Thank you Katjav.

you're right, [sigmund~] works fairly well. But octave shifts are not rare, and are my major concern. Violin is rather noisy, especially in transients, but I imagine I shouldn't dare dreaming of sensible improvements in transients, as I know this is a common problem in pitch tracking.

Thank you for your comments and explanation about pitch tracking, you have the eagle eye's view while I have the bug's eye view Now I understand that YIN is less interesting topic to be explored than others.
I'll keep on studying !

Thank you,

Nau

@nau said:

[sigmund~] works fairly well. But octave shifts are not rare, and are my major concern.

I'm still working on the pitch tracking topic, and curious to know about your experience, Nau. If you have octave jumps with [sigmund~], are they downward jumps or upward jumps in general? Or both?

Transients are a special case indeed. They tend to have fairly high frequency content, and a noisy character as compared to sustained notes. [sigmund~] has 'maxfreq' and 'stabletime' parameters which can help excluding transients and noise. Steep filtering at (for example) ~1 KHz also helps to get more stable pitch detection. But the best solution would be to combine transient detection and pitch detection in one comprehensive analysis process. Most transients can be detected within one or two milliseconds, as their magnitude rise is very sudden.

Nau, I have another, off topic question for you. I've seen it stated more than once that violin tones played together slightly out of tune do not produce the 'beat' or 'roughness' phenomenon, or at least less so than other instruments. Can you confirm this from experience? My father and I had once set up a Pd experiment to test the statement and quantify amplitude modulation. But the point is, we're no violin players. I wonder, could you record two slightly out of tune violin tones, a few seconds each, for this test? That would be cool.

Katja

@nau said:

Thank you Katjav.
you have the eagle eye's view while I have the bug's eye view

perfect analogy. love it

Hi Katjav,

I remember mainly higher-octave jumps, but maybe is this just of matter of perception. I'd need to play with it again and check it properly. I'll try steep filtering along. I forgot to mention another common problem encountered with my basic [sigmund~] implementation : as note-off are sent using a fixed low threshold value, sometimes a unique note is played with two or more separated notes. A classic case is the strong pizzicato on the lower string (G), as sound amplidtude exhibits an envelope that decreases below the threshold and goes back up again, producing a new note on, and then decreases again to produce the last note off. But I learned things in the meanwhile that could lead to improvements of my old patch. Amplitude calibration has taken me some time, and this make me think now that I should 'pre-process' the incoming sound to maximize stability and performance (compression ?, steep lo-pass,...)

About perception of beats with violins, I never heard that violins play 'unison' in a less ugly fashion than other instruments, and when I play with a friend of mine 'unison' are always tricky. But I'm just an amateur player, as you will hear with the three samples I recorded for you.
I had to power off the heating fan for recording and my playing is quite shivery I used the fine tuners to detune, but I'm not sure the amount of detune was well chosen in regard to the test. Tell me, I can re-record something better if you give me guidelines.

http://www.sendspace.com/file/4l9sag

Nau

Thanks for recording the violin notes for me Nau. I hope you did not catch a cold after turning the heating off.

The "beatless violin hypothesis" can be found in Arthur Benade's 'Fundamentals of Musical Acoustics' amongst others. See a possibly unofficial copy of the relevant chapter here (at about 3/4 down in the text):

http://www.zainea.com/violinfamily.htm

some citations:

"While I was still an undergraduate I noticed that one does not hear clear-cut beats between mistuned violin tones of the sort that painfully advertise slight errors between two wind instrument sounds. 1t was not difficult for me to recognize at that time that the weakness of the beats implied unsteadinesses in the sticking and slipping of the rosined bow on the string." ...
" On the one hand it allows larger tuning errors to be made in ensemble playing before the discrepancies become unacceptable, and on the other it permits the composer to write a wide variety of chords having many degrees of consonance and dissonance."

Recently I've seen that statement repeated more often, but don't remember where exactly. Anyway, I did a quick patch to check the amplitude modulations in the sum signal of your two original recordings. Of course, there's already a lot of amplitude modulations in the originals. Still it is very well possible to observe and quantify the beat phenomenon. See attached patch (to be used with your recordings).

'Thank you for watching this episode of MusicalMythBusters'.

Katja

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

I had to surrender. IPPASS is a nice idea, but Philip McLeod's SNAC is unbeatable in terms of speed and accuracy. Find a new pitch tracker [helmholtz~] based on SNAC here:

http://puredata.hurleur.com/sujet-6776-helmholtz-guess

Katja