The group delay function can be effectively used for various speech processing tasks only when the signal under consideration is a minimum phase signal.
What kind of speech processing you are currently doing? whether speech enhancement, speech recognition or speaker recognition or verification?. It is considered that Phase information of the speech signal has very little importance (Though there are papers which is against this argument). In general minimum phase systems introduce minimal group delay which is a requirement for the distortion less transmission lines.
I personally consider phase to be important. Two vowels sounding same have less perceptual difference, but the phase difference may be present. I am referring to distinguishing natural and HMM speech on phase based features for the problem of imposter attack in speaker verification.
If some of the zeros that correspond to the system, that might have generated the (speech) signal, lie outside the unit-cirle (non-minimum phase system), then the group delay spectrum is expected to be noisy (spiky at the angular locations of these zeros). To avoid this, researchers try to make the signal as a minimum-phase signal before the computation of group delay spectrum. However, there are ways to handle this scenario, without deriving an equivalent minimum phase signal. This is only my view.
Yes sir, I am aware of requirement of minimum-phase if spikes in group delay is to be avoided. I wanted to know, if it is compulsory. But I guess it will depend on the application.. Thank you.
Yes, it is compulsory. The group delay is, to certain extent, similar to the magnitude spectrum of the signal. Those spikes are due to wrapped phase and not actual, and it has to be avoided.
- Badges
- Science topic