Multiple Vocoder Translation Software Application

Period of Performance: 10/28/2008 - 10/28/2010

$1.5MM

Phase 2 SBIR

Recipient Firm

Arcon Corp.
260 Bear Hill Road
Waltham, MA 02451
Principal Investigator

Abstract

A set of vocoder translation methods will be developed for use in voice communications applications where different vocoder algorithms are being used across the application. The translation methods will provide reduced complexity and increased performance from that of digital tandem approach. The translations will address system configurations involving both speech model parametric based vocoders and direct waveform representation types of vocoder. The complexity of a translation method for a given configuration will be targeted to be less than that of the corresponding tandem for that configuration. The goal for the performance of the voice communication for a given configuration will be greater than or equal to that of the corresponding digital tandem. Previous research has demonstrated that by adjusting for parametric vocoder algorithm differences in procedures such as pre-emphasis and characteristics such as analysis window size, an increased performance can be achieved by transcoding. In a transcode, the model parameters from the source vocoder are translated to the model parameters of the destination vocoder. For those parametric vocoders that do not incorporate acoustic noise reduction techniques, the performance of specific configuration is limited to that of the lower data rate coder of the configuration. With waveform types of vocoders, with no parametric basis, it is not possible to use a true transcode approach to translation. For theses configurations, smart tandem approaches will be developed. The smart tandems will match the digital speech streams for best performance while reducing complexity with mixed synthesis approaches.