We present a connectionist system for text independent speaker identification. We have developed an architecture based on the cooperation of several connectionist modules to achieve this identification. The system is composed of a typology detector and a set of expert modules. Each expert module of the system is concerned with the discrimination between speakers of the same typology. The score used in the final decision is obtained weighting the scores of the typology detection module with those of the expert modules. The system has been tested on a population of 102 speakers extracted from the DARPA-TIMIT database. Perfect identification has been observed, specifically, an interval of confidence 95% for [99.9%, 100.0%] recognition with a precision of 0.1%. The performances of our system are compared with those of a system based on multivariate auto-regressive models.