Hi
I have project in pd
to develop a speech analyser that outputs relevant parameters into a .pfs file that can then be played back using the formant synthesiser

it is the inverse operation of Holmes parallel formant synthesiser

can any one help me??

no one can help me ?!!

Besan, with all respect, your question can be answered with another question... I'm looking for the meaning of life. Can you answer this for me?

oh, and you have one day before I start getting frustrated with your lack of answer.

and I won't bother to offer you any advice I can give you from my prior research that may help you in your search for my answer.

what parameters do you want to process?
should the formant synth re-create sort of understandable "speech" or rather abstract sounds?

The 'Holmes' parallel formant synthesiser Capable of producing speech from12 parameters (updated every 10 msecs):
-- FN low-frequency formant (fixed at 250 Hz)
-- ALF amplitude of low-frequency region
-- F1 frequency of 1st formant
-- A1 amplitude of 1st formant
-- F2 frequency of 2nd formant
-- A2 amplitude of 2nd formant
-- F3 frequency of 3rd formant
-- A3 amplitude of 3rd formant
-- AHF amplitude of 4th formant (fixed at 3500 Hz)
-- V degree of voicing
-- F0 fundamental frequency
-- MS glottal pulse mark/space ratio (fixed)

the required patch should accept wave file and produce pfs file that contains these parameters

how can I do it?

please ...........

I just need the algorithm or some hints

Well... though I do not know the exact solution of your problems as it is long time that I am away from DSP world. Anyhow few years back, I was too at your position and really curse the PD tool a lot with my friends ..... that what the hell this PD is ....:) there is no main function as we are often familiar with C/JAVA etc... it is a different world but believe me after a very short period of time, I found it very interesting and simple to program in PD. so just gear up yourself, it is new world for the orthodox programmers but very easy world. I must say.

Now come to the problem one by one. suppose we want to calculate the amplitude of voice. Do the following steps. ( some senior/junior DSP programmer can correct me, if I am wrong)

step 1: Use the adc~ object to take input
step 2: perform the fft~ to convert the input into a complex number
step 3: calculate the magnitude by using the formula sqrt (real*real + imaginary* imaginary). you have to use the object +, * and sqrt for this purpose.

step 4: show the output in some display object.

step 5: try to learn write this info by using write~ or some other kind of object into file.
step 6: use the dac~ object in the end to listen your voice...

Thank you so much ahmad
I've followed your steps
and it works and I've manged to write the result in file

but how can I compute the 12 parameters??

-- FN low-frequency formant (fixed at 250 Hz)
-- ALF amplitude of low-frequency region
-- F1 frequency of 1st formant
-- A1 amplitude of 1st formant
-- F2 frequency of 2nd formant
-- A2 amplitude of 2nd formant
-- F3 frequency of 3rd formant
-- A3 amplitude of 3rd formant
-- AHF amplitude of 4th formant (fixed at 3500 Hz)
-- V degree of voicing
-- F0 fundamental frequency
-- MS glottal pulse mark/space ratio (fixed)

http://www.pdpatchrepo.info/hurleur/wav2pfs.pd

can some one give me the equations to compute these 12 parameters?

Have you tried looking through this?

http://books.google.com/books?id=8yFi1kZD5ucC&lpg=PA56&ots=DaiBfcQftD&dq=holmes%20formant%20synth&pg=PP1#v=onepage&q=&f=false

http://www.busim.ee.boun.edu.tr/~speech/research.html

these guys are actively working in speech processing domain. Few years back I also studied some of their papers to extract information. Have a look over their research work and even make a request for guidance , may be some one guide you as well.

thank you so much
I'll study these papers and if I have any questions I'll post them here

hello there,

new to this forum, and to programing, really, so this may be way off .... but there is a usefull object for easy fft analysis (since it looks like you don't go past the 3rd formant). that's the fiddle~ object.

Hey Besan, did you get anywhere with this project? I'm having to do something very similar at the moment.

I just wanted to "revive" this post.
I am trying to work something similar myself.
Unfortunately the estimations for the formants that I get are not really good which results to a crappy output that makes no sense on the Formant synthesiser.
Could anyone please give ideas on how to get an estimation of the first two or three formant frequencies and their amplitudes as well?
LPC is almost impossible in Pd, since the documentation is really poor.
So, two ways I can think of is either using Fourier or just filtering.
Unfortunately I haven't been able to implement the Fourier transform and the filtering approach I use does not work really well..
So, ideas anyone????

ANY help will be highly appreciated!

Thank you in advance!!!

I'm doing something extremely similar. Did anyone get any results?