Vuelta al Principio Última
actualización de esta página:
viernes, 18 de septiembre de 2015
Visitantes:
Esta página está en construcción: perdonen los errores, repeticiones y temas inacabados.
This page is being developed: I am sorry for errors, duplications and unfinished subjects.
5. Random contiguous pitch values (included in An_aadf_pitch_estimator)
A. The nature of pitch (let us not forget the essentially subjective nature of this parameter) is continuous because the tension of arytenoids cartilages (the one which tension regulates vocal cords tension and then for pitch) varies continuously, no matter how quickly. Thus, the estimated pitch must vary also continuously and in a limited quantity by time unit, that is, its gradient is limited.
Therefore, once we are fairly sure that we got the right pitch (what we believe to be true when consecutive pitch values are similar ‒within a ten percent of the values), we must look only for neighboring values of the last calculated pitch. We find in that way a sort of looking window within which we look for the next pitch value.
In that way, we find a spectacularly quick algorithm because we only calculate a few potential values of next pitch.
B. Another approach of the general problem of choosing between several candidates of pitch for the same time segment (harmonics of fundamental) is to keep for several continuous time segments these candidates to check afterwards the relative consistency of each family which we could call pitch tracks. It is probable that some of them will vanish or at least break the continuity more than others, which we will prefer as the correct pitches.
The problem becomes now the evaluation of simultaneous pitch signals to choose the better, that is the one that maximize or minimizes a formula of algorithm of pitch quality evaluation. Of course this transfer the problem to a more general one, because you must optimize functions instead of values.
The only problem can be to take a wrong value, the octave for instance, so we must focus in finding a robust estimator of initial voiced segments.
Vuelta al Principio Última actualización de esta página: viernes, 18 de septiembre de 2015 Visitantes:
