Figure 3From: Audio visual speech source separation via improved context dependent association modelVideo coherent speech detection on pure relevant and irrelevant signals. (a) Results for different configurations of model type (MLP,GMM), ka (2,4,6,8,10,12), kv (2,4,6,8) and T (4,8,16). (b) Error bars (95% confidence interval) for curves with T=16. Other free parameters are marginalized by maximum selection (see text for more details).Back to article page