Search

Phenotypic antimicrobial resistance testing of 160 Campylobacter jejuni isolates from broiler caeca and neck skin samples collected at key stages during processing from 32 batches was conducted to: (i) determine the prevalence of antimicrobial resistance in C. jejuni isolates (ii) investigate if C. jejuni strains with particular resistance profiles were more frequently recovered at the end of processing, (iii) to establish if multiple strains with different resistance profiles were present within individual batches of broilers and (iv) to determine the virulence profiles of isolates. C. jejuni isolates were tested for resistance to erythromycin (ERY), ciprofloxacin (CIP), tetracycline (TET), gentamicin (GEN), nalidixic acid (NAL) and streptomycin (STR) using a broth dilution phenotypic method. Resistance to TET was most prevalent (46%), followed by resistance to CIP and NAL (29%). All isolates were susceptible to ERY and GEN, while resistance to STR was low (2%). Strains with different resistance profiles were found within 8 individual batches of broilers. Both resistant and susceptible strains were present within 4 individual batches of broilers. Whole genome sequencing of 132 C. jejuni isolates was conducted to: (i) determine relationship between resistant phenotypes and genotypes, (ii) examine genomic diversity of strains present in caeca and in neck skin samples collected at different processing stages and (iii) gain insights into the genomic profiles, including the resistome and virulome of C. jejuni isolates. A high level of correlation between resistant phenotypes and genotypes was observed (99.2%). Resistance determinants identified in sequenced isolates using ResFinder included the tet(O) gene, tet(O/32/O) gene, gyrA p.T86I mutation and blaOXA-61 like genes. A high correlation between sequence type (ST) and antimicrobial resistance was observed. The 132 sequenced C. jejuni isolates were assigned to 10 sequence types (STs), which belonged to 7 clonal complexes, based on multi-locus sequence typing (MLST). ST-257 was prevalent with 58 isolates assigned to it, followed by ST-51 with 25 isolates, ST-10089 with 16 isolates, ST-48 with 13 isolates and ST-50 with 12 isolates. Four isolates were assigned to ST-354, while 1 isolate was assigned to STs 464, 42, 1489 and 6209. All sequenced isolates were assigned to 17 cgSTs. Virulence genes which might have a role in Campylobacter survival and environmental adaptation during key processing stages and a number of virulence genes associated with human infection and colonization of broilers were detected. This study is one of the first to provide data on genomic diversity of C. jejuni isolates from broilers collected during processing in Ireland.

This study examined the impact of key processing stages and flock variables on the prevalence of Campylobacter on broiler carcasses. Overall, the prevalence of Campylobacter was 62% in caeca, and 68%, 65% and 62% in neck skin samples collected after evisceration, final wash and carcass chilling, respectively. Campylobacter were found in 32% of caeca, and 52%, 40% and 32% of neck skin samples collected after evisceration, final wash and carcass chilling, respectively from first thin broiler batches. Final thin broiler batches were more frequently contaminated with prevalences of 83% found in caeca, 80% in neck skin samples collected after evisceration and 83% found in neck skin samples collected after both final wash and carcass chilling stages (p < 0.05). Thinning status had a significant effect on Campylobacter counts with significantly higher counts observed in samples from final thin batches (p < 0.05). Highest Campylobacter concentrations in neck skin samples were observed at the evisceration stage in both first and final thin samples, with counts ranging from 2.0 to 3.8 log10 CFU/g and 2.3 to 4.8 log10 CFU/g in first and final thin batches, respectively. All first thin samples had counts below the European Union (EU) Process Hygiene Criterion threshold level of 3 log10 CFU/g after chilling while 52% of final thin batches had counts above this limit.

We describe the use of support vector machines (SVMs) for continuous speech recognition by incorporating them in segmental minimum Bayes risk decoding. Lattice cutting is used to convert the Automatic Speech Recognition search space into sequences of smaller recognition problems. SVMs are then trained as discriminative models over each of these problems and used in a rescoring framework. We pose the estimation of a posterior distribution over hypotheses in these regions of acoustic confusion as a logistic regression problem. We also show that GiniSVMs can be used as an approximation technique to estimate the parameters of the logistic regression problem. On a small vocabulary recognition task we show that the use of GiniSVMs can improve the performance of a well trained hidden Markov model system trained under the Maximum Mutual Information criterion. We also find that it is possible to derive reliable confidence scores over the GiniSVM hypotheses and that these can be used to good effect in hypothesis combination. We discuss the problems that we expect to encounter in extending this approach to large vocabulary continuous speech recognition and describe initial investigation of constrained estimation techniques to derive feature spaces for SVMs.

Lattice segmentation techniques developed for Minimum Bayes Risk decoding in large vocabulary speech recognition tasks are used to compute the statistics needed for discriminative training algorithms that estimate HMM parameters so as to reduce the overall risk over the training data. New estimation procedures are developed and evaluated for both small and large vocabulary recognition tasks, and additive performance improvements are shown relative to maximum mutual information estimation. These relative gains are explained through a detailed analysis of individual word recognition errors.

The widely used maximum likelihood linear regression speaker adaptation procedure suffers from overtraining when used for rapid adaptation tasks in which the amount of adaptation data is severely limited. This is a well known difficulty associated with the expectation maximization algorithm. We use an information geometric analysis of the expectation maximization algorithm as an alternating minimization of a Kullback–Leibler-type divergence to see the cause of this difficulty, and propose a more robust discounted likelihood estimation procedure. This gives rise to a discounted likelihood linear regression procedure, which is a variant of maximum likelihood linear regression suited for small adaptation sets. Our procedure is evaluated on an unsupervised rapid adaptation task defined on the Switchboard conversational telephone speech corpus, where our proposed procedure improves word error rate by 1.6% (absolute) with as little as 5 seconds of adaptation data, which is a situation in which maximum likelihood linear regression overtrains in the first iteration of adaptation. We compare several realizations of discounted likelihood linear regression with maximum likelihood linear regression and other simple maximum likelihood linear regression variants, and discuss issues that arise in implementing our discounted likelihood procedures.

In the early 1990s, the availability of the TIMIT read-speech phonetically transcribed corpus led to work at AT&T on the automatic inference of pronunciation variation. This work, briefly summarized here, used stochastic decision trees trained on phonetic and linguistic features, and was applied to the DARPA North American Business News read-speech ASR task. More recently, the ICSI spontaneous-speech phonetically transcribed corpus was collected at the behest of the 1996 and 1997 LVCSR Summer Workshops held at Johns Hopkins University. A 1997 workshop (WS97) group focused on pronunciation inference from this corpus for application to the DoD Switchboard spontaneous telephone speech ASR task. We describe several approaches taken there. These include (1) one analogous to the AT&T approach, (2) one, inspired by work at WS96 and CMU, that involved adding pronunciation variants of a sequence of one or more words (`multiwords') in the corpus (with corpus-derived probabilities) into the ASR lexicon, and (1+2) a hybrid approach in which a decision-tree model was used to automatically phonetically transcribe a much larger speech corpus than ICSI and then the multiword approach was used to construct an ASR recognition pronunciation lexicon. Anfang der neunziger Jahre hat die Verfugbarkeit des phonetisch umschrifteten TIMIT Korpus fur gelesene Sprache bei AT&T zur Forschung uber die automatische Inferenz von Aussprachemodellen gefuhrt. In dieser Arbeit wurden Decision Trees auf der Basis von phonetischen wie linguistischen Merkmalen trainiert und auf den DARPA North American Business News Korpus fur die automatische Erkennung gelesener Sprache angewendet. In jungerer Zeit wurde fur die an der Johns Hopkins Universitat ausgerichteten LVCSR Summer Workshops 1996 und 1997 der phonetisch transkribierte ICSI Korpus fur spontane Sprache erstellt. Eine Arbeitsgruppe des Workshops 1997 (WS97) hat sich auf Basis dieses Korpus mit der Inferenz von Aussprachemodellen zur Anwendung auf den DoD Switchboard Korpus fur spontane Telefonsprache befa?st. Wir beschreiben im folgenden verschiedene Ansatze in diese Richtung. Sie umfassen im einzelnen: ein zu dem AT&T Ansatz (1) analoges Modell, einen Ansatz (2), der auf der Arbeit am WS96 und an der CMU aufbaut und die automatische Generierung und Aufnahme von Aussprachevarianten fur Folgen von einem oder mehreren Worten (`multiwords') in das ASR Ausspracheworterbuch (mit aus dem Trainingskorpus abgeleiteten Wahrscheinlichkeiten) behandelt, und ein hybrides Modell (1+2), in dem ein Decision Tree basiertes Modell zur automatischen phonetischen Transkription eines sehr viel gro?seren als des ICSI Korpus verwendet wird um dann mittels des `multiword'-Ansatzes ein ASR Ausspracheworterbuch zu generieren. Au debut des annees 90, la disponibilite du corpus TIMIT de transcription phonetique de parole dictee conduisit les chercheurs de AT&T a travailler sur l'inference automatique des variations de prononciation. Ce travail, brievement resume ici, utilisa des arbres de decision construit a partir de traits phonetiques et linguistiques, et fut applique a la reconaissance de la parole dictee dans le domaine du DARPA North American Business News. Plus recemment, le corpus ICSI de transcription phonetique de parole spontanee fut mis au point a l'occasion des ateliers de travails d'ete a l'universite Johns Hopkins (WS) en 1996 et 1997. Un groupe de travail du WS97 mit l'accent sur l'inference de la prononciation a partir de ce corpus afin de l'utiliser dans le domaine de la parole spontanee telephonique DoD Switchboard. Nous decrivons plusieurs approches suivies a cette occasion. Ces approches incluent (1) une approche analogue a celle de AT&T, (2) une autre approche inspiree par le travail accompli a WS96 et a CMU qui necessite l'addition de variantes de prononciation d'une sequence d'un ou plusieurs mots du corpus (`multimots', avec des probabilites extraites de ce corpus) au dictionnaire de prononciation du systeme de reconnaissance de la parole, et (1+2) une approche hybride ou un model d'arbres de decision est utilise pour determiner automatiquement la transcription phonetique d'un corpus bien plus long que celui d'ICSI, et ou l'approche multimots est utilisee pour construire un dictionnaire de prononciation pour la reconnaissance automatique de la parole.