Cross likelihood ratio based speaker clustering using eigenvoice models

This paper proposes the use of eigenvoice modeling techniques with the Cross Likelihood Ratio (CLR) as a criterion for speaker clustering within a speaker diarization system. The CLR has previously been shown to be a robust decision criterion for speaker clustering using Gaussian Mixture Models. Recently, eigenvoice modeling techniques have become increasingly popular, due to its ability to adequately represent a speaker based on sparse training data, as well as an improved capture of differences in speaker characteristics. This paper hence proposes that it would be beneficial to capitalize on the advantages of eigenvoice modeling in a CLR framework. Results obtained on the 2002 Rich Transcription (RT-02) Evaluation dataset show an improved clustering performance, resulting in a 35.1% relative improvement in the overall Diarization Error Rate (DER) compared to the baseline system.

Inflammatory cytokine responses to progressive resistance training and supplementation with fortified milk in men aged 50+ years : an 18-month randomized controlled trial

We examined the effects of progressive resistance training (PRT) and supplementation with calcium-vitamin D(3) fortified milk on markers of systemic inflammation, and the relationship between inflammation and changes in muscle mass, size and strength. Healthy men aged 50-79 years (n = 180) participated in this 18-month randomized controlled trial that comprised a factorial 2 x 2 design. Participants were randomized to (1) PRT + fortified milk supplement, (2) PRT, (3) fortified milk supplement, or (4) a control group. Participants assigned to PRT trained 3 days per week, while those in the supplement groups consumed 400 ml day(-1) of milk containing 1,000 mg calcium plus 800 IU vitamin D(3). We collected venous blood samples at baseline, 12 and 18 months to measure the serum concentrations of IL-6, TNF-alpha and hs-CRP. There were no exercise x supplement interactions, but serum IL-6 was 29% lower (95% CI, -62, 0) in the PRT group compared with the control group after 12 months. Conversely, IL-6 was 31% higher (95% CI, -2, 65) in the supplement group compared with the non-supplemented groups after 12 and 18 months. These between-group differences did not persist after adjusting for changes in fat mass. In the PRT group, mid-tibia muscle cross-sectional area increased less in men with higher pre-training inflammation compared with those men with lower inflammation (net difference similar to 2.5%, p < 0.05). In conclusion, serum IL-6 concentration decreased following PRT, whereas it increased after supplementation with fortified milk concomitant with changes in fat mass. Furthermore, low-grade inflammation at baseline restricted muscle hypertrophy following PRT.

Eigenvoice modeling for cross likelihood ratio based speaker clustering : a Bayesian approach

This paper proposes the use of Bayesian approaches with the cross likelihood ratio (CLR) as a criterion for speaker clustering within a speaker diarization system, using eigenvoice modeling techniques. The CLR has previously been shown to be an effective decision criterion for speaker clustering using Gaussian mixture models. Recently, eigenvoice modeling has become an increasingly popular technique, due to its ability to adequately represent a speaker based on sparse training data, as well as to provide an improved capture of differences in speaker characteristics. The integration of eigenvoice modeling into the CLR framework to capitalize on the advantage of both techniques has also been shown to be beneficial for the speaker clustering task. Building on that success, this paper proposes the use of Bayesian methods to compute the conditional probabilities in computing the CLR, thus effectively combining the eigenvoice-CLR framework with the advantages of a Bayesian approach to the diarization problem. Results obtained on the 2002 Rich Transcription (RT-02) Evaluation dataset show an improved clustering performance, resulting in a 33.5% relative improvement in the overall Diarization Error Rate (DER) compared to the baseline system.

The role of intensity of interval training on fat oxidation and eating behaviour in overweight/obese men

The increasing prevalence of obesity in society has been associated with a number of atherogenic risk factors such as insulin resistance. Aerobic training is often recommended as a strategy to induce weight loss, with a greater impact of high-intensity levels on cardiovascular function and insulin sensitivity, and a greater impact of moderate-intensity levels on fat oxidation. Anaerobic high-intensity (supramaximal) interval training has been advocated to improve cardiovascular function, insulin sensitivity and fat oxidation. However, obese individuals tend to have a lower tolerance of high-intensity exercise due to discomfort. Furthermore, some obese individuals may compensate for the increased energy expenditure by eating more and/or becoming less active. Recently, both moderate- and high-intensity aerobic interval training have been advocated as alternative approaches. However, it is still uncertain as to which approach is more effective in terms of increasing fat oxidation given the issues with levels of fitness and motivation, and compensatory behaviours. Accordingly, the objectives of this thesis were to compare the influence of moderate- and high-intensity interval training on fat oxidation and eating behaviour in overweight/obese men. Two exercise interventions were undertaken by 10-12 overweight/obese men to compare their responses to study variables, including fat oxidation and eating behaviour during moderate- and high-intensity interval training (MIIT and HIIT). The acute training intervention was a methodological study designed to examine the validity of using exercise intensity from the graded exercise test (GXT) - which measured the intensity that elicits maximal fat oxidation (FATmax) - to prescribe interval training during 30-min MIIT. The 30-min MIIT session involved 5-min repetitions of workloads 20% below and 20% above the FATmax. The acute intervention was extended to involve HIIT in a cross-over design to compare the influence of MIIT and HIIT on eating behaviour using subjective appetite sensation and food preference through the liking and wanting test. The HIIT consisted of 15-sec interval training at 85 %VO2peak interspersed by 15-sec unloaded recovery, with a total mechanical work equal to MIIT. The medium term training intervention was a cross-over 4-week (12 sessions) MIIT and HIIT exercise training with a 6-week detraining washout period. The MIIT sessions consisted of 5-min cycling stages at ±20% of mechanical work at 45 %VO2peak, and the HIIT sessions consisted of repetitive 30-sec work at 90 %VO2peak and 30-sec interval rests, during identical exercise sessions of between 30 and 45 mins. Assessments included a constant-load test (45 %VO2peak for 45 mins) followed by 60-min recovery at baseline and the end of 4-week training, to determine fat oxidation rate. Participants’ responses to exercise were measured using blood lactate (BLa), heart rate (HR) and rating of perceived exertion (RPE) and were measured during the constant-load test and in the first intervention training session of every week during training. Eating behaviour responses were assessed by measuring subjective appetite sensations, liking and wanting and ad libitum energy intake. Results of the acute intervention showed that FATmax is a valid method to estimate VO2 and BLa, but is not valid to estimate HR and RPE in the MIIT session. While the average rate of fat oxidation during 30-min MIIT was comparable with the rate of fat oxidation at FATmax (0.16 ±0.09 and 0.14 ±0.08 g/min, respectively), fat oxidation was significantly higher at minute 25 of MIIT (P≤0.01). In addition, there was no significant difference between MIIT and HIIT in the rate of appetite sensations after exercise, but there was a tendency towards a lower rate of hunger after HIIT. Different intensities of interval exercise also did not affect explicit liking or implicit wanting. Results of the medium-term intervention indicated that current interval training levels did not affect body composition, fasting insulin and fasting glucose. Maximal aerobic capacity significantly increased (P≤0.01) (2.8 and 7.0% after MIIT and HIIT respectively) during GXT, and fat oxidation significantly increased (P≤0.01) (96 and 43% after MIIT and HIIT respectively) during the acute constant-load exercise test. RPE significantly decreased after HIIT greater than MIIT (P≤0.05), and the decrease in BLa was greater during the constant-load test after HIIT than MIIT, but this difference did not reach statistical significance (P=0.09). In addition, following constant-load exercise, exercise-induced hunger and desire to eat decreased after HIIT greater than MIIT but were not significant (p value for desire to eat was 0.07). Exercise-induced liking of high-fat sweet (HFSW) and high-fat non-sweet (HFNS) foods increased after MIIT and decreased after HIIT (p value for HFNS was 0.09). The intervention explained 12.4% of the change in fat intake (p = 0.07). This research is significant in that it confirmed two points in the acute study. While the rate of fat oxidation increased during MIIT, the average rate of fat oxidation during 30-min MIIT was comparable with the rate of fat oxidation at FATmax. In addition, manipulating the intensity of acute interval exercise did not affect appetite sensations and liking and wanting. In the medium-term intervention, constant-load exercise-induced fat oxidation significantly increased after interval training, independent of exercise intensity. In addition, desire to eat, explicit liking for HFNS and fat intake collectively confirmed that MIIT is accompanied by a greater compensation of eating behaviour than HIIT. Findings from this research will assist in developing exercise strategies to provide obese men with various training options. In addition, the finding that overweight/obese men expressed a lower RPE and decreased BLa after HIIT compared with MIIT is contrary to the view that obese individuals may not tolerate high-intensity interval training. Therefore, high-intensity interval training can be advocated among the obese adult male population. Future studies may extend this work by using a longer-term intervention.

Effect of interval training intensity on fat oxidation, blood lactate and the rate of perceived exertion in obese men

Purpose The objectives of this study were to examine the effect of 4-week moderate- and high-intensity interval training (MIIT and HIIT) on fat oxidation and the responses of blood lactate (BLa) and rating of perceived exertion (RPE). Methods Ten overweight/obese men (age = 29 ±3.7 years, BMI = 30.7 ±3.4 kg/m2) participated in a cross-over study of 4-week MIIT and HIIT training. The MIIT training sessions consisted of 5-min cycling stages at mechanical workloads 20% above and 20% below 45%VO2peak. The HIIT sessions consisted of intervals of 30-s work at 90%VO2peak and 30-s rest. Pre- and post-training assessments included VO2max using a graded exercise test (GXT) and fat oxidation using a 45-min constant-load test at 45%VO2max. BLa and RPE were also measured during the constant-load exercise test. Results There were no significant changes in body composition with either intervention. There were significant increases in fat oxidation after MIIT and HIIT (p ≤ 0.01), with no effect of intensity. BLa during the constant-load exercise test significantly decreased after MIIT and HIIT (p ≤ 0.01), and the difference between MIIT and HIIT was not significant (p = 0.09). RPE significantly decreased after HIIT greater than MIIT (p ≤ 0.05). Conclusion Interval training can increase fat oxidation with no effect of exercise intensity, but BLa and RPE decreased after HIIT to greater extent than MIIT.

Scaffolding techniques : a teacher training for cooperative learning in Thailand primary education

Using cooperative learning in classrooms promotes academic achievement, communication skills, problem-solving, social skills and student motivation. Yet it is reported that cooperative learning as a Western educational concept may be ineffective in Asian cultural contexts. The study aims to investigate the utilisation of scaffolding techniques for cooperative learning in Thailand primary mathematics classes. A teacher training program was designed to foster Thai primary school teachers’ cooperative learning implementation. Two teachers participated in this experimental program for one and a half weeks and then implemented cooperative learning strategies in their mathematics classes for six weeks. The data collected from teacher interviews and classroom observations indicates that the difficulty or failure of implementing cooperative learning in Thailand education may not be directly derived from cultural differences. Instead, it does indicate that Thai culture can be constructively merged with cooperative learning through a teacher training program and practices of scaffolding techniques.

Training Researchers : Visits and Training Schools

COST IS0801, like all COST Actions, provided many opportunities for training of early career researchers, and initiation of new research projects. Some of these were supported by short-term visits of usually one or two weeks, up to a maximum of three months, by an Action member to another venue, for purposes that supported the overall aims of the Action. The first part of this chapter provides some description of these, illustrated by a number of case studies. The second part of the chapter overviews the organization and outcome of two Training Schools for early career researchers, one in Australia entitled Research to policy and practice: Innovation and sustainability in cyberbullying prevention, and one in Finland, entitled Adolescents and Social Media: Guidelines and Coping Strategies for Cyberbullying. The organization of these Training Schools, the educational approaches used, and their evaluation and impact, will be summarised.

Alcohol ingestion impairs maximal post-exercise rates of myofibrillar protein synthesis following a single bout of concurrent training

Introduction The culture in many team sports involves consumption of large amounts of alcohol after training/competition. The effect of such a practice on recovery processes underlying protein turnover in human skeletal muscle are unknown. We determined the effect of alcohol intake on rates of myofibrillar protein synthesis (MPS) following strenuous exercise with carbohydrate (CHO) or protein ingestion. Methods In a randomized cross-over design, 8 physically active males completed three experimental trials comprising resistance exercise (8×5 reps leg extension, 80% 1 repetition maximum) followed by continuous (30 min, 63% peak power output (PPO)) and high intensity interval (10×30 s, 110% PPO) cycling. Immediately, and 4 h post-exercise, subjects consumed either 500 mL of whey protein (25 g; PRO), alcohol (1.5 g·kg body mass−1, 12±2 standard drinks) co-ingested with protein (ALC-PRO), or an energy-matched quantity of carbohydrate also with alcohol (25 g maltodextrin; ALC-CHO). Subjects also consumed a CHO meal (1.5 g CHO·kg body mass−1) 2 h post-exercise. Muscle biopsies were taken at rest, 2 and 8 h post-exercise. Results Blood alcohol concentration was elevated above baseline with ALC-CHO and ALC-PRO throughout recovery (P<0.05). Phosphorylation of mTORSer2448 2 h after exercise was higher with PRO compared to ALC-PRO and ALC-CHO (P<0.05), while p70S6K phosphorylation was higher 2 h post-exercise with ALC-PRO and PRO compared to ALC-CHO (P<0.05). Rates of MPS increased above rest for all conditions (~29–109%, P<0.05). However, compared to PRO, there was a hierarchical reduction in MPS with ALC-PRO (24%, P<0.05) and with ALC-CHO (37%, P<0.05). Conclusion We provide novel data demonstrating that alcohol consumption reduces rates of MPS following a bout of concurrent exercise, even when co-ingested with protein. We conclude that alcohol ingestion suppresses the anabolic response in skeletal muscle and may therefore impair recovery and adaptation to training and/or subsequent performance.

A 3D virtual medical imaging suite : bridging the academic-clinical training boundary

Aims The Medical Imaging Training Immersive Environment (MITIE) system is a recently developed virtual reality (VR) platform that allows students to practice a range of medical imaging techniques. The aim of this pilot study was to harvest user feedback about the educational value of the application and inform future pedagogical development. This presentation explores the use of this technology for skills training and blurring the boundaries between academic learning and clinical skills training. Background MITIE is a 3D VR environment that allows students to manipulate a patient and radiographic equipment in order to produce a VR-generated image for comparison with a gold standard. As with VR initiatives in other health disciplines (1-6) the software mimics clinical practice as much as possible and uses 3D technology to enhance immersion and realism. The software was developed by the Medical Imaging Course Team at a provider University with funding from a Health Workforce Australia “Simulated Learning Environments” grant. Methods Over 80 students undertaking the Bachelor of Medical Imaging Course were randomised to receive practical experience with either MITIE or radiographic equipment in the medical radiation laboratory. Student feedback about the educational value of the software was collected and performance with an assessed setup was measured for both groups for comparison. Ethical approval for the project was provided by the university ethics panel. Results This presentation provides qualitative analysis of student perceptions relating to satisfaction, usability and educational value as well as comparative quantitative performance data. Students reported high levels of satisfaction and both feedback and assessment results confirmed the application’s significance as a pre-clinical training tool. There was a clear emerging theme that MITIE could be a useful learning tool that students could access to consolidate their clinical learning, either during their academic timetables or their clinical placement. Conclusion Student feedback and performance data indicate that MITIE has a valuable role to play in the clinical skills training for medical imaging students both in the academic and the clinical environment. Future work will establish a framework for an appropriate supporting pedagogy that can cross the boundary between the two environments. This project was possible due to funding made available by Health Workforce Australia.

Exploring potential factors leading to effective training : an exclusive study on commercial banks in Cambodia

Purpose A previous study found that the quality of education in Cambodia is poor compared to other developing countries. However, the working performance of commercial banks in Cambodia is high. It was speculated that effective training was the main factor underlying this contradiction. Therefore, the main purpose of this article is to explore the elements of training conducted by commercial banks in Cambodia and to examine their relationship with training effectiveness. Design/methodology/approach The research focuses on six factors: training needs assessment; training program; flexibility of training; self-efficacy; social support; and transfer of knowledge. The data came in the form of questionnaires and desk research. A descriptive analytical approach is then used to describe these six factors. Findings The banking industry in Cambodia offers very effective training to its employees. It is also worth noting that more than 80 percent of employees are satisfied with the training, despite few attempts on the part of management to elicit opinions from employees on what training methods should be employed. Research limitations/implications As research studies involving Cambodia are relatively rare, it was difficult for to gather primary data. Because of this limitation and the purpose of this study, descriptive data interpretation was employed. Practical implications – Even though training can make up for poor education, it is only a short-term solution. In the long term, education needs to be enhanced to increase working performance. Originality/value This research provides a good framework for commercial banks in other developing countries to compare. A cross-cultural study is also proposed for future research.

Post-exercise cold water immersion attenuates acute anabolic signalling and long-term adaptations in muscle to strength training

We investigated functional, morphological and molecular adaptations to strength training exercise and cold water immersion (CWI) through two separate studies. In one study, 21 physically active men strength trained for 12 weeks (2 d⋅wk–1), with either 10 min of CWI or active recovery (ACT) after each training session. Strength and muscle mass increased more in the ACT group than in the CWI group (P<0.05). Isokinetic work (19%), type II muscle fibre cross-sectional area (17%) and the number of myonuclei per fibre (26%) increased in the ACT group (all P<0.05) but not the CWI group. In another study, nine active men performed a bout of single-leg strength exercises on separate days, followed by CWI or ACT. Muscle biopsies were collected before and 2, 24 and 48 h after exercise. The number of satellite cells expressing neural cell adhesion molecule (NCAM) (10−30%) and paired box protein (Pax7)(20−50%) increased 24–48 h after exercise with ACT. The number of NCAM+ satellitecells increased 48 h after exercise with CWI. NCAM+- and Pax7+-positivesatellite cell numbers were greater after ACT than after CWI (P<0.05). Phosphorylation of p70S6 kinaseThr421/Ser424 increased after exercise in both conditions but was greater after ACT (P<0.05). These data suggest that CWI attenuates the acute changes in satellite cell numbers and activity of kinases that regulate muscle hypertrophy, which may translate to smaller long-term training gains in muscle strength and hypertrophy. The use of CWI as a regular post-exercise recovery strategy should be reconsidered.

End-of-life care practices of critical care nurses: A national cross-sectional survey

Background: The critical care context presents important opportunities for nurses to deliver skilled, comprehensive care to patients at the end of life and their families. Limited research has identified the actual end-of-life care practices of critical care nurses. Objective: To identify the end-of-life care practices of critical care nurses. Design: A national cross-sectional online survey. Methods: The survey was distributed to members of an Australian critical care nursing association and 392 critical care nurses (response rate 25%) completed the survey. Exploratory factor analysis using principal axis factoring with oblique rotation was undertaken on survey responses to identify the domains of end-of-life care practice. Descriptive statistics were calculated for individual survey items. Results: Exploratory factor analysis identified six domains of end-of-life care practice: information sharing, environmental modification, emotional support, patient and family centred decision-making, symptom management and spiritual support. Descriptive statistics identified a high level of engagement in information sharing and environmental modification practices and less frequent engagement in items from the emotional support and symptom management practice areas. Conclusions: The findings of this study identified domains of end-of-life care practice, and critical care nurse engagement in these practices. The findings highlight future training and practice development opportunities, including the need for experiential learning targeting the emotional support practice domain. Further research is needed to enhance knowledge of symptom management practices during the provision of end-of-life care to inform and improve practice in this area.

Airway Inflammation and Bronchial Hyperresponsiveness in Elite Cross-Country Skiers and in Patients with Newly Diagnosed Asthma: A Bronchial Biopsy Study

The objective of these studies was to evaluate possible airway inflammation and remodeling at the bronchial level in cross-country skiers without a prior diagnosis of asthma, and relate the findings to patients with mild chronic asthma and patients with newly diagnosed asthma. We also studied the association of airway inflammatory changes and bronchial hyperresponsivess (BHR), and treatment effects in cross-country skiers and in patients with newly diagnosed asthma. Bronchial biopsies were obtained from the subjects by flexible bronchoscopy, and the inflammatory cells (eosinophils, mast cells, T-lymphocytes, macrophages, and neutrophils) were identified by immunohistochemistry. Tenascin (Tn) immunoreactivity in the bronchial basement membrane (BM) was identified by immunofluorescence staining. Lung function was measured with spirometry, and BHR was assessed by methacholine (skiers) or histamine (asthmatics) challenges. Skiers with BHR and asthma-like symptoms were recruited to a drug-intervention study. Skiers were given treatment (22 weeks) with placebo or budesonide (400 µg bid). Patients with newly diagnosed asthma were given treatment for 16 weeks with placebo, salmeterol (SLM) (50 µg bid), fluticasone propionate (FP) (250 µg bid), or disodium cromoglicate (DSCG) (5 mg qid). Bronchial biopsies were obtained at baseline and at the end of the treatment period. In the skiers a distinct airway inflammation was evident. In their bronchial biopsy specimens, T-lymphocyte, macrophage, and eosinophil counts were, respectively greater by 43-fold (P<0.001), 26-fold (P<0.001, and 2-fold (P<0.001) in skiers, and by 70-fold (p>0.001), 63-fold (P<0.001), and 8-fold (P<0.001) in asthmatic subjects than in controls. In skiers, neutrophil counts were more than 2-fold greater than in asthmatic subjects (P<0.05). Tn expression was higher in skiers than in controls and lower in skiers than in mild asthmatics. No significant changes were seen between skiers with or without BHR in the inflammatory cell counts or Tn expression. Treatment with inhaled budesonide did not attenuate asthma-like symptoms, the inflammatory cell infiltration, or BM Tn expression in the skiers. In newly diagnosed asthmatic patients, SLM, FP, and DSCG reduced asthma symptoms, and need for rescue medication (P<0.04). BHR was reduced by doubling doses 2.78, 5.22, and 1.35 respectively (all P<0.05). SLM and placebo had no effect on cell counts or Tn expression. FP and DSCG reduced eosinophil counts in the bronchial biopsy specimens (P<0.02 and <0.048, respectively). No significant change in tenascin expression appeared in any treatment group. Regarding to atopy, no significant differences existed in the inflammatory cell counts in the bronchial mucosa of subjects with newly diagnosed asthma or in elite cross country skiers. Tn expression in the BM was significantly higher in atopic asthma than in those with nonatopic asthma. Airway inflammation occurred in elite cross-country skiers with and without respiratory symptoms or BHR. Their inflammatory cell pattern differed from that in asthma. Infiltration with eosinophils, macrophages, and mast cells was milder, but lymphocyte counts did not differ from counts in asthmatic airways. Neutrophilic infiltration was more extensive in skiers than in asthmatics. Remodeling took place in the skiers’ airways, as reflected by increased expression of BM tenascin These inflammatory changes and Tn expression may be caused by prolonged exposure of the lower airways to inadequately humidified cold air. In skiers inflammatory changes and remodeling were not reversed with anti-inflammatory treatment. In contrast, in patients with newly diagnosed asthma, anti-inflammatory treatment did attenuate eosinophilic inflammation in the bronchial mucosa. In skiers, anti-inflammatory treatment did not attenuate BHR as it did in asthmatic patients. The BHR in skiers was attenuated spontaneously during placebo treatment, with no difference from budesonide treatment. Lower training intensity during the treatment period may explain this spontaneous decrease in BHR. The origin of BHR probably differs in skiers and in asthmatics. No significant association between BHR and inflammatory cell counts or between BHR and Tn expression was evident in cross-country skiers or asthmatic subjects. Airway remodeling differed between atopic and nonatopic asthma. As opposed to nonatopic asthma, Tn expression was higher in atopic asthma and is related to inflammatory cell densities.

Cross-Lingual spoken language understanding from unaligned data using discriminative classification models and machine translation

This paper investigates several approaches to bootstrapping a new spoken language understanding (SLU) component in a target language given a large dataset of semantically-annotated utterances in some other source language. The aim is to reduce the cost associated with porting a spoken dialogue system from one language to another by minimising the amount of data required in the target language. Since word-level semantic annotations are costly, Semantic Tuple Classifiers (STCs) are used in conjunction with statistical machine translation models both of which are trained from unaligned data to further reduce development time. The paper presents experiments in which a French SLU component in the tourist information domain is bootstrapped from English data. Results show that training STCs on automatically translated data produced the best performance for predicting the utterance's dialogue act type, however individual slot/value pairs are best predicted by training STCs on the source language and using them to decode translated utterances. © 2010 ISCA.

Unsupervised intra-lingual and cross-lingual speaker adaptation for HMM-based speech synthesis using two-pass decision tree construction

Hidden Markov model (HMM)-based speech synthesis systems possess several advantages over concatenative synthesis systems. One such advantage is the relative ease with which HMM-based systems are adapted to speakers not present in the training dataset. Speaker adaptation methods used in the field of HMM-based automatic speech recognition (ASR) are adopted for this task. In the case of unsupervised speaker adaptation, previous work has used a supplementary set of acoustic models to estimate the transcription of the adaptation data. This paper first presents an approach to the unsupervised speaker adaptation task for HMM-based speech synthesis models which avoids the need for such supplementary acoustic models. This is achieved by defining a mapping between HMM-based synthesis models and ASR-style models, via a two-pass decision tree construction process. Second, it is shown that this mapping also enables unsupervised adaptation of HMM-based speech synthesis models without the need to perform linguistic analysis of the estimated transcription of the adaptation data. Third, this paper demonstrates how this technique lends itself to the task of unsupervised cross-lingual adaptation of HMM-based speech synthesis models, and explains the advantages of such an approach. Finally, listener evaluations reveal that the proposed unsupervised adaptation methods deliver performance approaching that of supervised adaptation.