I'm working on a voice biometrics project and am looking for some guidance on two key issues:
First, does anyone know if a standard exists for voice biometric system accuracy? I have been unable to find any specific numbers, but it looks like most publications aim to achieve an accuracy of at least 80%.
Second, what is considered the most "state-of-the-art" system? From what I've gathered, current work on feature extraction has largely ranked on MFCC's over LPC's. Acoustic modeling seems split between GMM's, HMM's, Vector Quantization, etc. How does Dynamic Time Warping strengthen accuracy?
If anyone is knowledgable, I would greatly appreciate some help.