Herbig, Tobias.
Self-Learning Speaker Identification A System for Enhanced Speech Recognition / [electronic resource] : by Tobias Herbig, Franz Gerl, Wolfgang Minker. - XII, 172 p. online resource. - Signals and Communication Technology, 1860-4862 . - Signals and Communication Technology, .
Introduction -- State of the Art -- Fundamentals -- Speech Production -- Front-End -- Speaker Change -- Speaker Identification.-Speaker Adaptation.
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.
9783642198991
10.1007/978-3-642-19899-1 doi
Engineering.
Computer science.
Biometrics.
Telecommunication.
Engineering.
Signal, Image and Speech Processing.
Biometrics.
Communications Engineering, Networks.
User Interfaces and Human Computer Interaction.
TK5102.9 TA1637-1638 TK7882.S65
621.382
Self-Learning Speaker Identification A System for Enhanced Speech Recognition / [electronic resource] : by Tobias Herbig, Franz Gerl, Wolfgang Minker. - XII, 172 p. online resource. - Signals and Communication Technology, 1860-4862 . - Signals and Communication Technology, .
Introduction -- State of the Art -- Fundamentals -- Speech Production -- Front-End -- Speaker Change -- Speaker Identification.-Speaker Adaptation.
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.
9783642198991
10.1007/978-3-642-19899-1 doi
Engineering.
Computer science.
Biometrics.
Telecommunication.
Engineering.
Signal, Image and Speech Processing.
Biometrics.
Communications Engineering, Networks.
User Interfaces and Human Computer Interaction.
TK5102.9 TA1637-1638 TK7882.S65
621.382