State of the art | EPFL Graph Search

The state of the art (SOTA, sometimes cutting edge or leading edge) refers to the highest level of general development, as of a device, technique, or scientific field achieved at a particular time. However, in some contexts it can also refer to a level of development reached at any particular time as a result of the common methodologies employed at the time. The term has been used since 1910, and has become both a common term in advertising and marketing, and a legally significant phrase with respect to both patent law and tort liability. In advertising, the phrase is often used to convey that a product is made with the best or latest available technology, but it has been noted that "the term 'state of the art' requires little proof on the part of advertisers", as it is considered mere puffery. The use of the term in patent law "does not connote even superiority, let alone the superlative quality the ad writers would have us ascribe to the term". Origin and history The c

Guessing the fingerings

Raphaël Nicolas Latty

On most musical instruments, especially on the guitar, it is possible to play the same note or chord in multiple ways. In this project, we develop a simple audio-based method to estimate the fingering of a note played on an acoustic guitar from a recording. Such a system has several applications, including musical training, musical education, or analysis of historical recordings. The advantage of our method with respect to the state-of-the-art is that it is simple and efficient: it uses a single feature, the inharmonicity coefficient. In spite of its simplicity, this method performs well in both synthetic and real audio recordings: the misclassification rate that we achieve is of 12.3 %.

2017

Language Independent Query by Example Spoken Term Detection

Dhananjay Ram

Language independent query-by-example spoken term detection (QbE-STD) is the problem of retrieving audio documents from an archive, which contain a spoken query provided by a user. This is usually casted as a hypothesis testing and pattern matching problem, also referred to as a ``zero-resource task'' since no specific training or lexical information is required to represent the spoken query. Thus it enables multilingual search on unconstrained speech without requiring a full speech recognition system. State-of-the-art solutions typically rely on Dynamic Time Warping (DTW) based template matching using phone posteriors features estimated by Deep Neural Networks (DNN). In this thesis, we aim at exploiting the low-dimensional subspace structure of speech signal, resulting from the constrained human speech production process. We exploit this subspace structure to improve over the state-of-the-art to (1) generate better phone or phonological posterior features, and (2) to improve the matching algorithm. To enhance phone posteriors, we learn the underlying phonetic subspaces in an unsupervised way, and use the sub-phonetic attributes to extract the phonological components in a supervised manner. To improve the matching algorithm, we model the subspaces of the spoken query using its phone posterior representation. The resulting model is used to compute distances between the subspaces of the query and the phone posteriors of each audio document. These distances are then used to detect occurrences of the spoken query, while also regularizing the DTW to improve the detection scores. In addition to optimizing different components of the state-of-the-art system, we propose a novel DNN-based QbE-STD system to provide an end-to-end learning framework. Towards that end, we replace the DTW based matching with a Convolutional Neural Network (CNN) architecture. We also learn multilingual features, aimed at obtaining language independent representation. Finally, we integrate the feature learning and CNN-based matching to jointly train and further improve the QbE-STD performance. We perform experiments using the challenging AMI meeting corpus (English), as well as multilingual datasets such as Spoken Web Search 2013 and Query by Example Search on Speech Task 2014, and show significant improvements over a very competitive state-of-the-art system.

EPFL2019

Integrating advanced discrete choice models in mixed integer linear optimization

Michel Bierlaire, Bernard Gendron, Meritxell Pacheco Paneque, Shadi Sharif Azadeh

The integration of customer behavioral models in operations research (OR) is appealing to operators and policy makers (the supply) because it provides a better understanding of the preferences of customers (the demand) while planning for their systems. These preferences are formalized with discrete choice models, which are the state-of-the-art for the mathematical modeling of demand, whereas mixed integer linear programming (MILP) models are considered to design and configure the systems. Notwithstanding the clear advantages of this integration, the complexity of discrete choice models leads to mathematical formulations that are highly nonlinear and nonconvex in the variables of interest, and therefore difficult to be included in MILP. In this paper, we present a general framework that overcomes these limitations by integrating in MILP advanced discrete choice models. A concrete application on benefit maximization from an operator selling services to a market is used to illustrate the employment of the framework. A case study from the recent literature is considered to perform various experiments, such as price differentiation by population segmentation. The results show that this approach is a powerful tool to configure systems based on the heterogeneous behavior of customers, and it allows to investigate advanced marketing strategies and business models.

2017

Language Independent Query by Example Spoken Term Detection

Dhananjay Ram

EPFL2019