Forums

waveform interpolation

Started by dr_simic September 30, 2003
I'm trying to implement WI codec,for now just layer I(without quantisation),as
it is described in Eddie Choy's thesis,but I don't get perfectly sounding
decoded speech,as they stated,there are audible errors,and these errors usually
occur when pitch doubles. Can anybody help me,or send me some code that is
working.
Thanks in advance.



I think nobody will send you the code of WI since now it is a
commercial secret.

About Choy's coder. In the site of McGill University Choy publishes
the demo files that was generated by his coder. On listening them I am
not sure that Choy's coder (if he implemented his thesis exactly) can
produce so good speech. So, be careful when you compare you coder to
Choy's one.

One of the problem of Choy's coder is an absolutely primitive pitch
detector; WI needs much more sophisticated algorithm. I recommend you
to look at the pitch detector of MELP: it efficiently solves the
pitch-doubling.

See also my message in this newsgroup #644.

Best regards,
Ilya Druker
--- In , "dr_simic" <dr_simic@y...> wrote:
> I'm trying to implement WI codec,for now just layer I(without
quantisation),as it is described in Eddie Choy's thesis,but I don't
get perfectly sounding decoded speech,as they stated,there are audible
errors,and these errors usually occur when pitch doubles. Can anybody
help me,or send me some code that is working.
> Thanks in advance.


Ilya-

> I think nobody will send you the code of WI since now it is a
> commercial secret.
>
> About Choy's coder. In the site of McGill University Choy publishes
> the demo files that was generated by his coder. On listening them I am
> not sure that Choy's coder (if he implemented his thesis exactly) can
> produce so good speech. So, be careful when you compare you coder to
> Choy's one.
>
> One of the problem of Choy's coder is an absolutely primitive pitch
> detector; WI needs much more sophisticated algorithm. I recommend you
> to look at the pitch detector of MELP: it efficiently solves the
> pitch-doubling.
>
> See also my message in this newsgroup #644.

The MELP pitch detector is very good. However, we've found some errors in it
and
corrected those, plus added some additional processing. Basically, our
objective was
to make a "MELPe-Plus" version that can handle music-on-hold, and we've made
good
progress towards that.

Jeff Brower
system engineer
Signalogic > --- In , "dr_simic" <dr_simic@y...> wrote:
> > I'm trying to implement WI codec,for now just layer I(without
> quantisation),as it is described in Eddie Choy's thesis,but I don't
> get perfectly sounding decoded speech,as they stated,there are audible
> errors,and these errors usually occur when pitch doubles. Can anybody
> help me,or send me some code that is working.
> > Thanks in advance.