A New Multi-modal Database for Developing Speech Recognition Systems for an Assistive Technology Application

In this paper we report on the acquisition and content of a new database intended for developing audio-visual speech recognition systems. This database supports a speaker dependent continuous speech recognition task, based on a small vocabulary, and was captured in the European Portuguese language. Along with the collected multi-modal speech materials, the respective orthographic transcription and time-alignment files are supplied. The package also includes data on stochastic language models and the generative grammar associated to the collected spoken sentences. The application addressed by this database, which consists of voice control of a basic scientific calculator, has the particularity of being designed for a person with a specific motor impairment, namely muscular dystrophy. This specificity is a remarkable characteristic, given the lack of such kind of data resources for developing assistive systems based on audio-visual speech recognition technology.

• Type: Conference
• Year: 2004
• Where: TSD
• Authors: António Moura, Diamantino Freitas, Vitor Pera
• Keywords: Audio-Visual Speech Recognition, Signal Processing, Speech Recognition, Speech Recognition Systems, TSD 2004