Structured discriminative models for noise robust continuous speech recognition

Recently there has been interest in structured discriminative models for speech recognition. In these models sentence posteriors are directly modelled, given a set of features extracted from the observation sequence, and hypothesised word sequence. In previous work these discriminative models have been combined with features derived from generative models for noise-robust speech recognition for continuous digits. This paper extends this work to medium to large vocabulary tasks. The form of the score-space extracted using the generative models, and parameter tying of the discriminative model, are both discussed. Update formulae for both conditional maximum likelihood and minimum Bayes' risk training are described. Experimental results are presented on small and medium to large vocabulary noise-corrupted speech recognition tasks: AURORA 2 and 4. © 2011 IEEE.

Word boundary modelling and full covariance gaussians for Arabic Speech-to-Text systems

This paper describes recent improvements to the Cambridge Arabic Large Vocabulary Continuous Speech Recognition (LVCSR) Speech-to-Text (STT) system. It is shown that wordboundary context markers provide a powerful method to enhance graphemic systems by implicit phonetic information, improving the modelling capability of graphemic systems. In addition, a robust technique for full covariance Gaussian modelling in the Minimum Phone Error (MPE) training framework is introduced. This reduces the full covariance training to a diagonal covariance training problem, thereby solving related robustness problems. The full system results show that the combined use of these and other techniques within a multi-branch combination framework reduces the Word Error Rate (WER) of the complete system by up to 5.9% relative. Copyright © 2011 ISCA.

Syllable language models for Mandarin speech recognition: exploiting character language models.

Mandarin Chinese is based on characters which are syllabic in nature and morphological in meaning. All spoken languages have syllabiotactic rules which govern the construction of syllables and their allowed sequences. These constraints are not as restrictive as those learned from word sequences, but they can provide additional useful linguistic information. Hence, it is possible to improve speech recognition performance by appropriately combining these two types of constraints. For the Chinese language considered in this paper, character level language models (LMs) can be used as a first level approximation to allowed syllable sequences. To test this idea, word and character level n-gram LMs were trained on 2.8 billion words (equivalent to 4.3 billion characters) of texts from a wide collection of text sources. Both hypothesis and model based combination techniques were investigated to combine word and character level LMs. Significant character error rate reductions up to 7.3% relative were obtained on a state-of-the-art Mandarin Chinese broadcast audio recognition task using an adapted history dependent multi-level LM that performs a log-linearly combination of character and word level LMs. This supports the hypothesis that character or syllable sequence models are useful for improving Mandarin speech recognition performance.

Discriminative spoken language understanding using word confusion networks

Current commercial dialogue systems typically use hand-crafted grammars for Spoken Language Understanding (SLU) operating on the top one or two hypotheses output by the speech recogniser. These systems are expensive to develop and they suffer from significant degradation in performance when faced with recognition errors. This paper presents a robust method for SLU based on features extracted from the full posterior distribution of recognition hypotheses encoded in the form of word confusion networks. Following [1], the system uses SVM classifiers operating on n-gram features, trained on unaligned input/output pairs. Performance is evaluated on both an off-line corpus and on-line in a live user trial. It is shown that a statistical discriminative approach to SLU operating on the full posterior ASR output distribution can substantially improve performance both in terms of accuracy and overall dialogue reward. Furthermore, additional gains can be obtained by incorporating features from the previous system output. © 2012 IEEE.

Using sub-word-level information for confidence estimation with conditional random field models

The task of word-level confidence estimation (CE) for automatic speech recognition (ASR) systems stands to benefit from the combination of suitably defined input features from multiple information sources. However, the information sources of interest may not necessarily operate at the same level of granularity as the underlying ASR system. The research described here builds on previous work on confidence estimation for ASR systems using features extracted from word-level recognition lattices, by incorporating information at the sub-word level. Furthermore, the use of Conditional Random Fields (CRFs) with hidden states is investigated as a technique to combine information for word-level CE. Performance improvements are shown using the sub-word-level information in linear-chain CRFs with appropriately engineered feature functions, as well as when applying the hidden-state CRF model at the word level.

ART 2-A: An Adaptive Resonance Algorithm for Rapid Category Learning and Recognition

This article introduces ART 2-A, an efficient algorithm that emulates the self-organizing pattern recognition and hypothesis testing properties of the ART 2 neural network architecture, but at a speed two to three orders of magnitude faster. Analysis and simulations show how the ART 2-A systems correspond to ART 2 dynamics at both the fast-learn limit and at intermediate learning rates. Intermediate learning rates permit fast commitment of category nodes but slow recoding, analogous to properties of word frequency effects, encoding specificity effects, and episodic memory. Better noise tolerance is hereby achieved without a loss of learning stability. The ART 2 and ART 2-A systems are contrasted with the leader algorithm. The speed of ART 2-A makes practical the use of ART 2 modules in large-scale neural computation.

Inter-Frame Contextual Modelling For Visual Speech Recognition

In this paper, we present a new approach to visual speech recognition which improves contextual modelling by combining Inter-Frame Dependent and Hidden Markov Models. This approach captures contextual information in visual speech that may be lost using a Hidden Markov Model alone. We apply contextual modelling to a large speaker independent isolated digit recognition task, and compare our approach to two commonly adopted feature based techniques for incorporating speech dynamics. Results are presented from baseline feature based systems and the combined modelling technique. We illustrate that both of these techniques achieve similar levels of performance when used independently. However significant improvements in performance can be achieved through a combination of the two. In particular we report an improvement in excess of 17% relative Word Error Rate in comparison to our best baseline system.

FPGA Implementation of a Pipelined Gaussian Calculation for HMM-Based Large Vocabulary Speech Recognition

A scalable large vocabulary, speaker independent speech recognition system is being developed using Hidden Markov Models (HMMs) for acoustic modeling and a Weighted Finite State Transducer (WFST) to compile sentence, word, and phoneme models. The system comprises a software backend search and an FPGA-based Gaussian calculation which are covered here. In this paper, we present an efficient pipelined design implemented both as an embedded peripheral and as a scalable, parallel hardware accelerator. Both architectures have been implemented on an Alpha Data XRC-5T1, reconfigurable computer housing a Virtex 5 SX95T FPGA. The core has been tested and is capable of calculating a full set of Gaussian results from 3825 acoustic models in 9.03 ms which coupled with a backend search of 5000 words has provided an accuracy of over 80%. Parallel implementations have been designed with up to 32 cores and have been successfully implemented with a clock frequency of 133?MHz.

Optimization of Weighted Finite State Transducer for Speech Recognition

There is considerable interest in creating embedded, speech recognition hardware using the weighted finite state transducer (WFST) technique but there are performance and memory usage challenges. Two system optimization techniques are presented to address this; one approach improves token propagation by removing the WFST epsilon input arcs; another one-pass, adaptive pruning algorithm gives a dramatic reduction in active nodes to be computed. Results for memory and bandwidth are given for a 5,000 word vocabulary giving a better practical performance than conventional WFST; this is then exploited in an adaptive pruning algorithm that reduces the active nodes from 30,000 down to 4,000 with only a 2 percent sacrifice in speech recognition accuracy; these optimizations lead to a more simplified design with deterministic performance.

The idea, the machine, and the art: word and image in the 21st century

Taking in recent advances in neuroscience and digital technology, Gander and Garland assess the state of the inter-arts in America and the Western world, exploring and questioning the primacy of affect in an increasingly hypertextual everyday environment. In this analysis they signal a move beyond W. J. T. Mitchell’s coinage of the ‘imagetext’ to an approach that centres the reader-viewer in a recognition, after John Dewey, of ‘art as experience’. New thinking in cognitive and computer sciences about the relationship between the body and the mind challenges any established definitions of ‘embodiment’, ‘materiality’, ‘virtuality’ and even ‘intelligence, they argue, whilst ‘Extended Mind Theory’, they note, marries our cognitive processes with the material forms with which we engage, confirming and complicating Marshall McLuhan’s insight, decades ago, that ‘all media are “extensions of man”’. In this chapter, Gander and Garland open paths and suggest directions into understandings and critical interpretations of new and emerging imagetext worlds and experiences.

Visual word learning in good and poor readers

Forty students from regular, grade five classes were divided into two groups of twenty, a good reader group and a' poor reader group, on the basis. of their reading scores on Canadian Achievement Tests. .The subjects took. part in four experimental conditions iM which they .learned lists of pronounceable and unprono~nceable pseudowords, some with semantic referents, and responded to questions designed tci test visual perceptu~l learning and lexical ·and semantic association learning. It' was hypothesized "that the good reade~ group would be able to make use of graphemic and phonemic redundancy patterns in order to improv~·visuSl perceptual learning and lexical and semantic association lea~ningto a greater extent. than would .the poor reader gr6up. The data supported this hypothesis, and also indicated that, although the poor readers were less adept at using familiar sound and letter patterns, they were more dependent on· such pa~terns as an aid to visual recognition memory and semantic recall than were the good readers. It wa.s postulated that poor readers are in a double- ~ . bind situatio~ of having to choose between using weak graphemic-semantic associations or gr~pheme-phoneme associations which are also weak and which have hindered them in developing automaticity in. reading.

Behavioural and neural correlates of emotion recognition as a function of psychopathic personality traits

Psychopathy is associated with well-known characteristics such as a lack of empathy and impulsive behaviour, but it has also been associated with impaired recognition of emotional facial expressions. The use of event-related potentials (ERPs) to examine this phenomenon could shed light on the specific time course and neural activation associated with emotion recognition processes as they relate to psychopathic traits. In the current study we examined the PI , N170, and vertex positive potential (VPP) ERP components and behavioural performance with respect to scores on the Self-Report Psychopathy (SRP-III) questionnaire. Thirty undergraduates completed two tasks, the first of which required the recognition and categorization of affective face stimuli under varying presentation conditions. Happy, angry or fearful faces were presented under with attention directed to the mouth, nose or eye region and varied stimulus exposure duration (30, 75, or 150 ms). We found that behavioural performance to be unrelated to psychopathic personality traits in all conditions, but there was a trend for the Nl70 to peak later in response to fearful and happy facial expressions for individuals high in psychopathic traits. However, the amplitude of the VPP was significantly negatively associated with psychopathic traits, but only in response to stimuli presented under a nose-level fixation. Finally, psychopathic traits were found to be associated with longer N170 latencies in response to stimuli presented under the 30 ms exposure duration. In the second task, participants were required to inhibit processing of irrelevant affective and scrambled face distractors while categorizing unrelated word stimuli as living or nonliving. Psychopathic traits were hypothesized to be positively associated with behavioural performance, as it was proposed that individuals high in psychopathic traits would be less likely to automatically attend to task-irrelevant affective distractors, facilitating word categorization. Thus, decreased interference would be reflected in smaller N170 components, indicating less neural activity associated with processing of distractor faces. We found that overall performance decreased in the presence of angry and fearful distractor faces as psychopathic traits increased. In addition, the amplitude of the N170 decreased and the latency increased in response to affective distractor faces for individuals with higher levels of psychopathic traits. Although we failed to find the predicted behavioural deficit in emotion recognition in Task 1 and facilitation effect in Task 2, the findings of increased N170 and VPP latencies in response to emotional faces are consistent wi th the proposition that abnormal emotion recognition processes may in fact be inherent to psychopathy as a continuous personality trait.

k-NN based On-Line Handwritten Character recognition system

On-line handwriting recognition has been a frontier area of research for the last few decades under the purview of pattern recognition. Word processing turns to be a vexing experience even if it is with the assistance of an alphanumeric keyboard in Indian languages. A natural solution for this problem is offered through online character recognition. There is abundant literature on the handwriting recognition of western, Chinese and Japanese scripts, but there are very few related to the recognition of Indic script such as Malayalam. This paper presents an efficient Online Handwritten character Recognition System for Malayalam Characters (OHR-M) using K-NN algorithm. It would help in recognizing Malayalam text entered using pen-like devices. A novel feature extraction method, a combination of time domain features and dynamic representation of writing direction along with its curvature is used for recognizing Malayalam characters. This writer independent system gives an excellent accuracy of 98.125% with recognition time of 15-30 milliseconds

Development & evaluation of different acoustic models for Malayalam continuous speech recognition

Performance of any continuous speech recognition system is dependent on the accuracy of its acoustic model. Hence, preparation of a robust and accurate acoustic model lead to satisfactory recognition performance for a speech recognizer. In acoustic modeling of phonetic unit, context information is of prime importance as the phonemes are found to vary according to the place of occurrence in a word. In this paper we compare and evaluate the effect of context dependent tied (CD tied) models, context dependent (CD) and context independent (CI) models in the perspective of continuous speech recognition of Malayalam language. The database for the speech recognition system has utterance from 21 speakers including 11 female and 10 males. Our evaluation results show that CD tied models outperforms CI models over 21%.

The Degree of Word-Expansion of Lexicalized RRWW-Automata

Restarting automata can be seen as analytical variants of classical automata as well as of regulated rewriting systems. We study a measure for the degree of nondeterminism of (context-free) languages in terms of deterministic restarting automata that are (strongly) lexicalized. This measure is based on the number of auxiliary symbols (categories) used for recognizing a language as the projection of its characteristic language onto its input alphabet. This type of recognition is typical for analysis by reduction, a method used in linguistics for the creation and verification of formal descriptions of natural languages. Our main results establish a hierarchy of classes of context-free languages and two hierarchies of classes of non-context-free languages that are based on the expansion factor of a language.