Optimal weight tuning method for unit selection cost functions in syllable based text-to-speech synthesis

This paper proposes a method for tuning the weights of unit selection cost functions in syllable based text-to-speech (TTS) synthesis system. In this work, unit selection cost functions, namely target cost and concatenation cost, are designed appropriate to syllables. The method tunes the weights in such a way that perceptual preference patterns are appropriately considered while selecting the units. The method uses genetic algorithm to derive the optimal weights. Fitness function is designed to map perceptual preference patterns into weights of unit selection cost functions. The effectiveness of proposed method is evaluated by both subjective and objective measures. From the results, it is observed that the derived optimal weights can synthesize good quality speech compared to manually tuned weights.     

Optimal Utterance Selection for Unit Selection Speech Synthesis Databases

This paper describes techniques to find an optimal data set for building high quality unit-selection speech synthesis inventories. As the quality of unit-selection speech synthesis is dependent on the coverage of the database used in the selection, it is important to select the right data to record. In this paper we describe some simple techniques as well as a more complex acoustic modeling technique based on the database speaker's acoustic characteristics. Result of a simple evaluation procedure are presented justifying the technique.     

基于用户偏好的权重搜索及告警选择方法

用户在现有交互方式下选择最为严重的告警时完全依据其个人偏好,而未考虑处理不同告警所需成本的差异性问题.为此,提出一种基于用户偏好的权重搜索及告警选择方法.挖掘用户对不同严重程度告警的偏好值,针对问题的复杂性建立评估函数,并给出偏好权重的选择策略.对不同告警及其对应的用户偏好权重建立效用函数,确定需优先解决的告警,并在成本约束下完成基于用户偏好的告警选择,提高告警处理效率.实验结果表明,该方法能够合理有效地做出告警选择,与基于背包式和设定阈值的方法相比,其告警选择的表现更优.     

基于评分和项目特征的群组推荐方法

针对现有群组推荐方法准确率偏低的问题,提出了一种基于评分和项目特征相结合的方法。首先综合时间因素对评分的影响和项目领域特征,利用改进的TF-IDF方法构建成员在各个特征上的偏好模型。其次考虑群体用户间的相互作用,从项目特征属性均值相似性权重和特征属性频度权重两个方面来得到群体偏好模型。最后计算群组在项目特征上和评分上的综合相似度,进行预测评分并推荐。通过在MovieLens数据集上进行实验,表明本方法比现有方法的准确率有明显提高。     

On-line experimental methods to evaluate text-to-speech (TTS) synthesis: effects of voice gender and signal quality on intelligibility,naturalness and preference

Three experiments are reported that use new experimental methods for the evaluation of text-to-speech (TTS) synthesis from the user's perspective. Experiment 1, using sentence stimuli, and Experiment 2, using discrete “call centre” word stimuli, investigated the effect of voice gender and signal quality on the intelligibility of three concatenative TTS synthesis systems. Accuracy and search time were recorded as on-line, implicit indices of intelligibility during phoneme detection tasks. It was found that both voice gender and noise affect intelligibility. Results also indicate interactions of voice gender, signal quality, and TTS synthesis system on accuracy and search time. In Experiment 3 the method of paired comparisons was used to yield ranks of naturalness and preference. As hypothesized, preference and naturalness ranks were influenced by TTS system, signal quality and voice, in isolation and in combination. The pattern of results across the four dependent variables – accuracy, search time, naturalness, preference – was consistent. Natural speech surpassed synthetic speech, and TTS system C elicited relatively high scores across all measures. Intelligibility, judged naturalness and preference are modulated by several factors and there is a need to tailor systems to particular commercial applications and environmental conditions.     

A rough set-based interval-valued intuitionistic fuzzy conceptual design decision approach with considering diverse customer preference distribution

Conceptual design evaluation plays a crucial role in new product development (NPD) and determines the quality of downstream design activities. Currently, most existing methods focus on fuzzy quantitative the evaluation information of multi-objectives in conceptual schemes selection. However, the above process ignores the various customers' preferences for each scheme under the evaluation objective, causing inconsistent preference weights in the various schemes, which cannot guarantee the market value of the optimal scheme. Furthermore, the ambiguous attitude from experts in the early design stage is not well taken into account. To this end, a conceptual scheme decision model with considering diverse customer preference distribution based on interval-valued intuitionistic fuzzy set (IVIFS) is proposed. The model is divided into three parts. Firstly, the initial decision matrix of multi-experts concerning the qualitative and quantitative design attributes is constructed based on intuitionistic fuzzy sets, and then the IFS decision matrix with interval boundaries is formed by using rough set technology. Secondly, the mapping model of design attribute to customer preference is constructed, and then the demand preference strategy implied by design attribute is judged. Thirdly, based on the demand preference strategy, the preferences’ weights for each scheme are calculated. Next, integrating the evaluation data with the same preference in the scheme, the comprehensive satisfaction of the scheme is obtained through IVIFS weighted aggregation operator, and then the optimal scheme is decided. Eventually, a case study of mobile phone form feature schemes is further employed to verify the proposed decision model, and results are sensitivity analyzed and compared.     

Emotional Vietnamese Speech Synthesis Using Style-Transfer Learning

In recent years, speech synthesis systems have allowed for the production of very high-quality voices. Therefore, research in this domain is now turning to the problem of integrating emotions into speech. However, the method of constructing a speech synthesizer for each emotion has some limitations. First, this method often requires an emotional-speech data set with many sentences. Such data sets are very time-intensive and labor-intensive to complete. Second, training each of these models requires computers with large computational capabilities and a lot of effort and time for model tuning. In addition, each model for each emotion failed to take advantage of data sets of other emotions. In this paper, we propose a new method to synthesize emotional speech in which the latent expressions of emotions are learned from a small data set of professional actors through a Flowtron model. In addition, we provide a new method to build a speech corpus that is scalable and whose quality is easy to control. Next, to produce a high-quality speech synthesis model, we used this data set to train the Tacotron 2 model. We used it as a pre-trained model to train the Flowtron model. We applied this method to synthesize Vietnamese speech with sadness and happiness. Mean opinion score (MOS) assessment results show that MOS is 3.61 for sadness and 3.95 for happiness. In conclusion, the proposed method proves to be more effective for a high degree of automation and fast emotional sentence generation, using a small emotional-speech data set.     

A new method of obtaining the priority weights from an interval fuzzy preference relation

In analyzing a multiple criteria decision-making problem, the decision maker may express her/his opinions as an interval fuzzy or multiplicative preference relation. Then it is an interesting and important issue to investigate the consistency of the preference relations and obtain the reliable priority weights. In this paper, a new consistent interval fuzzy preference relation is defined, and the corresponding properties are derived. The transformation formulae between interval fuzzy and multiplicative preference relations are further given, which show that two preference relations, consistent interval fuzzy and multiplicative preference relations, can be transformed into each other. Based on the transformation formula, the definition of acceptably consistent interval fuzzy preference relation is given. Furthermore a new algorithm for obtaining the priority weights from consistent or inconsistent interval fuzzy preference relations is presented. Finally, three numerical examples are carried out to compare the results using the proposed method with those using other existing procedures. The numerical results show that the given procedure is feasible, effective and not requisite to solve any mathematical programing.     

Supplier selection and order allocation with green criteria: An MCDM and multi-objective optimization approach

This research provides a decision-making tool to solve a multi-period green supplier selection and order allocation problem. The tool contains three integrated components. First, fuzzy TOPSIS (technique for order of preference by similarity to ideal solution) is used to assign two preference weights to each potential supplier according to two sets of criteria taken separately: traditional and green. Second, top management uses an analytic hierarchy process (AHP) to assign a global importance weight to each of the two sets of criteria based on the strategy of the company and independently of the potential suppliers. Third, for each supplier, the preference weight obtained from fuzzy TOPSIS regarding the traditional criteria is then multiplied by the global importance weight of the set of traditional criteria. The same is done for the green criteria. The two combined preference weights obtained for each supplier are then used in addition to total cost to select the best suppliers and to allocate orders using multi-period bi-objective and multi-objective optimization. The mathematical models are solved using the weighted comprehensive criterion method and the branch-and-cut algorithm. The approach of this research has a major advantage: it provides top management with flexibility in giving more or less importance weight to green or traditional criteria regardless of the number of criteria in each category through the use of AHP, which reduces the effect of the number of criteria on the preference weight of the suppliers. Contrary to the case in which each supplier is evaluated on the basis of all criteria at the same time, the proposed approach would not necessarily result in a supplier with poor green performance being ranked among the best for a situation in which the number of green criteria is smaller than the number of traditional criteria. In this case, the final ranking would mainly depend on the global weight of the green criteria set given by the top management using AHP as well as on the ranking of the supplier in terms of green criteria obtained from fuzzy TOPSIS. Extensive numerical experiments are conducted in which the bi-objective and multi-objective models are compared and the effect of the separation between green and traditional criteria is investigated. The results show that the two optimization approaches provide very close solutions, which leads to a preference for the bi-objective approach because of its lower computation time. Moreover, the results confirm the flexibility of the proposed approach and show that combining all criteria in one set is a special case. Finally, a time study is performed, which shows that the bi-objective integer linear programming model has a polynomial computation time.     

Application of Genetic Algorithm in unit selection for Malay speech synthesis system

Corpus based speech synthesis can produce high quality synthetic speech due to it high sensitivity to unit context. Large speech database is embedded in synthesis system and search algorithm (unit selection) is needed to search for the optimal unit sequence. Speech feature which served as target cost is estimated from the input text. The acoustic parameters which served as join cost are derived from mel frequency cepstral coefficients (MFCCs) and Euclidean distance. In this paper, a new method which is Genetic Algorithm is proposed to search for optimal unit sequence. Genetic Algorithm (GA) is a population based search algorithm that is based on the biological principles of selection, reproduction, crossover and mutation. It is a stochastic search algorithm for solving optimization problem. The speech unit sequence that has minimum join cost will be synthesized into complete waveform data.     

On compatibility of uncertain additive linguistic preference relations and its application in the group decision making

We develop a new compatibility for the uncertain additive linguistic preference relations and utilize it to determine the optimal weights of experts in the group decision making (GDM). Based on some operational laws for the uncertain additive linguistic preference labels, we propose some new concepts of the compatibility degree and acceptable compatibility index for the two uncertain additive linguistic preference relations. We also prove the properties that the synthetic preference relation is also of acceptable compatibility under the condition that additive linguistic preference relations provided by experts are all of acceptable compatibility with the specific linguistic preference relation, which provides a theoretic basis for the application of the uncertain additive linguistic preference relations in the GDM. Furthermore, we establish a mathematical model to obtain the weights of experts based on the criterion of minimizing the compatibility in the GDM, and we discuss the solution to the model. Finally, we give a numerical example to make comparative analysis on compatibility index using the optimal experts’ weights approach and the equal experts’ weights approach, which indicates that the model is feasible and effective.     

A New Korean Corpus-Based Text-to-Speech System

This paper describes a new Korean Text-to-Speech (TTS) system based on a large speech corpus. Conventional concatenative TTS systems still produce machine-like synthetic speech. The poor naturalness is caused by excessive prosodic modification using a small speech database. To cope with this problem, we utilized a dynamic unit selection method based on a large speech database without prosodic modification. The proposed TTS system adopts triphones as synthesis units. We designed a new sentence set maximizing phonetic or prosodic coverage of Korean triphones. All the utterances were segmented automatically into phonemes using a speech recognizer. With the segmented phonemes, we achieved a synthesis unit cost of zero if two synthesis units were placed consecutively in an utterance. This reduces the number of concatenating points that may occur due to concatenating mismatches. In this paper, we present data concerning the realization of major prosodic variations through a consideration of prosodic phrase break strength. The phrase break was divided into four kinds of strength based on pause length. Using phrase break strength, triphones were further classified to reflect major prosodic variations. To predict phrase break strength on texts, we adopted an HMM-like Part-of-Speech (POS) sequence model. The performance of the model showed 73.5% accuracy for 4-level break strength prediction. For unit selection, a Viterbi beam search was performed to find the most appropriate triphone sequence, which has the minimum continuation cost of prosody and spectrum at concatenating boundaries. From the informal listening test, we found that the proposed Korean corpus-based TTS system showed better naturalness than the conventional demisyllable-based one.     

OWA weight updating in repeated decision making under the influence of additional information

This paper is concerned with the problem of evaluating the outcomes of comparing alternatives in repeated decision making under uncertainty. In order to apply the ordered weighted averaging (OWA) operator to this problem, we propose to update the weights each time the decision maker obtains additional information on environmental behavior, such as feedback and finite number of probabilistic distributions. An attempt was made to develop an algorithm of OWA weigh correction in case of decision making under risk and ignorance, assuming that the decision maker sets up a preference relation on state of nature set. Also a new approach to OWA weights calculation is suggested based on the assumption that weights form a geometric progression, which is increasing for a risky decision maker and decreasing for a nonrisky one. © 2011 Wiley Periodicals, Inc.     

A systematic procedure to obtain a preferable and robust ranking of units

This paper introduces one systematic procedure for the manager of an organization to assess units under its governance using multiple performance indices. The goal of this systematic procedure is to assist the manager in obtaining a preferable and robust ranking result for units. In this procedure, for all units, one common set of weights attached to the performance indices is determined in order to maximize the group's comprehensive score. Then, using the common set of weights, each unit's comprehensive score is evaluated and compared for ranking. In order to obtain the preferable ranking, the manager's subjective preference is considered and formulated by the virtual weights restrictions while determining the common weights in the procedure. The procedure is applied in order to obtain a robust ranking by modifying the boundary of the feasible region of virtual weights restrictions in each assessment. The final statistical ranking of all assessments provides the manager with one robust ranking, which is invariant in different feasible regions of virtual weights restrictions in the numerical example.     

A fast neural net training algorithm and its application to speech classification

This paper describes a fast training algorithm for feedforward neural nets, as applied to a two-layer neural network to classify segments of speech as voiced, unvoiced, or silence. The speech classification method is based on five features computed for each speech segment and used as input to the network. The network weights are trained using a new fast training algorithm which minimizes the total least squares error between the actual output of the network and the corresponding desired output. The iterative training algorithm uses a quasi-Newtonian error-minimization method and employs a positive-definite approximation of the Hessian matrix to quickly converge to a locally optimal set of weights. Convergence is fast, with a local minimum typically reached within ten iterations; in terms of convergence speed, the algorithm compares favorably with other training techniques. When used for voiced-unvoiced-silence classification of speech frames, the network performance compares favorably with current approaches. Moreover, the approach used has the advantage of requiring no assumption of a particular probability distribution for the input features.     

Compatibility of interval fuzzy preference relations with the COWA operator and its application to group decision making

We develop a new compatibility for the interval fuzzy preference relations based on the continuous ordered weighted averaging (COWA) operator and use it to determine the weights of experts in group decision making (GDM). We define some concepts of the compatibility degree and the compatibility index for the two interval fuzzy preference relations based on the COWA operator. We study some desirable properties of the compatibility index and investigate the relationship between the each expert’s interval fuzzy preference relation and the synthetic interval fuzzy preference relation. The prominent characteristic of the compatibility index based on the COWA operator is that it can deal with the compatibility of all the arguments by using a controlled parameter considering the attitude of decision maker rather than the compatibility of the simply two points in intervals. To determine the experts’ weights in the GDM with the interval fuzzy preference relations, we propose an optimal model based on the criterion of minimizing the compatibility index. In the end, we give a numerical example to develop the new approach to GDM with interval fuzzy preference relations.     

A criteria weighting approach by combining fuzzy quality function deployment with relative preference relation

In past, fuzzy multi-criteria decision-making (FMCDM) models desired to find an optimal alternative from numerous feasible alternatives under fuzzy environment. However, researches seldom focused on determination of criteria weights, although they were also important components for FMCDM. In fact, criteria weights can be computed through extending quality function deployment (QFD) under fuzzy environment, i.e. fuzzy quality function deployment (FQFD). By FQFD, customer demanded qualities expressing the opinions of customers and service development capabilities presenting the opinions of experts can be integrated into criteria weights for FMCDM. However, deriving criteria weights in FQFD may be complex and different to multiply two fuzzy numbers in real world. To resolve the tie, we will combine FQFD with relative preference relation on FMCDM problems. With the relative preference relation on fuzzy numbers, it is not necessary multiplying two fuzzy numbers to derive criteria weights in FQFD. Alternatively, adjusted criteria weights will substitute for original criteria weights through relative preference relation. Obviously, adjusted criteria weights are clearly determined and then utilized in FMCDM models.     

A wise cost-effective supplying bandwidth policy for multilayer wireless cognitive networks

Inventory management is one of the most important research areas in Operations Research and Logistics. It mainly aims to efficiently manage inventories at different facilities (for example, warehouses and plants in Supply Chains (SCs)), minimizing the total cost and satisfying the service levels. Some exact inventories management approaches are successfully proposed and applied to different real scenarios, traditionally related to the SCs, even if the extreme versatility of these techniques could make them attractive to new challenging scenarios such as those related to telecommunications networks. Starting from this vision, the focus of this paper is to show the new benefits of applying an adaptive period inventory management policy to a wireless cognitive telecommunication scenario in which radio transmission resources are treated as short-term life time goods which supplies wisely in order to maximize both economic profit and quality of service offered to wireless users. The system behavior is tested using an agent-based simulator and computational results show that introducing this wise control on the bandwidth supplying mechanism guarantees a more reactive and effective telecommunication network, reaching a good compromise between the total profit and the service levels.     

Obtaining OWA operators starting from a linear order and preference quantifiers

The ordered weighted averaging operator has been widely studied for its practical use in decision problems. This operator has an associated weights vector with specific properties. Different variants have been developed to obtain it. Among these are those which use the order relationship between the criteria. This paper presents a method to obtain a weights vector, which has as inputs the weights vector obtained by the Borda–Kendall law and the quantified preference relation between the criteria given by the decision maker. Then, through a set of operations, the new weights vector is obtained; this vector is between the weights obtained by the Borda–Kendall law and the weighted average vector. In addition, the paper shows the properties that verify the vectors obtained by this method and its use is illustrated through an example. © 2012 Wiley Periodicals, Inc.     

Using LSTM neural networks for cross-lingual phonetic speech segmentation with an iterative correction procedure

This article describes experiments on speech segmentation using long short-term memory recurrent neural networks. The main part of the paper deals with multi-lingual and cross-lingual segmentation, that is, it is performed on a language different from the one on which the model was trained. The experimental data involves large Czech, English, German, and Russian speech corpora designated for speech synthesis. For optimal multi-lingual modeling, a compact phonetic alphabet was proposed by sharing and clustering phones of particular languages. Many experiments were performed exploring various experimental conditions and data combinations. We proposed a simple procedure that iteratively adapts the inaccurate default model to the new voice/language. The segmentation accuracy was evaluated by comparison with reference segmentation created by a well-tuned hidden Markov model-based framework with additional manual corrections. The resulting segmentation was also employed in a unit selection text-to-speech system. The generated speech quality was compared with the reference segmentation by a preference listening test.