Speaker characteristics in speech and speaker recognition

Summary form only given. This paper analyses the acoustic variability of speakers and its impact on the robustness of contemporary automatic speech recognition and speaker recognition systems. The physiological and behavioural constraints of individual speech articulation are reviewed and examples are given of how these constraints affect commonly used spectral and cepstral features. Speech data from the Australian National Database of Spoken Language is analysed to show how gender, age and dialect affect the acoustic models in speech and speaker recognition systems. Some adaptive strategies for speech recognition and corresponding speaker recognition methods are discussed