INIGEM   23989
INSTITUTO DE INMUNOLOGIA, GENETICA Y METABOLISMO
Unidad Ejecutora - UE
congresos y reuniones científicas
Título:
Speaker-independent embedded speech recognition using Hidden Markov Models
Autor/es:
MARIANO MARUFO DA SILVA; SEBASTIÁN VERRASTRO; DIEGO A. EVIN
Lugar:
Buenos Aires
Reunión:
Congreso; Congreso Argentino de Ciencias de la Informática y Desarrollos de Investigación; 2016
Institución organizadora:
Centro de Investigación y Desarrollo en Informática de la Escuela de Ciencia y Tecnología de la Universidad Nacional de San Martín- Universidad CAECE - Universidad Central de Chile
Resumen:
Even though Automatic Speech Recognition (ASR) on embedded platforms has reached performance levels consistent withthe requirements of commercial applications, in most cases this performance is achieved at the expense of transferring the complexity of the problem to remote servers. For some applications this represents a problem, since the system requires an internet connection and its speed is strongly limited by the available bandwidth. Furthermore, there are privacy concerns related to data transfers. In this paper we detail a solution which implements the recognition directly on the device.