Improve vocal recognition and detection

[GoogleCodeExporter]

I tried vocals in the new beta, but something is not right. It jumps 
around a bit, even when singing in perfect tune. Is there a high pass 
filter on the vocals? If not there should be. I'd say the game should 
filter below 80Hz. This way it won't pick up any loud pops from P's, B's, 
etc. throwing your score off, when you might have actually sung the part 
perfectly.

Maybe add an option for HPF on/off in the options?

Original issue reported on code.google.com by mrhappen...@gmail.com on 29 Jun 2009 at 7:06

[GoogleCodeExporter]

I'll let other coders comment on this one (haven't checked the implementation) 
but
this sounds like a good idea and i doubt this would justify an option.

I think it should be on by default.

And actually, i'd go with a bandpass filter to filter out the obviously 
non-voice
high frequencies as well.

Original comment by evily...@gmail.com on 29 Jun 2009 at 8:14

[GoogleCodeExporter]

Not sure where the cutoff for high freq. should be. I don't think that would do 
as 
much though, seeing that harmonics are usually quieter than the notes being 
sung.

Original comment by mrhappen...@gmail.com on 29 Jun 2009 at 9:59

[GoogleCodeExporter]

Would be a nice addition for 3.120 final. 

Original comment by evily...@gmail.com on 13 Jul 2009 at 5:23

• Changed state: Accepted
• Added labels: Milestone-Release-3.120, Type-Enhancement, Usability

[GoogleCodeExporter]

Original comment by evily...@gmail.com on 13 Jul 2009 at 5:36

• Added labels: Priority-Medium
• Removed labels: Priority-Low

[GoogleCodeExporter]

Though it's essentially already included, and not a part of the pitch detection.

Original comment by aked...@gmail.com on 13 Jul 2009 at 11:54

[GoogleCodeExporter]

Ah, I see!
It's a matter of changing the bandpass cutoff frequencies in pypitch:
http://svn.jstump.com/websvn/pypitch/trunk/src/pitch.cpp

Should we make that configurable [allowing configuration]?

Original comment by evily...@gmail.com on 14 Jul 2009 at 3:56

[GoogleCodeExporter]

It's worth experimenting with, but probably not something that should be user-
configurable. 5000 is rather a good cutoff in theory. Human vocal frequencies 
are 
relevant up to about that range, and as the pitch detection works on harmonic 
detection cutting that too low we would risk losing out to instrument tracks 
with 
stronger harmonic frequencies in the lower range.

Original comment by aked...@gmail.com on 15 Jul 2009 at 12:10

[GoogleCodeExporter]

Vocal pitch detection should be seperate from the other audio if possible. The 
other 
tracks shouldn't have any cutoffs.

Original comment by mrhappen...@gmail.com on 15 Jul 2009 at 2:02

[GoogleCodeExporter]

Well yes... The only extent to which other instruments are relevant are in that 
the 
mic often picks up the played-back tracks from the game - we need to be able to 
ignore that.

Original comment by aked...@gmail.com on 15 Jul 2009 at 6:48

[GoogleCodeExporter]

Original comment by aked...@gmail.com on 17 Aug 2009 at 1:17

• Changed title: Improve vocal recognition and detection
• Added labels: Milestone-Release-4.0.0
• Removed labels: Milestone-Release-3.120

[GoogleCodeExporter]

Issue 927 has been merged into this issue.

Original comment by aked...@gmail.com on 17 Aug 2009 at 4:12

[GoogleCodeExporter]

I too have found vocals to be pretty unusable in its current state.  The arrow 
is by
no means stable and also has difficulties adjusting to various singers (e.g. 
volume
and so on).
Maybe a bandpass with a limiter is in order, or an option enabling fofix to 
control
the mic input volume (like skype does)?

Original comment by DNSchneid@gmail.com on 12 Oct 2009 at 9:23

[GoogleCodeExporter]

also, this is using an external mic that was made for some karaoke game, that 
as far
as I can tell with my tests in audacity has quite low noise levels, and does a 
pretty
good job of not feeding back.

Original comment by DNSchneid@gmail.com on 12 Oct 2009 at 9:28

[GoogleCodeExporter]

I agree with akedrou, isssue 927.
I'm using the RB Xbox Logitech mike, 
and when i sing "A" the pitch recognised by FOFIX is totally off.

Original comment by meister....@gmail.com on 8 Nov 2009 at 9:20

[GoogleCodeExporter]

I've experienced the exact same issue. I've tried...

My basic headset (normal input/output sockets)
A normal microphone (normal input)
USB Singstar microphone
USB Logitech microphone (from Guitar Hero World Tour)

...nothing works. The arrow is very jumpy but I can barely play. It can't 
detect what
I sing.

Original comment by aus_fla...@live.com.au on 13 Nov 2009 at 8:55

[GoogleCodeExporter]

@ aussie I found a workaround!

Under audio settings, there are two options: "pypitch" and "pypitchanalyser".

I changed it, still using my Logitech mike, and it works fine now.

Though the arrows are still kinda jumpy, but at least the pitch is on.

Original comment by meister....@gmail.com on 14 Nov 2009 at 3:30

[GoogleCodeExporter]

I shall try it immediately :D

Thanks johnny

Original comment by aus_fla...@live.com.au on 17 Nov 2009 at 2:07