Figure 4From: The self-taught vocal interfaceNMF-based learning against GMM-based learning. Upper part: for severe dysarthric speakers. Lower part: for speakers with extended training sets in the lower part. Numbered circles represent PID, and their locations indicate F-scores as a function of the number of utterances in the training sets. Furthermore, the smoothed curves are interpolations of the scattered F-scores using the LOWESS procedure, and they exemplify the performance of an average speaker.Back to article page