Search

Language Identification (LID), a recommended initial step to Automatic Speech Recognition (ASR), is used to detect a spoken language from audio specimens. In state-of-the-art systems capable of multilingual speech processing, however, users have to explicitly set one or more languages before using them. LID, therefore, plays a very important role in situations where ASR based systems cannot parse the uttered language in multilingual contexts causing failure in speech recognition. We propose an attention based convolutional recurrent neural network (CRNN with Attention) that works on Mel-frequency Cepstral Coefficient (MFCC) features of audio specimens. Additionally, we reproduce some state-of-the-art approaches, namely Convolutional Neural Network (CNN) and Convolutional Recurrent Neural Network (CRNN), and compare them to our proposed method. We performed extensive evaluation on thirteen different Indian languages and our model achieves classification accuracy over 98%. Our LID model is robust to noise and provides 91.2% accuracy in a noisy scenario. The proposed model is easily extensible to new languages.

Adequate nutrition is one of the basic requirements for survival and is generally regarded as a core dimension in any evaluation of well-being. In the context of India, a country with high prevalence of poor nutrition, there is a dearth of nutrition studies with adequate coverage and comparability. Using primary data on food consumption from a village in a poorer state of India, we study the consumption of key nutrients, namely, calories, protein, carbohydrate, calcium and iron. Investigating the determinants of nutrition intake, we find evidence of consumption smoothing. By correcting for potential endogeneity in both expenditure and consumption of subsidised food through the Public Distribution System (PDS) we demonstrate a causal link from expenditure and the availability of subsidised food grains to nutritional intake. We also find evidence that other individual characteristics such as gender and occupation have an important role to play as determinants of nutrition.

This thesis aims at pushing the performances of a cold atom interferometer principally sensitive to rates of rotation in a particular axis. In our experiment, Cesium atoms are laser cooled, trapped and launched in a fountain configuration. According to the Sagnac effect, the sensitivity of the interferometer to rotation is proportional to the area enclosed by the interferometer arms. We use stimulated Raman transitions to split the atoms in two paths and to form a folded Mach-Zehnder-like interferometer architecture using four Raman pulses. With an interrogation time of the atoms of 800 ms, we achieve a Sagnac area as high as 11 cm^2. The thesis describes the improvements to the experimental setup to operate the gyroscope with such a high Sagnac area. A procedure for the relative alignment of the Raman beams at the microrad level is presented, which is critical to meet the interference condition of the cold atoms at the interferometer output. The characterization and mitigation of the vibration noise, affecting the gyroscope, is also demonstrated. We finally demonstrate a short term rotation stability of 160 nrad/s at 1 s and a long term stability of 1.8 nrad/s after 10 000 s of integration time. This stability level represents a factor 5 improvement compared to the previous SYRTE gyroscope experiment of 2009 and a factor 15 compared to other published results. The thesis work also presents a new method of interrogation to operate the gyroscope without dead times, which is important for various applications of cold atom sensors in inertial navigation, geophysics and in fundamental physics., Cette thèse a pour objet de repousser les performances d'un interféromètre à atomes froids principalement sensible aux rotations selon un axe particulier. Des atomes de Cesium sont refroidis par laser, piégés, et lancés verticalement selon une configuration en fontaine. La sensibilité du gyromètre repose sur l'effet Sagnac et est proportionnelle à l'aire physique qu'entourent les deux bras de l'interféromètre. Nous utilisons des transitions Raman stimulées pour séparer les ondes atomiques et former une géométrie d'interféromètre de type Mach-Zehnder replié. Avec un temps d'interrogation de 800 ms, nous parvenons à une aire physique de 11 cm^2. Le manuscrit décrit les améliorations apportées au dispositif expérimental pour faire fonctionner le gyromètre avec une telle aire Sagnac. Une procédure d'alignement relatif des faisceaux Raman au niveau du microrad est présentée et est particulièrement importante pour permettre aux ondes de matière d'interférer. La caractérisation des bruits de vibration impactant la sensibilité du gyromètre, ainsi que sa réjection sont également décrites. Nous démontrons une sensibilité de 160 nrad/s à 1 s, et une stabilité long terme de 1.8 nrad/s après 10 000 s d'intégration. Ce niveau de stabilité représente une amélioration d'un facteur 5 par rapport à la précédente expérience de gyromètre du SYRTE de 2009, et d'un facteur 15 par rapport aux autres résultats publiés. Cette thèse présente également une nouvelle méthode d'interrogation des atomes pour opérer le gyromètre sans temps morts, un aspect important pour diverses applications des capteurs à atomes froids en navigation inertielle, en géophysique et en physique fondamentale.

This paper presents a theoretical model to show how distributional concerns can engender social conflict. We have a two-period model that highlights the crucial role of future inequality. Equality of assets and income in the current period does not stop conflict from arising the anticipated future inequality is significant. Further we find that the impact of inequality on conflict is not straightforward. Societies with low levels of inequality show no conflict; groups engage in conflict only when inequality exceeds a certain threshold level. Additionally the model shows that the link between inequality and conflict may be non-monotonic.

There is a growing interest on dynamic and broader concepts of deprivation such as vulnerability, which takes in to account the destitution of individuals from future shocks. We use the framework of decision making under uncertainty to arrive at a new measure of vulnerability to poverty. We highlight the importance of current standard of living to better capture the notion of vulnerability. In conceptualizing the new class of measures of vulnerability we thus move beyond the standard expected poverty measures that is commonly found in the literature. We also axiomatically characterize the new class of measure and discuss some of it's properties.

In this paper we explore what impact, if any, government debts have on achieving the Millennium Development Goals for the Indian states. To fulfill the goals, national governments, especially in the developing world, have to undertake major investments in the social sector; but how much they will really be able to do so will depend on the conditions of their finances. For the Indian states we find that government investment in the social sector is extremely important to reduce poverty, but the government's debt burden is actually stopping several states from attaining the MDG targets. Although, in the medium term the impact of the debt on poverty is not very harmful, in the longer run it has a significant negative impact. Therefore for policy purposes reduction in debt should be given a priority.

Food insecurity and hunger have traditionally been measured by aggregate food supplies or by variables correlated with food insecurity. Because these measures often poorly reflect individuals’ true deprivation, economists have turned to surveys with direct questions about food insecurity. Using these surveys, households have then been classified into broad categories, a classification system which ignores the richness of the multiple questions. In this paper, we propose food insecurity measures, along the lines of the well established poverty measures, which incorporates this richness and allow us to reflect the depth and severity, in addition to the incidence, of food insecurity. Using these indices, we calculate the extent of food insecurity and hunger in the United States. Along with giving a richer picture of food insecurity in the US, these food insecurity measures demonstrates that the ordering of various demographic categories differs depends on the choice of measure. – deprivation ; food insecurity ; hunger ; poverty measures

This paper presents a simple model to show how distributional concerns can engender social conflict. We have a two period model, where the cost of conflict is endogenous in the sense that parties involved have full control over how much conflict they can create. We find that anticipated future inequality plays a crucial role in determining the level of conflict in the current period. The model also provides an explanation for why similar levels of inequality may exhibit drastically different levels of conflict. Further, we argue that the link between inequality and conflict may be non-monotonic. – conflict ; wealth inequality ; Nash bargaining