Next: CELP coders
Up: Speech coding
Previous: Multi-pulse coders
The basic waveform coders presented so far compute and transmit LPC and
residual parameters on a frame by frame basis. Backwards adaptation
uses past decoded speech to estimate the LP model. Issues are:
-
No need to transmit LP paramterers
-
LP Model is poorer - trained on delayed data
-
LP paramters are hard to quantise - consume appreciable bandwidth
-
Simpler bitstream format
-
Longer model order is possible - not limited by quantisation error
-
Decoder needs more power to generate the LP model
Works best for low delay systems.
Speech Vision Robotics group/Tony Robinson