STATINA – multi-channel acoustic room impulse response for speech processing

Abstract

The objective of this task is to develop a speaker independent automatic speech recognition system (ASR) for STATINA to make all the facilities easily accessible by the voice commands. The aim is to develop an ASR which is robust to background noises and reverberation. With the integrated speech recognizer, different applications or tasks can be invoked using voice commands. An HMM (Hidden Markov Model)-based ASR is being developed. A set of commands has been defined for STATINA with the flexibility to increase the size of the command set at any point of time. A specific grammar has also been set for the commands. A database consists of Asian subjects was created to cater the needs of the system which will be used in Asian region. The system is able to achieve up to 100% accuracy with a simple grammar, which indicates that the good quality of training data is achieved. A study of feasibility to implement the system inside ADSP-BF533 DSP board is done, and the initial program of feature extraction is made. This is the first step towards the goal of making a portable speech recognition system for STATINA.