You will need OpenCV, Festival and pi-blaster installed for this to work:
sudo apt-get install python-opencv festival
pi-blaster is a manual install: https://github.com/sarfata/pi-blaster/
The text files named hi and bye are the files from which Festival speaks when a face is detected (hi) and when the face leaves the screen (bye). You can modify these files as necesary to get the desired speaking for each face event. Be aware that using pi-blaster for the servos means you can't use the analog audio anymore - it's busy generating the PWM signals for the servos. Audio must be via HDMI for this to work. This is a setback for now, sorry.
I used the OpenCV code on this webpage: http://www.technolabsz.com/2013/03/how-to-easily-install-opencv-on.html
Included is the Python script and the human face XML from which the OpenCV uses to identify the faces. I modified the facedetect.py script to work with the servos and festival output - this is the script that you have now from this Github repo.
pi-blaster makes all the GPIO pins PWM capable, you need to use 2 pins for the servos, one for pan the other for tilt. My facedetect.py uses pi-blaster pin 6 for tilt and 7 for pan (GPIO 24,25). You can change these pins as needed.
The servo values need to be calibrated for your individual servos. There is a mapping function that scales a range into another range, which given the calibrated servo values for both pan and tilt will give you accurate tracking. Play with these values until they work for your servos.
Any questions please feel free to ask.