Creating and sharing knowledge for telecommunications

Project: Acoustic Modelling for Speech Recognition

Acronym: AcuRec
Main Objective:
The main objective of this project is the automatic training of the acoustic model in a speech recognition system. The approach to follow consists in the reliable detection of acoustic events that identify anchor points and short segments for which phonetic labels will be estimated. This will be done in an iterative manner using previous phone models and confidence measures of its correct classification. The spectral analysis system will also include features with perceptual significance with integration periods higher and lower than the normally used ones. In order to increase the segmentation accuracy we intend to also use speaker normalization and adaptation techniques.
Funding: IT/LA
Start Date: 01-09-2005
End Date: 30-09-2007
Team: Fernando Manuel Santos Perdigao, Carla Alexandra Calado Lopes, Eduardo Marta, Luis António Serralva Vieira de Sá
Groups: Multimedia Signal Processing – Co
Local Coordinator: Fernando Manuel Santos Perdigao
Links: Internal Page
Associated Publications