Bitte wählen Sie ihr Lieferland und ihre Kundengruppe
Recursive methods for parameter estimation in order to facilitate implementation in custom digital signal processors are presented. A 'sample-by-sample' approach reduces the memory requirements at the cost of an increased computational load. An adaptive lattice filter have been found to be very suitable to follow the vocal tract behaviour. For each new speech sample a set of typically 10 coefficient values will be produced, which requires a strong data reduction before transmission. We define blocks of convenient length and sample the coefficients for each block. Both fixed and adaptive sampling schemes are investigated. It is essential to recreate coefficient values in the receiver for each sample or with small intervals. Several interpolation algorithms have been compared, in particular linear interpolation of coefficients and area functions. Those blocks where sudden changes in the spectral properties occur create special problems and we are studying means of identifying these by a technique known as event detection.