Iván Vladimir Meza Ruíz

Practical Speech Recognition for Contextualized Service Robots

In this work, we present the speech recognition module of a service robot that performs various tasks, such as being a host party, receiving multiple commands or giving a tour guide. These tasks take place in diverse acoustic environments, e.g., a home or a supermarket, in which speech is one of the main modalities of interaction. Our approach relies on three strategies: 1) making the recognizer aware of the task context, 2) providing prompting strategies to guide the recognition, and 3) calibrating the audio setting specific to the environment. We provide an evaluation with recordings from real interactions with a service robot in different environments.

Knee-Ankle Sensor for Gait Characterization: Gender Identification Case

Classification based on gait biomarkers is an area of study that includes approaches aimed at monitoring (vigilance), education and health. A correct classification is achieved depending on algorithms that serve that purpose, however, an accuracy must be available during the data acquisition of gait. In this study, a sensor network is proposed that allows to capture, in children, data of knee and right ankle. Results shows acceptable percentages of correct classification when implementing various machine learning algorithms, especially, combining the LogitBoost+Random Forest algorithms.

Emotion Based Features of Bird Singing for Turdus migratorius Identification

A possible solution for the current rate of animal extinction in the world is the use of new technologies in their monitoring in order to tackle problems in the reduction of their populations in a timely manner. In this work we present a system for the identification of the Turdus migratorius bird species based on their singing. The core of the system is based on turn-level features extracted from the audio signal of the bird songs. These features were adapted from the recognition of human emotion in speech, which are based on Support Vector Machines. The resulting system is a prototype module of acoustic identification of birds which goal is to monitor birds in their environment, and, in the future, estimate their populations.