal intelligence was a significant predictor for recognition accuracy in the verbalization group but not in the control group. This suggests that strengthening verbal intelligence in children can pay off in non-verbal domains as well, which might have educational implications.

Effects of modality, urgency and situation on responses to multimodal warnings for drivers

Signifying road-related events with warnings can be highly beneficial, especially when imminent attention is needed. This thesis describes how modality, urgency and situation can influence driver responses to multimodal displays used as warnings. These displays utilise all combinations of audio, visual and tactile modalities, reflecting different urgency levels. In this way, a new rich set of cues is designed, conveying information multimodally, to enhance reactions during driving, which is a highly visual task. The importance of the signified events to driving is reflected in the warnings, and safety-critical or non-critical situations are communicated through the cues. Novel warning designs are considered, using both abstract displays, with no semantic association to the signified event, and language-based ones, using speech. These two cue designs are compared, to discover their strengths and weaknesses as car alerts. The situations in which the new cues are delivered are varied, by simulating both critical and non-critical events and both manual and autonomous car scenarios. A novel set of guidelines for using multimodal driver displays is finally provided, considering the modalities utilised, the urgency signified, and the situation simulated.