Automatic speech recognition of the Mixtec language: An ubiquitous computing application

In this work we present the implementation by means of ubiquitous computing of an Automatic Speech Recognizer (ASR) for the Mixtec language. This is proposed as a key development for future speech translation systems between the Mixtec and the Spanish languages in remote communities. The core of the ASR system was built with Hidden Markov Models (HMMs) and the management of its resources was performed with an interface developed with MATLAB. The interface was integrated with scripts that allowed remote access to the ASR by means of the MLConnect for Android devices with Wi-Fi connection. Experiments performed with a smartphone and a tablet PC, both inside a building, showed fast accessibility and word recognition accuracies over 92% for non-native speakers.