speaker adaptation apparatus of continuation voice recognition system

摘要

PURPOSE: A speaker adaptation device of a continuous voice recognition system is provided to recognize a variety of human conversations and general sentences in real time. CONSTITUTION: A speaker adaptation device of a continuous voice recognition system includes a signal input unit(10) for receiving continuous voice signals, a training data collector(20) that collects a proper number of training data items depending on the number of clusters of a speaker independent model, and a speaker independent model clustering unit(30) that divides the speaker independent model into clusters as many as the training data items. The speaker adaptation device further includes a speaker adaptation unit(40) that forms a speaker adaptation model for each cluster using maximum-likelihood linear regression, a continuous voice recognition unit(50) that replaces a speaker-adapted model with the speaker independent model to obtain an improved voice recognition result for each speaker, and a recognition result execution unit(60) that executes an operation based on the recognition result.