Figure 20
From: Multi-pose lipreading and audio-visual speech recognition
Scatter between optimal video weight and lipreading performance for a corrupted audio with 7 dB of SNR. Scatter plot between the optimal value of the weight associated to the visual stream in the audio-visual classifier and the performance of the corresponding visual stream in lipreading experiments. The audio stream is a corrupted with 7 dB of SNR, while different visual streams are considered: an ideal frontal view of the speaker, the original lateral views at 30°, 60°, and 90° of head rotation and the corresponding streams after pose normalization.