[Martin Drautzburg]

Dear all

I've been looking for a free vocal pitch correction software or toolkit
for several weeks now, bit there does not seem to be anything. I am
even willing to program the stuff myself if I find a decent toolkit, so
I don't have to start from scratch.

Basically I am looking for pointers where to look next, or alternatively
a formum where they discuss such things.

Here is what I've done so far:

csound:

I managed to get a working pitch shifter by using lpanal (lpc-analysis)
and then resynthesize the signal with a programmed excitation signal
whose pitch I can control. The results were promising: formants were
not shifted, there is no chipmunk (aka Mickey Mouse) effect and you
could clearly identifify the singer. The overall quality however was
still too bad, the voice sounded kinda rough. I could get some
improvement by fine-tuning the spectrum of the excitation signal.

My conclusion was that the final step should always be an LPC
resynthesis (to get the formants right) and the tricky part is to get a
proper excitation signal. Ideally I would use the residue of the LPC
analysis and pitch-shift that by some standard method to get the
excitation signal, but alas, csound does not give me access to the
residue.

rubberband:

Rubberband does a fine job on pitch shifting, but it shifts formants
too, so you get the infamous chipmunk effect. There is an
option "preserve formants" but it does not work right.

audacity:

Audacity, like rubberband, does only do naive pitch shifting, with
formants shifted, i.e. you get a chipmunk effect.

clam:

Clam lets you graphically plug together an audio processing network and
it can even produce ladspa plugins (I haven't tried that yet), so it
seems like the ideal tool to me. It does a decent job on pitch-shifting
but it also shifts the formants. There are two LPC analyzers but I
can't see any LPC synthesizers. In fact there don't seem to be any
filters at all (an LPC resynthesizer is basically just a time-varying
digital filter).

praat:

praat does an excellent job on vocal pitch shifting but it is an
integrated tool (though sciptable). It does not work well on large
audio files (it becomes very slow) and there is no hope that pitch
shifting with praat could ever work in realtime.

commercial:

I haven't looked too deeply into commercial tools. I am aware of
- Antares autotune
- Melodyne
- Max/MSP

[Don Pearce]

Martin Drautzburg wrote:
Dear all

I've been looking for a free vocal pitch correction software or toolkit
for several weeks now, bit there does not seem to be anything. I am
even willing to program the stuff myself if I find a decent toolkit, so
I don't have to start from scratch.

Basically I am looking for pointers where to look next, or alternatively
a formum where they discuss such things.

Here is what I've done so far:

csound:

I managed to get a working pitch shifter by using lpanal (lpc-analysis)
and then resynthesize the signal with a programmed excitation signal
whose pitch I can control. The results were promising: formants were
not shifted, there is no chipmunk (aka Mickey Mouse) effect and you
could clearly identifify the singer. The overall quality however was
still too bad, the voice sounded kinda rough. I could get some
improvement by fine-tuning the spectrum of the excitation signal.

My conclusion was that the final step should always be an LPC
resynthesis (to get the formants right) and the tricky part is to get a
proper excitation signal. Ideally I would use the residue of the LPC
analysis and pitch-shift that by some standard method to get the
excitation signal, but alas, csound does not give me access to the
residue.

rubberband:

Rubberband does a fine job on pitch shifting, but it shifts formants
too, so you get the infamous chipmunk effect. There is an
option "preserve formants" but it does not work right.

audacity:

Audacity, like rubberband, does only do naive pitch shifting, with
formants shifted, i.e. you get a chipmunk effect.

clam:

Clam lets you graphically plug together an audio processing network and
it can even produce ladspa plugins (I haven't tried that yet), so it
seems like the ideal tool to me. It does a decent job on pitch-shifting
but it also shifts the formants. There are two LPC analyzers but I
can't see any LPC synthesizers. In fact there don't seem to be any
filters at all (an LPC resynthesizer is basically just a time-varying
digital filter).

praat:

praat does an excellent job on vocal pitch shifting but it is an
integrated tool (though sciptable). It does not work well on large
audio files (it becomes very slow) and there is no hope that pitch
shifting with praat could ever work in realtime.

commercial:

I haven't looked too deeply into commercial tools. I am aware of
- Antares autotune
- Melodyne
- Max/MSP




I use Melodyne and it is not far short of miraculous in what it can do.
There may be a trial version.

d

[Randy Yates]

Martin Drautzburg writes:

Dear all

I've been looking for a free vocal pitch correction software or toolkit
for several weeks now, bit there does not seem to be anything. I am
even willing to program the stuff myself if I find a decent toolkit, so
I don't have to start from scratch.

Basically I am looking for pointers where to look next, or alternatively
a formum where they discuss such things.

Here is what I've done so far:

csound:

I managed to get a working pitch shifter by using lpanal (lpc-analysis)
and then resynthesize the signal with a programmed excitation signal
whose pitch I can control. The results were promising: formants were
not shifted, there is no chipmunk (aka Mickey Mouse) effect and you
could clearly identifify the singer. The overall quality however was
still too bad, the voice sounded kinda rough. I could get some
improvement by fine-tuning the spectrum of the excitation signal.

My conclusion was that the final step should always be an LPC
resynthesis (to get the formants right) and the tricky part is to get a
proper excitation signal. Ideally I would use the residue of the LPC
analysis and pitch-shift that by some standard method to get the
excitation signal, but alas, csound does not give me access to the
residue.

rubberband:

Rubberband does a fine job on pitch shifting, but it shifts formants
too, so you get the infamous chipmunk effect. There is an
option "preserve formants" but it does not work right.

audacity:

Audacity, like rubberband, does only do naive pitch shifting, with
formants shifted, i.e. you get a chipmunk effect.

clam:

Clam lets you graphically plug together an audio processing network and
it can even produce ladspa plugins (I haven't tried that yet), so it
seems like the ideal tool to me. It does a decent job on pitch-shifting
but it also shifts the formants. There are two LPC analyzers but I
can't see any LPC synthesizers. In fact there don't seem to be any
filters at all (an LPC resynthesizer is basically just a time-varying
digital filter).

praat:

praat does an excellent job on vocal pitch shifting but it is an
integrated tool (though sciptable). It does not work well on large
audio files (it becomes very slow) and there is no hope that pitch
shifting with praat could ever work in realtime.

commercial:

I haven't looked too deeply into commercial tools. I am aware of
- Antares autotune
- Melodyne
- Max/MSP

Check out Wave Mechanics' "Pitch Doctor":

http://www.prodigy-pro.com/detail.aspx?ID=108

(OK, so I know the owner and a former development engineer...)

--Randy
--
% Randy Yates % "With time with what you've learned,
%% Fuquay-Varina, NC % they'll kiss the ground you walk
%%% 919-577-9882 % upon."
%%%% % '21st Century Man', *Time*, ELO
http://www.digitalsignallabs.com

[Martin Drautzburg]

Don Pearce wrote:

I use Melodyne and it is not far short of miraculous in what it can
do. There may be a trial version.

I downloaded the trial version and I was pleased to see that it runs
quite well under Linux using the wine emulator. Only problem is: I
cannot record directly into Melodyne, but thet is not essential for me
as I was looking for a postproduction tool anyways.

I agree it's amazing. It is lightyears ahead from what I could ever
program myself. I guess I'll stop this entire DIY project.

BTW: "vocaloids" is also pretty amazing, you don't need a human singer
at all. I just wonder when we get electronic listeners. Then we can
make them listen to electronic singers and we can sit by a campfire and
play guitar ;-)

Martin