agree with Monther. Could you give more detail ?

Dear Monther and Nicolas,

First of all, Thank you for your responses,

I am talking about using the SVM decision values as scores to allow or deny a user to access to the system, i.e. if the decision value is greater than a fixed threshold (wich is usually different to 0), the user is allowed, otherwise, the user is denied.

Hi Ayoub, 

These papers might may shed some light on this matter in the context of forensic speaker recognition (FSR). One of the frameworks used in automatic speaker recognition (and discussed in this paper) is based on a GMM-SVM for acoustic modelling and scoring. The interpretation of the SVM output (and other scoring procedures) are discussed in the context of FSR and some of the problems that arise in the decision process (calibration, thresholding, etc) are also detailed.

---

https://dspace.mit.edu/openaccess-disseminate/1721.1/52318

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.207.9328&rep=rep1&type=pdf

I must say I am a bit sceptical about the idea. On principle, yes, you could use a SVM for that, in practice, it means carefully choosing the features and having a very rigorous training, especially because false négative (a people that would normally be authorized to access but is denied by the system) is not a big issue (the user may contact the adminsitrator that can grant her/him access) but a false positive (granting access to someone who should not have it) is something you want to avoid at all costs 

Dear Waad,

Thank you very much for your response :)

Dear Nicolas,

I think that it depends on the context of application ..  Any way, thank you for your response.