VoxForge
VoxForge
My research need a speaker independent acoustic model, i wonder whether i should use the adaptation or to totally train it from scratch?
Say for example, i've try out voxforge acoustic model to recognize my voice data, but the result turn out to be not so accurate.
In this case, should i adapt my voice data to the voxforge AM or should i download the voxforge original data together with my own voice data to train them again?
TQ!
--- (Edited on 9/12/2009 12:32 am [GMT-0500] by degra) ---
> My research need a speaker independent acoustic model, i wonder whether i should use the adaptation or to totally train it from scratch?
Did you mean speaker-dependent?
> Say for example, i've try out voxforge acoustic model to recognize my voice data, but the result turn out to be not so accurate.In this case, should i adapt my voice data to the voxforge AM or should i download the voxforge original data together with my own voice data to train them again?
In this case you first need to measure the accuracy quantitatively and get precise word error rate.
--- (Edited on 9/12/2009 4:10 am [GMT-0500] by nsh) ---
Hi,
> Did you mean speaker-dependent?
Eventually i need to have speaker-independent, but i think it is because our local english accents, thus the Voxforge AM doesn't seems like returning accurate result, and that's why i need to try out if it recognize my voice (accents) first before i collect more data from other place.
> In this case you first need to measure the accuracy quantitatively and get precise word error rate.
As i'm still new to Speech Recoginition Engine, what do you mean by accuracy quantitative and how the precise word error rate can improve the accurracy of recognition?
TQ!
--- (Edited on 9/12/2009 11:56 pm [GMT-0500] by degra) ---