Age and gender recognition from speech patterns based on supervised non-negative matrix factorization
Mohamad Hasan Bahari, Hugo Van hamme · 2011
In many criminal cases, evidence might be in the form of recorded conversations, possibly over the telephone. Therefore, law enforcement agencies have been concerned about accurate methods to profile different characteristics of a speaker from recorded voice patterns, which facilitate to identify him/her or at least narrow down the number of suspects. This paper proposes a new gender and age group recognition approach based on Non-Negative Matrix Factorization (NMF) (Lee and Seung, 2001). First, an acoustic model is trained for all speakers in a training database including male and female speakers of different age. Then, Gaussian Mixture (GM) weights are extracted and concatenated to form a supervector for each speaker. Finally, Supervised NMF (SNMF) is applied to detect the gender and age group of unseen test speakers. Evaluation results on a corpus of read and spontaneous speech in Dutch confirms the effectiveness of proposed scheme.