A low bit rate speech coding method using a formant-articulatory parameter nomogram

In this paper, we propose a new method for low bit rate speech coding using a nomogram that is a pair of codebooks representing the functional relationship between formant frequencies and articulatory parameters. Significant features of our approach are 1) using the codebooks derived theoretically from the computation using a stylized vocal tract model and 2) independent coding by separating frequency information from the amplitude in a speech segment. From these features, the method is also characterized by little dependency upon speech databases and/or languages in the acoustic domain, so that it has a potential to construct a more flexible rule-based speech synthesis system. We have conducted articulatory encode-decode experiments with the bit rate range from 3.2kbps to 1.6kbps using speech samples in ASJ and TIMIT speech databases and confirmed that good quality speech synthesis is achieved with improvements on the bit allocation scheme and a frame sampling method.