Graph-based Semi-Supervised Learning in Acoustic Modeling for Automatic Speech Recognition

Thesis (Ph.D.)--University of Washington, 2016-06

Development & evaluation of different acoustic models for Malayalam continuous speech recognition

Performance of any continuous speech recognition system is dependent on the accuracy of its acoustic model. Hence, preparation of a robust and accurate acoustic model lead to satisfactory recognition performance for a speech recognizer. In acoustic modeling of phonetic unit, context information is of prime importance as the phonemes are found to vary according to the place of occurrence in a word. In this paper we compare and evaluate the effect of context dependent tied (CD tied) models, context dependent (CD) and context independent (CI) models in the perspective of continuous speech recognition of Malayalam language. The database for the speech recognition system has utterance from 21 speakers including 11 female and 10 males. Our evaluation results show that CD tied models outperforms CI models over 21%.

Estudi d'impacte acústic de les obres d'un intercanviador de metro

En el territori del Barcelonès les obres han estat a l’ordre del dia. Abans de l’inici de la crisi es van encetar i programar nombroses obres tan d’immobles com de grans infraestructures com la que aquí es presenta: un intercanviador de metro. L’alta densitat demogràfica d’aquesta regió ha fet que aquests treballs realitzats a les obres siguin poc compatibles amb el benestar i la salut de la població. Un dels principals factors causants del malestar als veïns és el soroll. Mitjançant estudis de modelització acústica, com el realitzat en aquest projecte, es pretén preveure i evitar que l’afecció acústica derivada de les obres sigui perjudicial per la persona per tal de compatibilitzar al màxim possible les feines de l’obra amb el benestar general. Mitjançant el programa de monitorització acústica Cadna‐A i nombroses mesures acústiques in situ s’ha pogut preveure la futura afecció sonora de 6 fases d’obra considerades com les més sorolloses i s’han proposat un seguit de mesures preventives i/o correctores per tal d’assolir uns nivells acústics assequibles amb la legislació vigent.

Automated Transcription System for Malayalam Language

Malayalam is one of the 22 scheduled languages in India with more than 130 million speakers. This paper presents a report on the development of a speaker independent, continuous transcription system for Malayalam. The system employs Hidden Markov Model (HMM) for acoustic modeling and Mel Frequency Cepstral Coefficient (MFCC) for feature extraction. It is trained with 21 male and female speakers in the age group ranging from 20 to 40 years. The system obtained a word recognition accuracy of 87.4% and a sentence recognition accuracy of 84%, when tested with a set of continuous speech data.

Modelagem acústica por diferenças finitas e elementos finitos em 2-D e 2,5-D

A modelagem acústica fornece dados úteis para avaliação de metodologias de processamento e imageamento sísmico, em modelos com estrutura geológica complexa. Esquemas de diferenças finitas (DF) e elementos finitos (EF) foram implementados e avaliados em modelos homogêneos e heterogêneos. O algoritmo de diferenças finitas foi estendido para o caso 2,5-D em modelos com densidade variável. Foi apresentada a modelagem de alvos geológicos de interesse exploratório existentes na Bacia Paleozóica do Solimões na Amazônia. Reflexões múltiplas de longo período produzidas entre a superfície livre e a discordância Cretáceo-Paleozóica, a baixa resolução da onda sísmica nas proximidades do reservatório e as fracas reflexões na interface entre as rochas reservatório e as rochas selantes são as principais características dos dados sintéticos obtidos, os quais representam um grande desafio ao imageamento sísmico.

Evaluación subjetiva de la diferenciación entre las respuestas impulsivas medidas y simuladas en diferentes recintos

Para estudiar la acústica en recintos se utilizan herramientas de modelado acústico de salas y la auralización de estos recintos. En muchos casos el mismo software de modelado genera las auralizaciones que simulan como se escucharía un determinado audio en un punto del recinto. En el presente trabajo se intenta estudiar la calidad de las respuestas impulsivas simuladas generadas por una de estas herramientas de modelado. El proyecto consiste en realizar una evaluación subjetiva de la diferenciación entre las respuestas impulsivas medidas y simuladas en diferentes recintos, comparándola con los parámetros acústicos objetivos que se han obtenido mediante una evaluación objetiva anterior. Para ello, se desarrolla una herramienta software que, con las respuestas impulsivas simuladas (generadas por un software de modelado acústico, ODEON) y las respuestas impulsivas medidas, genere: las auralizaciones correspondientes; un test de escucha para hacer la evaluación subjetiva de las auralizaciones obtenidas y una interfaz de usuario para interaccionar con los oyentes a evaluar; y que almacene los resultados del test de escucha. Además se llevará a cabo la evaluación subjetiva (test de escucha) con un grupo de oyentes, con los que posteriormente se obtendrán los resultados para poder comparar con los valores objetivos. A partir de los resultados del test de escucha, se estudiará la relación entre las diferencias de las respuestas impulsivas (medidas y simuladas) con respecto a: 1) parámetros objetivos (diferencias en valores de JND), 2) evaluación subjetiva (diferenciación subjetiva con el test de escucha) y, 3) comparación entre ambos. Finalmente, se obtendrán las conclusiones derivadas de este estudio. ABSTRACT. Tools of Room Acoustic Modeling and the auralization of these rooms have been used to study room Acoustics. In many cases, the modeling software itself generates the auralizations which simulate the sound of a particular audio in a particular position in a room. The quality of simulated impulse responses generated by a modelling tool has been studied in this project. The project carries out a subjective evaluation of the differences between measured and simulated impulse responses in different rooms, and in comparing them with objective acoustical parameters which have been obtained from the previous objective evaluation. For this, a software tool has been developed. This software, with simulated (generated by a modelling software, ODEON) and measured impulse responses generate: the correspondent auralizations, listening test (which carries out the subjective evaluation of the auralization) and a user interface to interact with listeners. This tool also stores the listening test results. In addition, the subjective evaluation (listening test) will be carried out with a group of listeners, from who we will get the results to compare subjective values. The differences between impulse responses (both measured and simulated) are obtained from the results of the listening test and these differences have been studied regarding: 1) objective parameters (difference in JND values); 2) subjetive evaluation (subjective differences with the listening test); 3) comparison between both. Eventually the conclussions derivative from this project will be obtained.

Virtual restoration of the sound of the Hispanic Rite

The virtual acoustic reality techniques are powerful tools for the recovery of acoustical heritage of historic buildings. Through the acoustic modeling and auralization techniques it´s possible to reconstruct the sound of disappeared buildings or the ones with significant modifications over the years, knowing the original geometry and the acoustic characteristics of their surfaces. This paper shows the results of a research project whose goal is the virtual recovery of the sound of the Hispanic Rite, the rite celebrated by Christians of the Iberian Peninsula before the imposition of the Roman Rite in the mid-eleventh century. For this purpose, acoustic models of a series of Pre-Romanesque churches were made. These acoustic models represent the churches in their original state, following the reconstruction hypothesis proposed by leading researchers in medieval liturgical archeology. Multichannel anechoic recordings of several pieces of the music of the Hispanic Rite have been carried out using a spherical array composed of 31 microphones. Finally, static and dynamic auralizations have been developed, involving the different liturgical configurations which were usual in this rite.

Guidelines for the design of efficient sono-microreactors

Possible drawbacks of microreactors are inefficient reactant mixing and the clogging of microchannels when solid-forming reactions are carried out or solid (catalysts) suspensions are used. Ultrasonic irradiation has been successfully implemented for solving these problems in microreactor configurations ranging from capillaries immersed in ultrasonic baths to devices with miniaturized piezoelectric transducers. Moving forward in process intensification and sustainable development, the acoustic energy implementation requires a strategy to optimize the microreactor from an ultrasound viewpoint during its design. In this work, we present a simple analytical model that can be used as a guide to achieving a proper acoustic design of stacked microreactors. An example of this methodology was demonstrated through finite element analysis and it was compared with an experimental study found in the literature.

Passive Acoustics: A Multifaceted Tool for Marine Mammal Conservation

Marine mammals exploit the efficiency of sound propagation in the marine environment for essential activities like communication and navigation. For this reason, passive acoustics has particularly high potential for marine mammal studies, especially those aimed at population management and conservation. Despite the rapid realization of this potential through a growing number of studies, much crucial information remains unknown or poorly understood. This research attempts to address two key knowledge gaps, using the well-studied bottlenose dolphin (Tursiops truncatus) as a model species, and underwater acoustic recordings collected on four fixed autonomous sensors deployed at multiple locations in Sarasota Bay, Florida, between September 2012 and August 2013. Underwater noise can hinder dolphin communication. The ability of these animals to overcome this obstacle was examined using recorded noise and dolphin whistles. I found that bottlenose dolphins are able to compensate for increased noise in their environment using a wide range of strategies employed in a singular fashion or in various combinations, depending on the frequency content of the noise, noise source, and time of day. These strategies include modifying whistle frequency characteristics, increasing whistle duration, and increasing whistle redundancy. Recordings were also used to evaluate the performance of six recently developed passive acoustic abundance estimation methods, by comparing their results to the true abundance of animals, obtained via a census conducted within the same area and time period. The methods employed were broadly divided into two categories – those involving direct counts of animals, and those involving counts of cues (signature whistles). The animal-based methods were traditional capture-recapture, spatially explicit capture-recapture (SECR), and an approach that blends the “snapshot” method and mark-recapture distance sampling, referred to here as (SMRDS). The cue-based methods were conventional distance sampling (CDS), an acoustic modeling approach involving the use of the passive sonar equation, and SECR. In the latter approach, detection probability was modelled as a function of sound transmission loss, rather than the Euclidean distance typically used. Of these methods, while SMRDS produced the most accurate estimate, SECR demonstrated the greatest potential for broad applicability to other species and locations, with minimal to no auxiliary data, such as distance from sound source to detector(s), which is often difficult to obtain. This was especially true when this method was compared to traditional capture-recapture results, which greatly underestimated abundance, despite attempts to account for major unmodelled heterogeneity. Furthermore, the incorporation of non-Euclidean distance significantly improved model accuracy. The acoustic modelling approach performed similarly to CDS, but both methods also strongly underestimated abundance. In particular, CDS proved to be inefficient. This approach requires at least 3 sensors for localization at a single point. It was also difficult to obtain accurate distances, and the sample size was greatly reduced by the failure to detect some whistles on all three recorders. As a result, this approach is not recommended for marine mammal abundance estimation when few recorders are available, or in high sound attenuation environments with relatively low sample sizes. It is hoped that these results lead to more informed management decisions, and therefore, more effective species conservation.

A state-space modeling approach for active structural acoustic control

The demands for improvement in sound quality and reduction of noise generated by vehicles are constantly increasing, as well as the penalties for space and weight of the control solutions. A promising approach to cope with this challenge is the use of active structural-acoustic control. Usually, the low frequency noise is transmitted into the vehicle`s cabin through structural paths, which raises the necessity of dealing with vibro-acoustic models. This kind of models should allow the inclusion of sensors and actuators models, if accurate performance indexes are to be accessed. The challenge thus resides in deriving reasonable sized models that integrate structural, acoustic, electrical components and the controller algorithm. The advantages of adequate active control simulation strategies relies on the cost and time reduction in the development phase. Therefore, the aim of this paper is to present a methodology for simulating vibro-acoustic systems including this coupled model in a closed loop control simulation framework that also takes into account the interaction between the system and the control sensors/actuators. It is shown that neglecting the sensor/actuator dynamics can lead to inaccurate performance predictions.

Robust acoustic and semantic modeling in a telephone-based spoken dialog system

Magdeburg, Univ., Fak. für Elektrotechnik und Informationstechnik, Diss., 2009

Finite Element Modeling of Spiral Frequency Steerable Acoustic Transducers (FSATs) for guided waves based Structural Health Monitoring of plate-like structures

Structural Health Monitoring (SHM) is an emerging area of research associated to improvement of maintainability and the safety of aerospace, civil and mechanical infrastructures by means of monitoring and damage detection. Guided wave structural testing method is an approach for health monitoring of plate-like structures using smart material piezoelectric transducers. Among many kinds of transducers, the ones that have beam steering feature can perform more accurate surface interrogation. A frequency steerable acoustic transducer (FSATs) is capable of beam steering by varying the input frequency and consequently can detect and localize damage in structures. Guided wave inspection is typically performed through phased arrays which feature a large number of piezoelectric transducers, complexity and limitations. To overcome the weight penalty, the complex circuity and maintenance concern associated with wiring a large number of transducers, new FSATs are proposed that present inherent directional capabilities when generating and sensing elastic waves. The first generation of Spiral FSAT has two main limitations. First, waves are excited or sensed in one direction and in the opposite one (180 ̊ ambiguity) and second, just a relatively rude approximation of the desired directivity has been attained. Second generation of Spiral FSAT is proposed to overcome the first generation limitations. The importance of simulation tools becomes higher when a new idea is proposed and starts to be developed. The shaped transducer concept, especially the second generation of spiral FSAT is a novel idea in guided waves based of Structural Health Monitoring systems, hence finding a simulation tool is a necessity to develop various design aspects of this innovative transducer. In this work, the numerical simulation of the 1st and 2nd generations of Spiral FSAT has been conducted to prove the directional capability of excited guided waves through a plate-like structure.

Modeling of functionally graded piezoelectric ultrasonic transducers

The application of functionally graded material (FGM) concept to piezoelectric transducers allows the design of composite transducers without interfaces, due to the continuous change of property values. Thus, large improvements can be achieved, as reduction of stress concentration, increasing of bonding strength, and bandwidth. This work proposes to design and to model FGM piezoelectric transducers and to compare their performance with non-FGM ones. Analytical and finite element (FE) modeling of FGM piezoelectric transducers radiating a plane pressure wave in fluid medium are developed and their results are compared. The ANSYS software is used for the FE modeling. The analytical model is based on FGM-equivalent acoustic transmission-line model, which is implemented using MATLAB software. Two cases are considered: (i) the transducer emits a pressure wave in water and it is composed of a graded piezoceramic disk, and backing and matching layers made of homogeneous materials; (ii) the transducer has no backing and matching layer; in this case, no external load is simulated. Time and frequency pressure responses are obtained through a transient analysis. The material properties are graded along thickness direction. Linear and exponential gradation functions are implemented to illustrate the influence of gradation on the transducer pressure response, electrical impedance, and resonance frequencies. (C) 2009 Elsevier B. V. All rights reserved.

Modeling of Effect of Plastic Deformation on Barkhausen Noise and Magnetoacoustic Emission in Iron With 2% Silicon

Before one models the effect of plastic deformation on magnetoacoustic emission (MAE), one must first treat non-180 degrees domain wall motion. In this paper, we take the Alessandro-Beatrice-Bertotti-Montorsi (ABBM) model and modify it to treat non-180 degrees wall motion. We then insert a modified stress-dependent Jiles-Atherton model, which treats plastic deformation, into the modified ABBM model to treat MAE and magnetic Barkhausen noise (HBN). In fitting the dependence of these quantities on plastic deformation, we apply a model for when deformation gets into the stage where dislocation tangles are formed, noting two chief effects, one due to increased density of emission centers owing to increased dislocation density, and the other due to a more gentle increase in the residual stress in the vicinity of the dislocation tangles as deformation is increased.

Mathematical modeling of pressurized system behaviour with entrapped air

The presence of entrapped air in pressurized hydraulic systems is considered a critical condition for the infrastructure security, due to the transient pressure enhancement related with its dynamic behaviour, similar to non-linear spring action. A mathematical model for the assessment of hydraulic transients resulting from rapid pressurizations, under referred condition is presented. Water movement was modeled through the elastic column theory considering a moving liquid boundary and the entrapped air pocket as lumped gas mass, where the acoustic effects are negligible. The method of characteristics was used to obtain the numerical solution of the liquid flow. The resulting model is applied to an experimental set-up having entrapped air in the top of a vertical pipe section and the numerical results are analyzed.