The use of temporal speech and lip information for multi-modal speaker identification via multi-stream HMMs

Investigates the use of temporal lip information, in conjunction with speech information, for robust, text-dependent speaker identification. We propose that significant speaker-dependent information can be obtained from moving lips, enabling speaker recognition systems to be highly robust in the presence of noise. The fusion structure for the audio and visual information is based around the use of multi-stream hidden Markov models (MSHMM), with audio and visual features forming two independent data streams. Recent work with multi-modal MSHMMs has been performed successfully for the task of speech recognition. The use of temporal lip information for speaker identification has been performed previously (T.J. Wark et al., 1998), however this has been restricted to output fusion via single-stream HMMs. We present an extension to this previous work, and show that a MSHMM is a valid structure for multi-modal speaker identification

Minimum description length, regularisation and multi-modal data

Conventional feed forward Neural Networks have used the sum-of-squares cost function for training. A new cost function is presented here with a description length interpretation based on Rissanen's Minimum Description Length principle. It is a heuristic that has a rough interpretation as the number of data points fit by the model. Not concerned with finding optimal descriptions, the cost function prefers to form minimum descriptions in a naive way for computational convenience. The cost function is called the Naive Description Length cost function. Finding minimum description models will be shown to be closely related to the identification of clusters in the data. As a consequence the minimum of this cost function approximates the most probable mode of the data rather than the sum-of-squares cost function that approximates the mean. The new cost function is shown to provide information about the structure of the data. This is done by inspecting the dependence of the error to the amount of regularisation. This structure provides a method of selecting regularisation parameters as an alternative or supplement to Bayesian methods. The new cost function is tested on a number of multi-valued problems such as a simple inverse kinematics problem. It is also tested on a number of classification and regression problems. The mode-seeking property of this cost function is shown to improve prediction in time series problems. Description length principles are used in a similar fashion to derive a regulariser to control network complexity.

Development of Multi-modal and Super-resolved Retinal Imaging Systems

Advancements in retinal imaging technologies have drastically improved the quality of eye care in the past couple decades. Scanning laser ophthalmoscopy (SLO) and optical coherence tomography (OCT) are two examples of critical imaging modalities for the diagnosis of retinal pathologies. However current-generation SLO and OCT systems have limitations in diagnostic capability due to the following factors: the use of bulky tabletop systems, monochromatic imaging, and resolution degradation due to ocular aberrations and diffraction.

Bulky tabletop SLO and OCT systems are incapable of imaging patients that are supine, under anesthesia, or otherwise unable to maintain the required posture and fixation. Monochromatic SLO and OCT imaging prevents the identification of various color-specific diagnostic markers visible with color fundus photography like those of neovascular age-related macular degeneration. Resolution degradation due to ocular aberrations and diffraction has prevented the imaging of photoreceptors close to the fovea without the use of adaptive optics (AO), which require bulky and expensive components that limit the potential for widespread clinical use.

In this dissertation, techniques for extending the diagnostic capability of SLO and OCT systems are developed. These techniques include design strategies for miniaturizing and combining SLO and OCT to permit multi-modal, lightweight handheld probes to extend high quality retinal imaging to pediatric eye care. In addition, a method for extending true color retinal imaging to SLO to enable high-contrast, depth-resolved, high-fidelity color fundus imaging is demonstrated using a supercontinuum light source. Finally, the development and combination of SLO with a super-resolution confocal microscopy technique known as optical photon reassignment (OPRA) is demonstrated to enable high-resolution imaging of retinal photoreceptors without the use of adaptive optics.

Speaker identification using higher order spectral phase features and their effectiveness vis-a-vis Mel-Cepstral features

The effectiveness of higher-order spectral (HOS) phase features in speaker recognition is investigated by comparison with Mel Cepstral features on the same speech data. HOS phase features retain phase information from the Fourier spectrum unlikeMel–frequency Cepstral coefficients (MFCC). Gaussian mixture models are constructed from Mel– Cepstral features and HOS features, respectively, for the same data from various speakers in the Switchboard telephone Speech Corpus. Feature clusters, model parameters and classification performance are analyzed. HOS phase features on their own provide a correct identification rate of about 97% on the chosen subset of the corpus. This is the same level of accuracy as provided by MFCCs. Cluster plots and model parameters are compared to show that HOS phase features can provide complementary information to better discriminate between speakers.

Assessment of Speech Dialog Systems using Multi-Modal Cognitive Load Analysis and Driving Performance Metrics

In this paper, cognitive load analysis via acoustic- and CAN-Bus-based driver performance metrics is employed to assess two different commercial speech dialog systems (SDS) during in-vehicle use. Several metrics are proposed to measure increases in stress, distraction and cognitive load and we compare these measures with statistical analysis of the speech recognition component of each SDS. It is found that care must be taken when designing an SDS as it may increase cognitive load which can be observed through increased speech response delay (SRD), changes in speech production due to negative emotion towards the SDS, and decreased driving performance on lateral control tasks. From this study, guidelines are presented for designing systems which are to be used in vehicular environments.

Multi-edge decimation in multi-modal 3D collaborative applications

Multi-resolution modelling has become essential as modern 3D applications demand 3D objects with higher LODs (LOD). Multi-modal devices such as PDAs and UMPCs do not have sufficient resources to handle the original 3D objects. The increased usage of collaborative applications has created many challenges for remote manipulation working with 3D objects of different quality. This paper studies how we can improve multi-resolution techniques by performing multiedge decimation and using annotative commands. It also investigates how devices with poorer quality 3D object can participate in collaborative actions.

Multi-modal object tracking using dynamic performance metrics

Intelligent surveillance systems typically use a single visual spectrum modality for their input. These systems work well in controlled conditions, but often fail when lighting is poor, or environmental effects such as shadows, dust or smoke are present. Thermal spectrum imagery is not as susceptible to environmental effects, however thermal imaging sensors are more sensitive to noise and they are only gray scale, making distinguishing between objects difficult. Several approaches to combining the visual and thermal modalities have been proposed, however they are limited by assuming that both modalities are perfuming equally well. When one modality fails, existing approaches are unable to detect the drop in performance and disregard the under performing modality. In this paper, a novel middle fusion approach for combining visual and thermal spectrum images for object tracking is proposed. Motion and object detection is performed on each modality and the object detection results for each modality are fused base on the current performance of each modality. Modality performance is determined by comparing the number of objects tracked by the system with the number detected by each mode, with a small allowance made for objects entering and exiting the scene. The tracking performance of the proposed fusion scheme is compared with performance of the visual and thermal modes individually, and a baseline middle fusion scheme. Improvement in tracking performance using the proposed fusion approach is demonstrated. The proposed approach is also shown to be able to detect the failure of an individual modality and disregard its results, ensuring performance is not degraded in such situations.

Spectral coding methods for speech compression and speaker identification

This thesis investigates aspects of encoding the speech spectrum at low bit rates, with extensions to the effect of such coding on automatic speaker identification. Vector quantization (VQ) is a technique for jointly quantizing a block of samples at once, in order to reduce the bit rate of a coding system. The major drawback in using VQ is the complexity of the encoder. Recent research has indicated the potential applicability of the VQ method to speech when product code vector quantization (PCVQ) techniques are utilized. The focus of this research is the efficient representation, calculation and utilization of the speech model as stored in the PCVQ codebook. In this thesis, several VQ approaches are evaluated, and the efficacy of two training algorithms is compared experimentally. It is then shown that these productcode vector quantization algorithms may be augmented with lossless compression algorithms, thus yielding an improved overall compression rate. An approach using a statistical model for the vector codebook indices for subsequent lossless compression is introduced. This coupling of lossy compression and lossless compression enables further compression gain. It is demonstrated that this approach is able to reduce the bit rate requirement from the current 24 bits per 20 millisecond frame to below 20, using a standard spectral distortion metric for comparison. Several fast-search VQ methods for use in speech spectrum coding have been evaluated. The usefulness of fast-search algorithms is highly dependent upon the source characteristics and, although previous research has been undertaken for coding of images using VQ codebooks trained with the source samples directly, the product-code structured codebooks for speech spectrum quantization place new constraints on the search methodology. The second major focus of the research is an investigation of the effect of lowrate spectral compression methods on the task of automatic speaker identification. The motivation for this aspect of the research arose from a need to simultaneously preserve the speech quality and intelligibility and to provide for machine-based automatic speaker recognition using the compressed speech. This is important because there are several emerging applications of speaker identification where compressed speech is involved. Examples include mobile communications where the speech has been highly compressed, or where a database of speech material has been assembled and stored in compressed form. Although these two application areas have the same objective - that of maximizing the identification rate - the starting points are quite different. On the one hand, the speech material used for training the identification algorithm may or may not be available in compressed form. On the other hand, the new test material on which identification is to be based may only be available in compressed form. Using the spectral parameters which have been stored in compressed form, two main classes of speaker identification algorithm are examined. Some studies have been conducted in the past on bandwidth-limited speaker identification, but the use of short-term spectral compression deserves separate investigation. Combining the major aspects of the research, some important design guidelines for the construction of an identification model when based on the use of compressed speech are put forward.