Improving spoken term detection using complementary information

This research has made contributions to the area of spoken term detection (STD), defined as the process of finding all occurrences of a specified search term in a large collection of speech segments. The use of visual information in the form of lip movements of the speaker in addition to audio and the use of topic of the speech segments, and the expected frequency of words in the target speech domain, are proposed. By using these complementary information, improvement in the performance of STD has been achieved which enables efficient search of key words in large collection of multimedia documents.

Digit recognition using trispectral features

Features derived from the trispectra of DFT magnitude slices are used for multi-font digit recognition. These features are insensitive to translation, rotation, or scaling of the input. They are also robust to noise. Classification accuracy tests were conducted on a common data base of 256× 256 pixel bilevel images of digits in 9 fonts. Randomly rotated and translated noisy versions were used for training and testing. The results indicate that the trispectral features are better than moment invariants and affine moment invariants. They achieve a classification accuracy of 95% compared to about 81% for Hu's (1962) moment invariants and 39% for the Flusser and Suk (1994) affine moment invariants on the same data in the presence of 1% impulse noise using a 1-NN classifier. For comparison, a multilayer perceptron with no normalization for rotations and translations yields 34% accuracy on 16× 16 pixel low-pass filtered and decimated versions of the same data.

Postcode segmentation and recognition using projections and bispectral features

A system to segment and recognize Australian 4-digit postcodes from address labels on parcels is described. Images of address labels are preprocessed and adaptively thresholded to reduce noise. Projections are used to segment the line and then the characters comprising the postcode. Individual digits are recognized using bispectral features extracted from their parallel beam projections. These features are insensitive to translation, scaling and rotation, and robust to noise. Results on scanned images are presented. The system is currently being improved and implemented to work on-line.

The effect of prior visual information on recognition of speech and sounds

To identify and categorize complex stimuli such as familiar objects or speech, the human brain integrates information that is abstracted at multiple levels from its sensory inputs. Using cross-modal priming for spoken words and sounds, this functional magnetic resonance imaging study identified 3 distinct classes of visuoauditory incongruency effects: visuoauditory incongruency effects were selective for 1) spoken words in the left superior temporal sulcus (STS), 2) environmental sounds in the left angular gyrus (AG), and 3) both words and sounds in the lateral and medial prefrontal cortices (IFS/mPFC). From a cognitive perspective, these incongruency effects suggest that prior visual information influences the neural processes underlying speech and sound recognition at multiple levels, with the STS being involved in phonological, AG in semantic, and mPFC/IFS in higher conceptual processing. In terms of neural mechanisms, effective connectivity analyses (dynamic causal modeling) suggest that these incongruency effects may emerge via greater bottom-up effects from early auditory regions to intermediate multisensory integration areas (i.e., STS and AG). This is consistent with a predictive coding perspective on hierarchical Bayesian inference in the cortex where the domain of the prediction error (phonological vs. semantic) determines its regional expression (middle temporal gyrus/STS vs. AG/intraparietal sulcus).

Analysis and detection of cognitive load and frustration in drivers' speech

Non-driving related cognitive load and variations of emotional state may impact a driver’s capability to control a vehicle and introduces driving errors. Availability of reliable cognitive load and emotion detection in drivers would benefit the design of active safety systems and other intelligent in-vehicle interfaces. In this study, speech produced by 68 subjects while driving in urban areas is analyzed. A particular focus is on speech production differences in two secondary cognitive tasks, interactions with a co-driver and calls to automated spoken dialog systems (SDS), and two emotional states during the SDS interactions - neutral/negative. A number of speech parameters are found to vary across the cognitive/emotion classes. Suitability of selected cepstral- and production-based features for automatic cognitive task/emotion classification is investigated. A fusion of GMM/SVM classifiers yields an accuracy of 94.3% in cognitive task and 81.3% in emotion classification.

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.

The delta-phase spectrum with application to voice activity detection and speaker recognition

For several reasons, the Fourier phase domain is less favored than the magnitude domain in signal processing and modeling of speech. To correctly analyze the phase, several factors must be considered and compensated, including the effect of the step size, windowing function and other processing parameters. Building on a review of these factors, this paper investigates a spectral representation based on the Instantaneous Frequency Deviation, but in which the step size between processing frames is used in calculating phase changes, rather than the traditional single sample interval. Reflecting these longer intervals, the term delta-phase spectrum is used to distinguish this from instantaneous derivatives. Experiments show that mel-frequency cepstral coefficients features derived from the delta-phase spectrum (termed Mel-Frequency delta-phase features) can produce broadly similar performance to equivalent magnitude domain features for both voice activity detection and speaker recognition tasks. Further, it is shown that the fusion of the magnitude and phase representations yields performance benefits over either in isolation.

Position, rotation, and scale invariant recognition of images using higher-order spectra

A new approach to recognition of images using invariant features based on higher-order spectra is presented. Higher-order spectra are translation invariant because translation produces linear phase shifts which cancel. Scale and amplification invariance are satisfied by the phase of the integral of a higher-order spectrum along a radial line in higher-order frequency space because the contour of integration maps onto itself and both the real and imaginary parts are affected equally by the transformation. Rotation invariance is introduced by deriving invariants from the Radon transform of the image and using the cyclic-shift invariance property of the discrete Fourier transform magnitude. Results on synthetic and actual images show isolated, compact clusters in feature space and high classification accuracies