Self-Learning Speaker Identification

Current speech recognition systems suffer from variation of voice
characteristics between speakers as they are usually based on speaker
independent speech models. In order to resolve this issue, adaptation
methods have been developed in many state-of-the-art systems. However,
information acquired over time is still lost whenever another speaker intermittently
uses the recognition system. This work therefore develops an integrated
approach for speech and speaker recognition in order to improve the
self-learning opportunities of the system. A speaker adaptation scheme
is introduced. It is suited for fast short-term and detailed long-term
adaptation. These adaptation profiles are then used for an efficient
speaker recognition system. The speaker identification enables the
speaker adaptation to track different speakers which results in an
optimal long-term adaptation.