Automatic speech recognition based on a new segmentation procedure