The "Fundamentals of Speech Recognition" text by Rabiner, mentions three approaches for Speech (Endpoints) detection - Explicit, Implicit, and Hybrid approaches. I am studying the block diagram for the explicit approach.

Steps involved in explicit approach are Speech signal -> feature extraction -> Speech Detection -> Pattern Comparision -> Decision Rule -> Speech Endpoints detected.

This process, which is intended to output Speech (Endpoints) detection, itself has a speech detection block in it. I find this counter intuitive.

Moreover, I do not see how it is possible to have a speech detection block just after the feature extraction block. The reason is that, for example, if we go with MFCC as features, we take features for every frame of the speech (frame duration usually is 20 - 30 ms). It is not possible to do speech detection (that too endpoints) with just 25 ms worth information.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.