Info

This question is closed. Reopen it to edit or answer.

Speaker Verification in matlab - posterior function intepretation

1 view (last 30 days)
My simplified code looks like:
params = 4;
%number of params to extract from spreach signal
components= 4;
% number of gmm components
for speaker = 1:212
path = ['..\speaker_' int2str(speaker) '.mat']
load(path,'X');
data{speaker} = X;
end
% end of preaparation known speakers from db
speakerID = input('Type your speaker ID: ');
if (isempty(speakerID))
disp('Error - incorrect ID!');
end
declarationParams = get_speaker_params_with_microphone(params);
%returns params(in this case 4) parameters of speach signal for one windowed frame,
%for example matrix 200 x 4
Y = data{speakerID}(:,1:params);
% get speakerID datas from db to Y
gmm_DB_pattern = mggmmemfit(Y, components, 0, 0);
% compute (using EM) gmm from speakerID datas, function returns same object as
gmdistribution.fit, link to better desc is below
https://cs.uwaterloo.ca/~mgrzes/code/mggmmemfit.m
Here is the problem:
Can someone explain me how should I compute/intepret results of posterior to get percentage or value between 0-1 of probability that speakerID typed at the beginning is proper for speaker recorded with get_speaker_params_with_microphone(params)?
[P, nlogl] = posterior(gmm_DB_pattern, declarationParams);
I think this function is proper to solve my problems, but I can not properly understand how to use it to acomplish my goal.
To sum up, I would like to ask how to compute likelihood that declared (testing sequence) params is being uttered by declared speaker ??

Answers (1)

Michal
Michal on 31 Aug 2014
I am facing with similar problem can anyone let me know how should I read nlogl ?

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!