The application of phonetic distribution normalisation to likelihood-maximising speech enhancement for robust ASR

Traditional speech enhancement methods optimise signal-level criteria such as signal-to-noise ratio, but such approaches are sub-optimal for noise-robust speech recognition. Likelihood-maximising (LIMA) frameworks on the other hand, optimise the parameters of speech enhancement algorithms based on state sequences generated by a speech recogniser for utterances of known transcriptions. Previous applications of LIMA frameworks have generated a set of global enhancement parameters for all model states without taking in account the distribution of model occurrence, making optimisation susceptible to favouring frequently occurring models, in particular silence. In this paper, we demonstrate the existence of highly disproportionate phonetic distributions on two corpora with distinct speech tasks, and propose to normalise the influence of each phone based on a priori occurrence probabilities. Likelihood analysis and speech recognition experiments verify this approach for improving ASR performance in noisy environments.

Automatic speaker recognition under adverse conditions

Speaker verification is the process of verifying the identity of a person by analysing their speech. There are several important applications for automatic speaker verification (ASV) technology including suspect identification, tracking terrorists and detecting a person’s presence at a remote location in the surveillance domain, as well as person authentication for phone banking and credit card transactions in the private sector. Telephones and telephony networks provide a natural medium for these applications. The aim of this work is to improve the usefulness of ASV technology for practical applications in the presence of adverse conditions. In a telephony environment, background noise, handset mismatch, channel distortions, room acoustics and restrictions on the available testing and training data are common sources of errors for ASV systems. Two research themes were pursued to overcome these adverse conditions: Modelling mismatch and modelling uncertainty. To directly address the performance degradation incurred through mismatched conditions it was proposed to directly model this mismatch. Feature mapping was evaluated for combating handset mismatch and was extended through the use of a blind clustering algorithm to remove the need for accurate handset labels for the training data. Mismatch modelling was then generalised by explicitly modelling the session conditions as a constrained offset of the speaker model means. This session variability modelling approach enabled the modelling of arbitrary sources of mismatch, including handset type, and halved the error rates in many cases. Methods to model the uncertainty in speaker model estimates and verification scores were developed to address the difficulties of limited training and testing data. The Bayes factor was introduced to account for the uncertainty of the speaker model estimates in testing by applying Bayesian theory to the verification criterion, with improved performance in matched conditions. Modelling the uncertainty in the verification score itself met with significant success. Estimating a confidence interval for the "true" verification score enabled an order of magnitude reduction in the average quantity of speech required to make a confident verification decision based on a threshold. The confidence measures developed in this work may also have significant applications for forensic speaker verification tasks.

Wide-baseline keypoint detection and matching with wide-angle images for vision based localisation

This thesis addresses the problem of detecting and describing the same scene points in different wide-angle images taken by the same camera at different viewpoints. This is a core competency of many vision-based localisation tasks including visual odometry and visual place recognition. Wide-angle cameras have a large field of view that can exceed a full hemisphere, and the images they produce contain severe radial distortion. When compared to traditional narrow field of view perspective cameras, more accurate estimates of camera egomotion can be found using the images obtained with wide-angle cameras. The ability to accurately estimate camera egomotion is a fundamental primitive of visual odometry, and this is one of the reasons for the increased popularity in the use of wide-angle cameras for this task. Their large field of view also enables them to capture images of the same regions in a scene taken at very different viewpoints, and this makes them suited for visual place recognition. However, the ability to estimate the camera egomotion and recognise the same scene in two different images is dependent on the ability to reliably detect and describe the same scene points, or ‘keypoints’, in the images. Most algorithms used for this purpose are designed almost exclusively for perspective images. Applying algorithms designed for perspective images directly to wide-angle images is problematic as no account is made for the image distortion. The primary contribution of this thesis is the development of two novel keypoint detectors, and a method of keypoint description, designed for wide-angle images. Both reformulate the Scale- Invariant Feature Transform (SIFT) as an image processing operation on the sphere. As the image captured by any central projection wide-angle camera can be mapped to the sphere, applying these variants to an image on the sphere enables keypoints to be detected in a manner that is invariant to image distortion. Each of the variants is required to find the scale-space representation of an image on the sphere, and they differ in the approaches they used to do this. Extensive experiments using real and synthetically generated wide-angle images are used to validate the two new keypoint detectors and the method of keypoint description. The best of these two new keypoint detectors is applied to vision based localisation tasks including visual odometry and visual place recognition using outdoor wide-angle image sequences. As part of this work, the effect of keypoint coordinate selection on the accuracy of egomotion estimates using the Direct Linear Transform (DLT) is investigated, and a simple weighting scheme is proposed which attempts to account for the uncertainty of keypoint positions during detection. A word reliability metric is also developed for use within a visual ‘bag of words’ approach to place recognition.

Aerial SLAM with a single camera using visual expectation

Micro aerial vehicles (MAVs) are a rapidly growing area of research and development in robotics. For autonomous robot operations, localization has typically been calculated using GPS, external camera arrays, or onboard range or vision sensing. In cluttered indoor or outdoor environments, onboard sensing is the only viable option. In this paper we present an appearance-based approach to visual SLAM on a flying MAV using only low quality vision. Our approach consists of a visual place recognition algorithm that operates on 1000 pixel images, a lightweight visual odometry algorithm, and a visual expectation algorithm that improves the recall of place sequences and the precision with which they are recalled as the robot flies along a similar path. Using data gathered from outdoor datasets, we show that the system is able to perform visual recognition with low quality, intermittent visual sensory data. By combining the visual algorithms with the RatSLAM system, we also demonstrate how the algorithms enable successful SLAM.

A survey of tagging techniques for music, speech and environmental sound

Sound tagging has been studied for years. Among all sound types, music, speech, and environmental sound are three hottest research areas. This survey aims to provide an overview about the state-of-the-art development in these areas.We discuss about the meaning of tagging in different sound areas at the beginning of the journey. Some examples of sound tagging applications are introduced in order to illustrate the significance of this research. Typical tagging techniques include manual, automatic, and semi-automatic approaches.After reviewing work in music, speech and environmental sound tagging, we compare them and state the research progress to date. Research gaps are identified for each research area and the common features and discriminations between three areas are discovered as well. Published datasets, tools used by researchers, and evaluation measures frequently applied in the analysis are listed. In the end, we summarise the worldwide distribution of countries dedicated to sound tagging research for years.

Robust automatic speaker linking and attribution

This research makes a major contribution which enables efficient searching and indexing of large archives of spoken audio based on speaker identity. It introduces a novel technique dubbed as “speaker attribution” which is the task of automatically determining ‘who spoke when?’ in recordings and then automatically linking the unique speaker identities within each recording across multiple recordings. The outcome of the research will also have significant impact in improving the performance of automatic speech recognition systems through the extracted speaker identities.

GrabCutSFM : how 3D information improves unsupervised object segmentation

In this paper, we present an unsupervised graph cut based object segmentation method using 3D information provided by Structure from Motion (SFM), called Grab- CutSFM. Rather than focusing on the segmentation problem using a trained model or human intervention, our approach aims to achieve meaningful segmentation autonomously with direct application to vision based robotics. Generally, object (foreground) and background have certain discriminative geometric information in 3D space. By exploring the 3D information from multiple views, our proposed method can segment potential objects correctly and automatically compared to conventional unsupervised segmentation using only 2D visual cues. Experiments with real video data collected from indoor and outdoor environments verify the proposed approach.

Linguistic and perceptual processing of communicative cues in Asperger Syndrome

Asperger Syndrome (AS) belongs to autism spectrum disorders where both verbal and non-verbal communication difficulties are at the core of the impairment. Social communication requires a complex use of affective, linguistic-cognitive and perceptual processes. In the four studies included in the current thesis, some of the linguistic and perceptual factors that are important for face-to-face communication were studied using behavioural methods. In all four studies the results obtained from individuals with AS were compared with typically developed age, gender and IQ matched controls. First, the language skills of school-aged children were characterized in detail with standardized tests that measured different aspects of receptive and expressive language (Study I). The children with AS were found to be worse than the controls in following complex verbal instructions. Next, the visual perception of facial expressions of emotion with varying degrees of visual detail was examined (Study II). Adults with AS were found to have impaired recognition of facial expressions on the basis of very low spatial frequencies which are important for processing global information. Following that, multisensory perception was investigated by looking at audiovisual speech perception (Studies III and IV). Adults with AS were found to perceive audiovisual speech qualitatively differently from typically developed adults, although both groups were equally accurate in recognizing auditory and visual speech presented alone. Finally, the effect of attention on audiovisual speech perception was studied by registering eye gaze behaviour (Study III) and by studying the voluntary control of visual attention (Study IV). The groups did not differ in eye gaze behaviour or in the voluntary control of visual attention. The results of the study series demonstrate that many factors underpinning face-to-face social communication are atypical in AS. In contrast with previous assumptions about intact language abilities, the current results show that children with AS have difficulties in understanding complex verbal instructions. Furthermore, the study makes clear that deviations in the perception of global features in faces expressing emotions as well as in the multisensory perception of speech are likely to harm face-to-face social communication.

A mode-shape classification technique for robust speech rate estimation and syllable nuclei detection

Acoustic feature based speech (syllable) rate estimation and syllable nuclei detection are important problems in automatic speech recognition (ASR), computer assisted language learning (CALL) and fluency analysis. A typical solution for both the problems consists of two stages. The first stage involves computing a short-time feature contour such that most of the peaks of the contour correspond to the syllabic nuclei. In the second stage, the peaks corresponding to the syllable nuclei are detected. In this work, instead of the peak detection, we perform a mode-shape classification, which is formulated as a supervised binary classification problem - mode-shapes representing the syllabic nuclei as one class and remaining as the other. We use the temporal correlation and selected sub-band correlation (TCSSBC) feature contour and the mode-shapes in the TCSSBC feature contour are converted into a set of feature vectors using an interpolation technique. A support vector machine classifier is used for the classification. Experiments are performed separately using Switchboard, TIMIT and CTIMIT corpora in a five-fold cross validation setup. The average correlation coefficients for the syllable rate estimation turn out to be 0.6761, 0.6928 and 0.3604 for three corpora respectively, which outperform those obtained by the best of the existing peak detection techniques. Similarly, the average F-scores (syllable level) for the syllable nuclei detection are 0.8917, 0.8200 and 0.7637 for three corpora respectively. (C) 2016 Elsevier B.V. All rights reserved.

Development of the 2003 CU-HTK conversational telephone speech transcription system

This paper describes the development of the 2003 CU-HTK large vocabulary speech recognition system for Conversational Telephone Speech (CTS). The system was designed based on a multi-pass, multi-branch structure where the output of all branches is combined using system combination. A number of advanced modelling techniques such as Speaker Adaptive Training, Heteroscedastic Linear Discriminant Analysis, Minimum Phone Error estimation and specially constructed Single Pronunciation dictionaries were employed. The effectiveness of each of these techniques and their potential contribution to the result of system combination was evaluated in the framework of a state-of-the-art LVCSR system with sophisticated adaptation. The final 2003 CU-HTK CTS system constructed from some of these models is described and its performance on the DARPA/NIST 2003 Rich Transcription (RT-03) evaluation test set is discussed.

Discriminative language model adaptation for Mandarin broadcast speech transcription and translation

This paper investigates unsupervised test-time adaptation of language models (LM) using discriminative methods for a Mandarin broadcast speech transcription and translation task. A standard approach to adapt interpolated language models to is to optimize the component weights by minimizing the perplexity on supervision data. This is a widely made approximation for language modeling in automatic speech recognition (ASR) systems. For speech translation tasks, it is unclear whether a strong correlation still exists between perplexity and various forms of error cost functions in recognition and translation stages. The proposed minimum Bayes risk (MBR) based approach provides a flexible framework for unsupervised LM adaptation. It generalizes to a variety of forms of recognition and translation error metrics. LM adaptation is performed at the audio document level using either the character error rate (CER), or translation edit rate (TER) as the cost function. An efficient parameter estimation scheme using the extended Baum-Welch (EBW) algorithm is proposed. Experimental results on a state-of-the-art speech recognition and translation system are presented. The MBR adapted language models gave the best recognition and translation performance and reduced the TER score by up to 0.54% absolute. © 2007 IEEE.

Language model combination and adaptation using weighted finite state transducers

In speech recognition systems language model (LMs) are often constructed by training and combining multiple n-gram models. They can be either used to represent different genres or tasks found in diverse text sources, or capture stochastic properties of different linguistic symbol sequences, for example, syllables and words. Unsupervised LM adaptation may also be used to further improve robustness to varying styles or tasks. When using these techniques, extensive software changes are often required. In this paper an alternative and more general approach based on weighted finite state transducers (WFSTs) is investigated for LM combination and adaptation. As it is entirely based on well-defined WFST operations, minimum change to decoding tools is needed. A wide range of LM combination configurations can be flexibly supported. An efficient on-the-fly WFST decoding algorithm is also proposed. Significant error rate gains of 7.3% relative were obtained on a state-of-the-art broadcast audio recognition task using a history dependently adapted multi-level LM modelling both syllable and word sequences. ©2010 IEEE.

Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple subsystems developed at different sites. Cross system adaptation can be used as an alternative to direct hypothesis level combination schemes such as ROVER. In normal cross adaptation it is assumed that useful diversity among systems exists only at acoustic level. However, complimentary features among complex LVCSR systems also manifest themselves in other layers of modelling hierarchy, e.g., subword and word level. It is thus interesting to also cross adapt language models (LM) to capture them. In this paper cross adaptation of multi-level LMs modelling both syllable and word sequences was investigated to improve LVCSR system combination. Significant error rate gains up to 6.7% rel. were obtained over ROVER and acoustic model only cross adaptation when combining 13 Chinese LVCSR subsystems used in the 2010 DARPA GALE evaluation. © 2010 ISCA.